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
;
359 EXPORT_SYMBOL_GPL(cpu_smt_control
);
361 static bool cpu_smt_available __read_mostly
;
363 void __init
cpu_smt_disable(bool force
)
365 if (cpu_smt_control
== CPU_SMT_FORCE_DISABLED
||
366 cpu_smt_control
== CPU_SMT_NOT_SUPPORTED
)
370 pr_info("SMT: Force disabled\n");
371 cpu_smt_control
= CPU_SMT_FORCE_DISABLED
;
373 cpu_smt_control
= CPU_SMT_DISABLED
;
378 * The decision whether SMT is supported can only be done after the full
379 * CPU identification. Called from architecture code before non boot CPUs
382 void __init
cpu_smt_check_topology_early(void)
384 if (!topology_smt_supported())
385 cpu_smt_control
= CPU_SMT_NOT_SUPPORTED
;
389 * If SMT was disabled by BIOS, detect it here, after the CPUs have been
390 * brought online. This ensures the smt/l1tf sysfs entries are consistent
391 * with reality. cpu_smt_available is set to true during the bringup of non
392 * boot CPUs when a SMT sibling is detected. Note, this may overwrite
393 * cpu_smt_control's previous setting.
395 void __init
cpu_smt_check_topology(void)
397 if (!cpu_smt_available
)
398 cpu_smt_control
= CPU_SMT_NOT_SUPPORTED
;
401 static int __init
smt_cmdline_disable(char *str
)
403 cpu_smt_disable(str
&& !strcmp(str
, "force"));
406 early_param("nosmt", smt_cmdline_disable
);
408 static inline bool cpu_smt_allowed(unsigned int cpu
)
410 if (topology_is_primary_thread(cpu
))
414 * If the CPU is not a 'primary' thread and the booted_once bit is
415 * set then the processor has SMT support. Store this information
416 * for the late check of SMT support in cpu_smt_check_topology().
418 if (per_cpu(cpuhp_state
, cpu
).booted_once
)
419 cpu_smt_available
= true;
421 if (cpu_smt_control
== CPU_SMT_ENABLED
)
425 * On x86 it's required to boot all logical CPUs at least once so
426 * that the init code can get a chance to set CR4.MCE on each
427 * CPU. Otherwise, a broadacasted MCE observing CR4.MCE=0b on any
428 * core will shutdown the machine.
430 return !per_cpu(cpuhp_state
, cpu
).booted_once
;
433 static inline bool cpu_smt_allowed(unsigned int cpu
) { return true; }
436 static inline enum cpuhp_state
437 cpuhp_set_state(struct cpuhp_cpu_state
*st
, enum cpuhp_state target
)
439 enum cpuhp_state prev_state
= st
->state
;
441 st
->rollback
= false;
446 st
->bringup
= st
->state
< target
;
452 cpuhp_reset_state(struct cpuhp_cpu_state
*st
, enum cpuhp_state prev_state
)
457 * If we have st->last we need to undo partial multi_instance of this
458 * state first. Otherwise start undo at the previous state.
467 st
->target
= prev_state
;
468 st
->bringup
= !st
->bringup
;
471 /* Regular hotplug invocation of the AP hotplug thread */
472 static void __cpuhp_kick_ap(struct cpuhp_cpu_state
*st
)
474 if (!st
->single
&& st
->state
== st
->target
)
479 * Make sure the above stores are visible before should_run becomes
480 * true. Paired with the mb() above in cpuhp_thread_fun()
483 st
->should_run
= true;
484 wake_up_process(st
->thread
);
485 wait_for_ap_thread(st
, st
->bringup
);
488 static int cpuhp_kick_ap(struct cpuhp_cpu_state
*st
, enum cpuhp_state target
)
490 enum cpuhp_state prev_state
;
493 prev_state
= cpuhp_set_state(st
, target
);
495 if ((ret
= st
->result
)) {
496 cpuhp_reset_state(st
, prev_state
);
503 static int bringup_wait_for_ap(unsigned int cpu
)
505 struct cpuhp_cpu_state
*st
= per_cpu_ptr(&cpuhp_state
, cpu
);
507 /* Wait for the CPU to reach CPUHP_AP_ONLINE_IDLE */
508 wait_for_ap_thread(st
, true);
509 if (WARN_ON_ONCE((!cpu_online(cpu
))))
512 /* Unpark the stopper thread and the hotplug thread of the target cpu */
513 stop_machine_unpark(cpu
);
514 kthread_unpark(st
->thread
);
517 * SMT soft disabling on X86 requires to bring the CPU out of the
518 * BIOS 'wait for SIPI' state in order to set the CR4.MCE bit. The
519 * CPU marked itself as booted_once in cpu_notify_starting() so the
520 * cpu_smt_allowed() check will now return false if this is not the
523 if (!cpu_smt_allowed(cpu
))
526 if (st
->target
<= CPUHP_AP_ONLINE_IDLE
)
529 return cpuhp_kick_ap(st
, st
->target
);
532 static int bringup_cpu(unsigned int cpu
)
534 struct task_struct
*idle
= idle_thread_get(cpu
);
538 * Some architectures have to walk the irq descriptors to
539 * setup the vector space for the cpu which comes online.
540 * Prevent irq alloc/free across the bringup.
544 /* Arch-specific enabling code. */
545 ret
= __cpu_up(cpu
, idle
);
549 return bringup_wait_for_ap(cpu
);
553 * Hotplug state machine related functions
556 static void undo_cpu_up(unsigned int cpu
, struct cpuhp_cpu_state
*st
)
558 for (st
->state
--; st
->state
> st
->target
; st
->state
--) {
559 struct cpuhp_step
*step
= cpuhp_get_step(st
->state
);
561 if (!step
->skip_onerr
)
562 cpuhp_invoke_callback(cpu
, st
->state
, false, NULL
, NULL
);
566 static int cpuhp_up_callbacks(unsigned int cpu
, struct cpuhp_cpu_state
*st
,
567 enum cpuhp_state target
)
569 enum cpuhp_state prev_state
= st
->state
;
572 while (st
->state
< target
) {
574 ret
= cpuhp_invoke_callback(cpu
, st
->state
, true, NULL
, NULL
);
576 st
->target
= prev_state
;
577 undo_cpu_up(cpu
, st
);
585 * The cpu hotplug threads manage the bringup and teardown of the cpus
587 static void cpuhp_create(unsigned int cpu
)
589 struct cpuhp_cpu_state
*st
= per_cpu_ptr(&cpuhp_state
, cpu
);
591 init_completion(&st
->done_up
);
592 init_completion(&st
->done_down
);
595 static int cpuhp_should_run(unsigned int cpu
)
597 struct cpuhp_cpu_state
*st
= this_cpu_ptr(&cpuhp_state
);
599 return st
->should_run
;
603 * Execute teardown/startup callbacks on the plugged cpu. Also used to invoke
604 * callbacks when a state gets [un]installed at runtime.
606 * Each invocation of this function by the smpboot thread does a single AP
609 * It has 3 modes of operation:
610 * - single: runs st->cb_state
611 * - up: runs ++st->state, while st->state < st->target
612 * - down: runs st->state--, while st->state > st->target
614 * When complete or on error, should_run is cleared and the completion is fired.
616 static void cpuhp_thread_fun(unsigned int cpu
)
618 struct cpuhp_cpu_state
*st
= this_cpu_ptr(&cpuhp_state
);
619 bool bringup
= st
->bringup
;
620 enum cpuhp_state state
;
623 * ACQUIRE for the cpuhp_should_run() load of ->should_run. Ensures
624 * that if we see ->should_run we also see the rest of the state.
628 if (WARN_ON_ONCE(!st
->should_run
))
631 cpuhp_lock_acquire(bringup
);
634 state
= st
->cb_state
;
635 st
->should_run
= false;
640 st
->should_run
= (st
->state
< st
->target
);
641 WARN_ON_ONCE(st
->state
> st
->target
);
645 st
->should_run
= (st
->state
> st
->target
);
646 WARN_ON_ONCE(st
->state
< st
->target
);
650 WARN_ON_ONCE(!cpuhp_is_ap_state(state
));
653 struct cpuhp_step
*step
= cpuhp_get_step(state
);
654 if (step
->skip_onerr
)
658 if (cpuhp_is_atomic_state(state
)) {
660 st
->result
= cpuhp_invoke_callback(cpu
, state
, bringup
, st
->node
, &st
->last
);
664 * STARTING/DYING must not fail!
666 WARN_ON_ONCE(st
->result
);
668 st
->result
= cpuhp_invoke_callback(cpu
, state
, bringup
, st
->node
, &st
->last
);
673 * If we fail on a rollback, we're up a creek without no
674 * paddle, no way forward, no way back. We loose, thanks for
677 WARN_ON_ONCE(st
->rollback
);
678 st
->should_run
= false;
682 cpuhp_lock_release(bringup
);
685 complete_ap_thread(st
, bringup
);
688 /* Invoke a single callback on a remote cpu */
690 cpuhp_invoke_ap_callback(int cpu
, enum cpuhp_state state
, bool bringup
,
691 struct hlist_node
*node
)
693 struct cpuhp_cpu_state
*st
= per_cpu_ptr(&cpuhp_state
, cpu
);
696 if (!cpu_online(cpu
))
699 cpuhp_lock_acquire(false);
700 cpuhp_lock_release(false);
702 cpuhp_lock_acquire(true);
703 cpuhp_lock_release(true);
706 * If we are up and running, use the hotplug thread. For early calls
707 * we invoke the thread function directly.
710 return cpuhp_invoke_callback(cpu
, state
, bringup
, node
, NULL
);
712 st
->rollback
= false;
716 st
->bringup
= bringup
;
717 st
->cb_state
= state
;
723 * If we failed and did a partial, do a rollback.
725 if ((ret
= st
->result
) && st
->last
) {
727 st
->bringup
= !bringup
;
733 * Clean up the leftovers so the next hotplug operation wont use stale
736 st
->node
= st
->last
= NULL
;
740 static int cpuhp_kick_ap_work(unsigned int cpu
)
742 struct cpuhp_cpu_state
*st
= per_cpu_ptr(&cpuhp_state
, cpu
);
743 enum cpuhp_state prev_state
= st
->state
;
746 cpuhp_lock_acquire(false);
747 cpuhp_lock_release(false);
749 cpuhp_lock_acquire(true);
750 cpuhp_lock_release(true);
752 trace_cpuhp_enter(cpu
, st
->target
, prev_state
, cpuhp_kick_ap_work
);
753 ret
= cpuhp_kick_ap(st
, st
->target
);
754 trace_cpuhp_exit(cpu
, st
->state
, prev_state
, ret
);
759 static struct smp_hotplug_thread cpuhp_threads
= {
760 .store
= &cpuhp_state
.thread
,
761 .create
= &cpuhp_create
,
762 .thread_should_run
= cpuhp_should_run
,
763 .thread_fn
= cpuhp_thread_fun
,
764 .thread_comm
= "cpuhp/%u",
768 void __init
cpuhp_threads_init(void)
770 BUG_ON(smpboot_register_percpu_thread(&cpuhp_threads
));
771 kthread_unpark(this_cpu_read(cpuhp_state
.thread
));
774 #ifdef CONFIG_HOTPLUG_CPU
776 * clear_tasks_mm_cpumask - Safely clear tasks' mm_cpumask for a CPU
779 * This function walks all processes, finds a valid mm struct for each one and
780 * then clears a corresponding bit in mm's cpumask. While this all sounds
781 * trivial, there are various non-obvious corner cases, which this function
782 * tries to solve in a safe manner.
784 * Also note that the function uses a somewhat relaxed locking scheme, so it may
785 * be called only for an already offlined CPU.
787 void clear_tasks_mm_cpumask(int cpu
)
789 struct task_struct
*p
;
792 * This function is called after the cpu is taken down and marked
793 * offline, so its not like new tasks will ever get this cpu set in
794 * their mm mask. -- Peter Zijlstra
795 * Thus, we may use rcu_read_lock() here, instead of grabbing
796 * full-fledged tasklist_lock.
798 WARN_ON(cpu_online(cpu
));
800 for_each_process(p
) {
801 struct task_struct
*t
;
804 * Main thread might exit, but other threads may still have
805 * a valid mm. Find one.
807 t
= find_lock_task_mm(p
);
810 cpumask_clear_cpu(cpu
, mm_cpumask(t
->mm
));
816 /* Take this CPU down. */
817 static int take_cpu_down(void *_param
)
819 struct cpuhp_cpu_state
*st
= this_cpu_ptr(&cpuhp_state
);
820 enum cpuhp_state target
= max((int)st
->target
, CPUHP_AP_OFFLINE
);
821 int err
, cpu
= smp_processor_id();
824 /* Ensure this CPU doesn't handle any more interrupts. */
825 err
= __cpu_disable();
830 * We get here while we are in CPUHP_TEARDOWN_CPU state and we must not
831 * do this step again.
833 WARN_ON(st
->state
!= CPUHP_TEARDOWN_CPU
);
835 /* Invoke the former CPU_DYING callbacks */
836 for (; st
->state
> target
; st
->state
--) {
837 ret
= cpuhp_invoke_callback(cpu
, st
->state
, false, NULL
, NULL
);
839 * DYING must not fail!
844 /* Give up timekeeping duties */
845 tick_handover_do_timer();
846 /* Park the stopper thread */
847 stop_machine_park(cpu
);
851 static int takedown_cpu(unsigned int cpu
)
853 struct cpuhp_cpu_state
*st
= per_cpu_ptr(&cpuhp_state
, cpu
);
856 /* Park the smpboot threads */
857 kthread_park(per_cpu_ptr(&cpuhp_state
, cpu
)->thread
);
860 * Prevent irq alloc/free while the dying cpu reorganizes the
861 * interrupt affinities.
866 * So now all preempt/rcu users must observe !cpu_active().
868 err
= stop_machine_cpuslocked(take_cpu_down
, NULL
, cpumask_of(cpu
));
870 /* CPU refused to die */
872 /* Unpark the hotplug thread so we can rollback there */
873 kthread_unpark(per_cpu_ptr(&cpuhp_state
, cpu
)->thread
);
876 BUG_ON(cpu_online(cpu
));
879 * The teardown callback for CPUHP_AP_SCHED_STARTING will have removed
880 * all runnable tasks from the CPU, there's only the idle task left now
881 * that the migration thread is done doing the stop_machine thing.
883 * Wait for the stop thread to go away.
885 wait_for_ap_thread(st
, false);
886 BUG_ON(st
->state
!= CPUHP_AP_IDLE_DEAD
);
888 /* Interrupts are moved away from the dying cpu, reenable alloc/free */
891 hotplug_cpu__broadcast_tick_pull(cpu
);
892 /* This actually kills the CPU. */
895 tick_cleanup_dead_cpu(cpu
);
896 rcutree_migrate_callbacks(cpu
);
900 static void cpuhp_complete_idle_dead(void *arg
)
902 struct cpuhp_cpu_state
*st
= arg
;
904 complete_ap_thread(st
, false);
907 void cpuhp_report_idle_dead(void)
909 struct cpuhp_cpu_state
*st
= this_cpu_ptr(&cpuhp_state
);
911 BUG_ON(st
->state
!= CPUHP_AP_OFFLINE
);
912 rcu_report_dead(smp_processor_id());
913 st
->state
= CPUHP_AP_IDLE_DEAD
;
915 * We cannot call complete after rcu_report_dead() so we delegate it
918 smp_call_function_single(cpumask_first(cpu_online_mask
),
919 cpuhp_complete_idle_dead
, st
, 0);
922 static void undo_cpu_down(unsigned int cpu
, struct cpuhp_cpu_state
*st
)
924 for (st
->state
++; st
->state
< st
->target
; st
->state
++) {
925 struct cpuhp_step
*step
= cpuhp_get_step(st
->state
);
927 if (!step
->skip_onerr
)
928 cpuhp_invoke_callback(cpu
, st
->state
, true, NULL
, NULL
);
932 static int cpuhp_down_callbacks(unsigned int cpu
, struct cpuhp_cpu_state
*st
,
933 enum cpuhp_state target
)
935 enum cpuhp_state prev_state
= st
->state
;
938 for (; st
->state
> target
; st
->state
--) {
939 ret
= cpuhp_invoke_callback(cpu
, st
->state
, false, NULL
, NULL
);
941 st
->target
= prev_state
;
942 undo_cpu_down(cpu
, st
);
949 /* Requires cpu_add_remove_lock to be held */
950 static int __ref
_cpu_down(unsigned int cpu
, int tasks_frozen
,
951 enum cpuhp_state target
)
953 struct cpuhp_cpu_state
*st
= per_cpu_ptr(&cpuhp_state
, cpu
);
954 int prev_state
, ret
= 0;
956 if (num_online_cpus() == 1)
959 if (!cpu_present(cpu
))
964 cpuhp_tasks_frozen
= tasks_frozen
;
966 prev_state
= cpuhp_set_state(st
, target
);
968 * If the current CPU state is in the range of the AP hotplug thread,
969 * then we need to kick the thread.
971 if (st
->state
> CPUHP_TEARDOWN_CPU
) {
972 st
->target
= max((int)target
, CPUHP_TEARDOWN_CPU
);
973 ret
= cpuhp_kick_ap_work(cpu
);
975 * The AP side has done the error rollback already. Just
976 * return the error code..
982 * We might have stopped still in the range of the AP hotplug
983 * thread. Nothing to do anymore.
985 if (st
->state
> CPUHP_TEARDOWN_CPU
)
991 * The AP brought itself down to CPUHP_TEARDOWN_CPU. So we need
992 * to do the further cleanups.
994 ret
= cpuhp_down_callbacks(cpu
, st
, target
);
995 if (ret
&& st
->state
> CPUHP_TEARDOWN_CPU
&& st
->state
< prev_state
) {
996 cpuhp_reset_state(st
, prev_state
);
1001 cpus_write_unlock();
1003 * Do post unplug cleanup. This is still protected against
1004 * concurrent CPU hotplug via cpu_add_remove_lock.
1006 lockup_detector_cleanup();
1011 static int cpu_down_maps_locked(unsigned int cpu
, enum cpuhp_state target
)
1013 if (cpu_hotplug_disabled
)
1015 return _cpu_down(cpu
, 0, target
);
1018 static int do_cpu_down(unsigned int cpu
, enum cpuhp_state target
)
1022 cpu_maps_update_begin();
1023 err
= cpu_down_maps_locked(cpu
, target
);
1024 cpu_maps_update_done();
1028 int cpu_down(unsigned int cpu
)
1030 return do_cpu_down(cpu
, CPUHP_OFFLINE
);
1032 EXPORT_SYMBOL(cpu_down
);
1035 #define takedown_cpu NULL
1036 #endif /*CONFIG_HOTPLUG_CPU*/
1039 * notify_cpu_starting(cpu) - Invoke the callbacks on the starting CPU
1040 * @cpu: cpu that just started
1042 * It must be called by the arch code on the new cpu, before the new cpu
1043 * enables interrupts and before the "boot" cpu returns from __cpu_up().
1045 void notify_cpu_starting(unsigned int cpu
)
1047 struct cpuhp_cpu_state
*st
= per_cpu_ptr(&cpuhp_state
, cpu
);
1048 enum cpuhp_state target
= min((int)st
->target
, CPUHP_AP_ONLINE
);
1051 rcu_cpu_starting(cpu
); /* Enables RCU usage on this CPU. */
1052 st
->booted_once
= true;
1053 while (st
->state
< target
) {
1055 ret
= cpuhp_invoke_callback(cpu
, st
->state
, true, NULL
, NULL
);
1057 * STARTING must not fail!
1064 * Called from the idle task. Wake up the controlling task which brings the
1065 * stopper and the hotplug thread of the upcoming CPU up and then delegates
1066 * the rest of the online bringup to the hotplug thread.
1068 void cpuhp_online_idle(enum cpuhp_state state
)
1070 struct cpuhp_cpu_state
*st
= this_cpu_ptr(&cpuhp_state
);
1072 /* Happens for the boot cpu */
1073 if (state
!= CPUHP_AP_ONLINE_IDLE
)
1076 st
->state
= CPUHP_AP_ONLINE_IDLE
;
1077 complete_ap_thread(st
, true);
1080 /* Requires cpu_add_remove_lock to be held */
1081 static int _cpu_up(unsigned int cpu
, int tasks_frozen
, enum cpuhp_state target
)
1083 struct cpuhp_cpu_state
*st
= per_cpu_ptr(&cpuhp_state
, cpu
);
1084 struct task_struct
*idle
;
1089 if (!cpu_present(cpu
)) {
1095 * The caller of do_cpu_up might have raced with another
1096 * caller. Ignore it for now.
1098 if (st
->state
>= target
)
1101 if (st
->state
== CPUHP_OFFLINE
) {
1102 /* Let it fail before we try to bring the cpu up */
1103 idle
= idle_thread_get(cpu
);
1105 ret
= PTR_ERR(idle
);
1110 cpuhp_tasks_frozen
= tasks_frozen
;
1112 cpuhp_set_state(st
, target
);
1114 * If the current CPU state is in the range of the AP hotplug thread,
1115 * then we need to kick the thread once more.
1117 if (st
->state
> CPUHP_BRINGUP_CPU
) {
1118 ret
= cpuhp_kick_ap_work(cpu
);
1120 * The AP side has done the error rollback already. Just
1121 * return the error code..
1128 * Try to reach the target state. We max out on the BP at
1129 * CPUHP_BRINGUP_CPU. After that the AP hotplug thread is
1130 * responsible for bringing it up to the target state.
1132 target
= min((int)target
, CPUHP_BRINGUP_CPU
);
1133 ret
= cpuhp_up_callbacks(cpu
, st
, target
);
1135 cpus_write_unlock();
1140 static int do_cpu_up(unsigned int cpu
, enum cpuhp_state target
)
1144 if (!cpu_possible(cpu
)) {
1145 pr_err("can't online cpu %d because it is not configured as may-hotadd at boot time\n",
1147 #if defined(CONFIG_IA64)
1148 pr_err("please check additional_cpus= boot parameter\n");
1153 err
= try_online_node(cpu_to_node(cpu
));
1157 cpu_maps_update_begin();
1159 if (cpu_hotplug_disabled
) {
1163 if (!cpu_smt_allowed(cpu
)) {
1168 err
= _cpu_up(cpu
, 0, target
);
1170 cpu_maps_update_done();
1174 int cpu_up(unsigned int cpu
)
1176 return do_cpu_up(cpu
, CPUHP_ONLINE
);
1178 EXPORT_SYMBOL_GPL(cpu_up
);
1180 #ifdef CONFIG_PM_SLEEP_SMP
1181 static cpumask_var_t frozen_cpus
;
1183 int freeze_secondary_cpus(int primary
)
1187 cpu_maps_update_begin();
1188 if (!cpu_online(primary
))
1189 primary
= cpumask_first(cpu_online_mask
);
1191 * We take down all of the non-boot CPUs in one shot to avoid races
1192 * with the userspace trying to use the CPU hotplug at the same time
1194 cpumask_clear(frozen_cpus
);
1196 pr_info("Disabling non-boot CPUs ...\n");
1197 for_each_online_cpu(cpu
) {
1200 trace_suspend_resume(TPS("CPU_OFF"), cpu
, true);
1201 error
= _cpu_down(cpu
, 1, CPUHP_OFFLINE
);
1202 trace_suspend_resume(TPS("CPU_OFF"), cpu
, false);
1204 cpumask_set_cpu(cpu
, frozen_cpus
);
1206 pr_err("Error taking CPU%d down: %d\n", cpu
, error
);
1212 BUG_ON(num_online_cpus() > 1);
1214 pr_err("Non-boot CPUs are not disabled\n");
1217 * Make sure the CPUs won't be enabled by someone else. We need to do
1218 * this even in case of failure as all disable_nonboot_cpus() users are
1219 * supposed to do enable_nonboot_cpus() on the failure path.
1221 cpu_hotplug_disabled
++;
1223 cpu_maps_update_done();
1227 void __weak
arch_enable_nonboot_cpus_begin(void)
1231 void __weak
arch_enable_nonboot_cpus_end(void)
1235 void enable_nonboot_cpus(void)
1239 /* Allow everyone to use the CPU hotplug again */
1240 cpu_maps_update_begin();
1241 __cpu_hotplug_enable();
1242 if (cpumask_empty(frozen_cpus
))
1245 pr_info("Enabling non-boot CPUs ...\n");
1247 arch_enable_nonboot_cpus_begin();
1249 for_each_cpu(cpu
, frozen_cpus
) {
1250 trace_suspend_resume(TPS("CPU_ON"), cpu
, true);
1251 error
= _cpu_up(cpu
, 1, CPUHP_ONLINE
);
1252 trace_suspend_resume(TPS("CPU_ON"), cpu
, false);
1254 pr_info("CPU%d is up\n", cpu
);
1257 pr_warn("Error taking CPU%d up: %d\n", cpu
, error
);
1260 arch_enable_nonboot_cpus_end();
1262 cpumask_clear(frozen_cpus
);
1264 cpu_maps_update_done();
1267 static int __init
alloc_frozen_cpus(void)
1269 if (!alloc_cpumask_var(&frozen_cpus
, GFP_KERNEL
|__GFP_ZERO
))
1273 core_initcall(alloc_frozen_cpus
);
1276 * When callbacks for CPU hotplug notifications are being executed, we must
1277 * ensure that the state of the system with respect to the tasks being frozen
1278 * or not, as reported by the notification, remains unchanged *throughout the
1279 * duration* of the execution of the callbacks.
1280 * Hence we need to prevent the freezer from racing with regular CPU hotplug.
1282 * This synchronization is implemented by mutually excluding regular CPU
1283 * hotplug and Suspend/Hibernate call paths by hooking onto the Suspend/
1284 * Hibernate notifications.
1287 cpu_hotplug_pm_callback(struct notifier_block
*nb
,
1288 unsigned long action
, void *ptr
)
1292 case PM_SUSPEND_PREPARE
:
1293 case PM_HIBERNATION_PREPARE
:
1294 cpu_hotplug_disable();
1297 case PM_POST_SUSPEND
:
1298 case PM_POST_HIBERNATION
:
1299 cpu_hotplug_enable();
1310 static int __init
cpu_hotplug_pm_sync_init(void)
1313 * cpu_hotplug_pm_callback has higher priority than x86
1314 * bsp_pm_callback which depends on cpu_hotplug_pm_callback
1315 * to disable cpu hotplug to avoid cpu hotplug race.
1317 pm_notifier(cpu_hotplug_pm_callback
, 0);
1320 core_initcall(cpu_hotplug_pm_sync_init
);
1322 #endif /* CONFIG_PM_SLEEP_SMP */
1326 #endif /* CONFIG_SMP */
1328 /* Boot processor state steps */
1329 static struct cpuhp_step cpuhp_bp_states
[] = {
1332 .startup
.single
= NULL
,
1333 .teardown
.single
= NULL
,
1336 [CPUHP_CREATE_THREADS
]= {
1337 .name
= "threads:prepare",
1338 .startup
.single
= smpboot_create_threads
,
1339 .teardown
.single
= NULL
,
1342 [CPUHP_PERF_PREPARE
] = {
1343 .name
= "perf:prepare",
1344 .startup
.single
= perf_event_init_cpu
,
1345 .teardown
.single
= perf_event_exit_cpu
,
1347 [CPUHP_WORKQUEUE_PREP
] = {
1348 .name
= "workqueue:prepare",
1349 .startup
.single
= workqueue_prepare_cpu
,
1350 .teardown
.single
= NULL
,
1352 [CPUHP_HRTIMERS_PREPARE
] = {
1353 .name
= "hrtimers:prepare",
1354 .startup
.single
= hrtimers_prepare_cpu
,
1355 .teardown
.single
= hrtimers_dead_cpu
,
1357 [CPUHP_SMPCFD_PREPARE
] = {
1358 .name
= "smpcfd:prepare",
1359 .startup
.single
= smpcfd_prepare_cpu
,
1360 .teardown
.single
= smpcfd_dead_cpu
,
1362 [CPUHP_RELAY_PREPARE
] = {
1363 .name
= "relay:prepare",
1364 .startup
.single
= relay_prepare_cpu
,
1365 .teardown
.single
= NULL
,
1367 [CPUHP_SLAB_PREPARE
] = {
1368 .name
= "slab:prepare",
1369 .startup
.single
= slab_prepare_cpu
,
1370 .teardown
.single
= slab_dead_cpu
,
1372 [CPUHP_RCUTREE_PREP
] = {
1373 .name
= "RCU/tree:prepare",
1374 .startup
.single
= rcutree_prepare_cpu
,
1375 .teardown
.single
= rcutree_dead_cpu
,
1378 * On the tear-down path, timers_dead_cpu() must be invoked
1379 * before blk_mq_queue_reinit_notify() from notify_dead(),
1380 * otherwise a RCU stall occurs.
1382 [CPUHP_TIMERS_PREPARE
] = {
1383 .name
= "timers:dead",
1384 .startup
.single
= timers_prepare_cpu
,
1385 .teardown
.single
= timers_dead_cpu
,
1387 /* Kicks the plugged cpu into life */
1388 [CPUHP_BRINGUP_CPU
] = {
1389 .name
= "cpu:bringup",
1390 .startup
.single
= bringup_cpu
,
1391 .teardown
.single
= NULL
,
1395 * Handled on controll processor until the plugged processor manages
1398 [CPUHP_TEARDOWN_CPU
] = {
1399 .name
= "cpu:teardown",
1400 .startup
.single
= NULL
,
1401 .teardown
.single
= takedown_cpu
,
1405 [CPUHP_BRINGUP_CPU
] = { },
1409 /* Application processor state steps */
1410 static struct cpuhp_step cpuhp_ap_states
[] = {
1412 /* Final state before CPU kills itself */
1413 [CPUHP_AP_IDLE_DEAD
] = {
1414 .name
= "idle:dead",
1417 * Last state before CPU enters the idle loop to die. Transient state
1418 * for synchronization.
1420 [CPUHP_AP_OFFLINE
] = {
1421 .name
= "ap:offline",
1424 /* First state is scheduler control. Interrupts are disabled */
1425 [CPUHP_AP_SCHED_STARTING
] = {
1426 .name
= "sched:starting",
1427 .startup
.single
= sched_cpu_starting
,
1428 .teardown
.single
= sched_cpu_dying
,
1430 [CPUHP_AP_RCUTREE_DYING
] = {
1431 .name
= "RCU/tree:dying",
1432 .startup
.single
= NULL
,
1433 .teardown
.single
= rcutree_dying_cpu
,
1435 [CPUHP_AP_SMPCFD_DYING
] = {
1436 .name
= "smpcfd:dying",
1437 .startup
.single
= NULL
,
1438 .teardown
.single
= smpcfd_dying_cpu
,
1440 /* Entry state on starting. Interrupts enabled from here on. Transient
1441 * state for synchronsization */
1442 [CPUHP_AP_ONLINE
] = {
1443 .name
= "ap:online",
1445 /* Handle smpboot threads park/unpark */
1446 [CPUHP_AP_SMPBOOT_THREADS
] = {
1447 .name
= "smpboot/threads:online",
1448 .startup
.single
= smpboot_unpark_threads
,
1449 .teardown
.single
= smpboot_park_threads
,
1451 [CPUHP_AP_IRQ_AFFINITY_ONLINE
] = {
1452 .name
= "irq/affinity:online",
1453 .startup
.single
= irq_affinity_online_cpu
,
1454 .teardown
.single
= NULL
,
1456 [CPUHP_AP_PERF_ONLINE
] = {
1457 .name
= "perf:online",
1458 .startup
.single
= perf_event_init_cpu
,
1459 .teardown
.single
= perf_event_exit_cpu
,
1461 [CPUHP_AP_WORKQUEUE_ONLINE
] = {
1462 .name
= "workqueue:online",
1463 .startup
.single
= workqueue_online_cpu
,
1464 .teardown
.single
= workqueue_offline_cpu
,
1466 [CPUHP_AP_RCUTREE_ONLINE
] = {
1467 .name
= "RCU/tree:online",
1468 .startup
.single
= rcutree_online_cpu
,
1469 .teardown
.single
= rcutree_offline_cpu
,
1473 * The dynamically registered state space is here
1477 /* Last state is scheduler control setting the cpu active */
1478 [CPUHP_AP_ACTIVE
] = {
1479 .name
= "sched:active",
1480 .startup
.single
= sched_cpu_activate
,
1481 .teardown
.single
= sched_cpu_deactivate
,
1485 /* CPU is fully up and running. */
1488 .startup
.single
= NULL
,
1489 .teardown
.single
= NULL
,
1493 /* Sanity check for callbacks */
1494 static int cpuhp_cb_check(enum cpuhp_state state
)
1496 if (state
<= CPUHP_OFFLINE
|| state
>= CPUHP_ONLINE
)
1502 * Returns a free for dynamic slot assignment of the Online state. The states
1503 * are protected by the cpuhp_slot_states mutex and an empty slot is identified
1504 * by having no name assigned.
1506 static int cpuhp_reserve_state(enum cpuhp_state state
)
1508 enum cpuhp_state i
, end
;
1509 struct cpuhp_step
*step
;
1512 case CPUHP_AP_ONLINE_DYN
:
1513 step
= cpuhp_ap_states
+ CPUHP_AP_ONLINE_DYN
;
1514 end
= CPUHP_AP_ONLINE_DYN_END
;
1516 case CPUHP_BP_PREPARE_DYN
:
1517 step
= cpuhp_bp_states
+ CPUHP_BP_PREPARE_DYN
;
1518 end
= CPUHP_BP_PREPARE_DYN_END
;
1524 for (i
= state
; i
<= end
; i
++, step
++) {
1528 WARN(1, "No more dynamic states available for CPU hotplug\n");
1532 static int cpuhp_store_callbacks(enum cpuhp_state state
, const char *name
,
1533 int (*startup
)(unsigned int cpu
),
1534 int (*teardown
)(unsigned int cpu
),
1535 bool multi_instance
)
1537 /* (Un)Install the callbacks for further cpu hotplug operations */
1538 struct cpuhp_step
*sp
;
1542 * If name is NULL, then the state gets removed.
1544 * CPUHP_AP_ONLINE_DYN and CPUHP_BP_PREPARE_DYN are handed out on
1545 * the first allocation from these dynamic ranges, so the removal
1546 * would trigger a new allocation and clear the wrong (already
1547 * empty) state, leaving the callbacks of the to be cleared state
1548 * dangling, which causes wreckage on the next hotplug operation.
1550 if (name
&& (state
== CPUHP_AP_ONLINE_DYN
||
1551 state
== CPUHP_BP_PREPARE_DYN
)) {
1552 ret
= cpuhp_reserve_state(state
);
1557 sp
= cpuhp_get_step(state
);
1558 if (name
&& sp
->name
)
1561 sp
->startup
.single
= startup
;
1562 sp
->teardown
.single
= teardown
;
1564 sp
->multi_instance
= multi_instance
;
1565 INIT_HLIST_HEAD(&sp
->list
);
1569 static void *cpuhp_get_teardown_cb(enum cpuhp_state state
)
1571 return cpuhp_get_step(state
)->teardown
.single
;
1575 * Call the startup/teardown function for a step either on the AP or
1576 * on the current CPU.
1578 static int cpuhp_issue_call(int cpu
, enum cpuhp_state state
, bool bringup
,
1579 struct hlist_node
*node
)
1581 struct cpuhp_step
*sp
= cpuhp_get_step(state
);
1585 * If there's nothing to do, we done.
1586 * Relies on the union for multi_instance.
1588 if ((bringup
&& !sp
->startup
.single
) ||
1589 (!bringup
&& !sp
->teardown
.single
))
1592 * The non AP bound callbacks can fail on bringup. On teardown
1593 * e.g. module removal we crash for now.
1596 if (cpuhp_is_ap_state(state
))
1597 ret
= cpuhp_invoke_ap_callback(cpu
, state
, bringup
, node
);
1599 ret
= cpuhp_invoke_callback(cpu
, state
, bringup
, node
, NULL
);
1601 ret
= cpuhp_invoke_callback(cpu
, state
, bringup
, node
, NULL
);
1603 BUG_ON(ret
&& !bringup
);
1608 * Called from __cpuhp_setup_state on a recoverable failure.
1610 * Note: The teardown callbacks for rollback are not allowed to fail!
1612 static void cpuhp_rollback_install(int failedcpu
, enum cpuhp_state state
,
1613 struct hlist_node
*node
)
1617 /* Roll back the already executed steps on the other cpus */
1618 for_each_present_cpu(cpu
) {
1619 struct cpuhp_cpu_state
*st
= per_cpu_ptr(&cpuhp_state
, cpu
);
1620 int cpustate
= st
->state
;
1622 if (cpu
>= failedcpu
)
1625 /* Did we invoke the startup call on that cpu ? */
1626 if (cpustate
>= state
)
1627 cpuhp_issue_call(cpu
, state
, false, node
);
1631 int __cpuhp_state_add_instance_cpuslocked(enum cpuhp_state state
,
1632 struct hlist_node
*node
,
1635 struct cpuhp_step
*sp
;
1639 lockdep_assert_cpus_held();
1641 sp
= cpuhp_get_step(state
);
1642 if (sp
->multi_instance
== false)
1645 mutex_lock(&cpuhp_state_mutex
);
1647 if (!invoke
|| !sp
->startup
.multi
)
1651 * Try to call the startup callback for each present cpu
1652 * depending on the hotplug state of the cpu.
1654 for_each_present_cpu(cpu
) {
1655 struct cpuhp_cpu_state
*st
= per_cpu_ptr(&cpuhp_state
, cpu
);
1656 int cpustate
= st
->state
;
1658 if (cpustate
< state
)
1661 ret
= cpuhp_issue_call(cpu
, state
, true, node
);
1663 if (sp
->teardown
.multi
)
1664 cpuhp_rollback_install(cpu
, state
, node
);
1670 hlist_add_head(node
, &sp
->list
);
1672 mutex_unlock(&cpuhp_state_mutex
);
1676 int __cpuhp_state_add_instance(enum cpuhp_state state
, struct hlist_node
*node
,
1682 ret
= __cpuhp_state_add_instance_cpuslocked(state
, node
, invoke
);
1686 EXPORT_SYMBOL_GPL(__cpuhp_state_add_instance
);
1689 * __cpuhp_setup_state_cpuslocked - Setup the callbacks for an hotplug machine state
1690 * @state: The state to setup
1691 * @invoke: If true, the startup function is invoked for cpus where
1692 * cpu state >= @state
1693 * @startup: startup callback function
1694 * @teardown: teardown callback function
1695 * @multi_instance: State is set up for multiple instances which get
1698 * The caller needs to hold cpus read locked while calling this function.
1701 * Positive state number if @state is CPUHP_AP_ONLINE_DYN
1702 * 0 for all other states
1703 * On failure: proper (negative) error code
1705 int __cpuhp_setup_state_cpuslocked(enum cpuhp_state state
,
1706 const char *name
, bool invoke
,
1707 int (*startup
)(unsigned int cpu
),
1708 int (*teardown
)(unsigned int cpu
),
1709 bool multi_instance
)
1714 lockdep_assert_cpus_held();
1716 if (cpuhp_cb_check(state
) || !name
)
1719 mutex_lock(&cpuhp_state_mutex
);
1721 ret
= cpuhp_store_callbacks(state
, name
, startup
, teardown
,
1724 dynstate
= state
== CPUHP_AP_ONLINE_DYN
;
1725 if (ret
> 0 && dynstate
) {
1730 if (ret
|| !invoke
|| !startup
)
1734 * Try to call the startup callback for each present cpu
1735 * depending on the hotplug state of the cpu.
1737 for_each_present_cpu(cpu
) {
1738 struct cpuhp_cpu_state
*st
= per_cpu_ptr(&cpuhp_state
, cpu
);
1739 int cpustate
= st
->state
;
1741 if (cpustate
< state
)
1744 ret
= cpuhp_issue_call(cpu
, state
, true, NULL
);
1747 cpuhp_rollback_install(cpu
, state
, NULL
);
1748 cpuhp_store_callbacks(state
, NULL
, NULL
, NULL
, false);
1753 mutex_unlock(&cpuhp_state_mutex
);
1755 * If the requested state is CPUHP_AP_ONLINE_DYN, return the
1756 * dynamically allocated state in case of success.
1758 if (!ret
&& dynstate
)
1762 EXPORT_SYMBOL(__cpuhp_setup_state_cpuslocked
);
1764 int __cpuhp_setup_state(enum cpuhp_state state
,
1765 const char *name
, bool invoke
,
1766 int (*startup
)(unsigned int cpu
),
1767 int (*teardown
)(unsigned int cpu
),
1768 bool multi_instance
)
1773 ret
= __cpuhp_setup_state_cpuslocked(state
, name
, invoke
, startup
,
1774 teardown
, multi_instance
);
1778 EXPORT_SYMBOL(__cpuhp_setup_state
);
1780 int __cpuhp_state_remove_instance(enum cpuhp_state state
,
1781 struct hlist_node
*node
, bool invoke
)
1783 struct cpuhp_step
*sp
= cpuhp_get_step(state
);
1786 BUG_ON(cpuhp_cb_check(state
));
1788 if (!sp
->multi_instance
)
1792 mutex_lock(&cpuhp_state_mutex
);
1794 if (!invoke
|| !cpuhp_get_teardown_cb(state
))
1797 * Call the teardown callback for each present cpu depending
1798 * on the hotplug state of the cpu. This function is not
1799 * allowed to fail currently!
1801 for_each_present_cpu(cpu
) {
1802 struct cpuhp_cpu_state
*st
= per_cpu_ptr(&cpuhp_state
, cpu
);
1803 int cpustate
= st
->state
;
1805 if (cpustate
>= state
)
1806 cpuhp_issue_call(cpu
, state
, false, node
);
1811 mutex_unlock(&cpuhp_state_mutex
);
1816 EXPORT_SYMBOL_GPL(__cpuhp_state_remove_instance
);
1819 * __cpuhp_remove_state_cpuslocked - Remove the callbacks for an hotplug machine state
1820 * @state: The state to remove
1821 * @invoke: If true, the teardown function is invoked for cpus where
1822 * cpu state >= @state
1824 * The caller needs to hold cpus read locked while calling this function.
1825 * The teardown callback is currently not allowed to fail. Think
1826 * about module removal!
1828 void __cpuhp_remove_state_cpuslocked(enum cpuhp_state state
, bool invoke
)
1830 struct cpuhp_step
*sp
= cpuhp_get_step(state
);
1833 BUG_ON(cpuhp_cb_check(state
));
1835 lockdep_assert_cpus_held();
1837 mutex_lock(&cpuhp_state_mutex
);
1838 if (sp
->multi_instance
) {
1839 WARN(!hlist_empty(&sp
->list
),
1840 "Error: Removing state %d which has instances left.\n",
1845 if (!invoke
|| !cpuhp_get_teardown_cb(state
))
1849 * Call the teardown callback for each present cpu depending
1850 * on the hotplug state of the cpu. This function is not
1851 * allowed to fail currently!
1853 for_each_present_cpu(cpu
) {
1854 struct cpuhp_cpu_state
*st
= per_cpu_ptr(&cpuhp_state
, cpu
);
1855 int cpustate
= st
->state
;
1857 if (cpustate
>= state
)
1858 cpuhp_issue_call(cpu
, state
, false, NULL
);
1861 cpuhp_store_callbacks(state
, NULL
, NULL
, NULL
, false);
1862 mutex_unlock(&cpuhp_state_mutex
);
1864 EXPORT_SYMBOL(__cpuhp_remove_state_cpuslocked
);
1866 void __cpuhp_remove_state(enum cpuhp_state state
, bool invoke
)
1869 __cpuhp_remove_state_cpuslocked(state
, invoke
);
1872 EXPORT_SYMBOL(__cpuhp_remove_state
);
1874 #if defined(CONFIG_SYSFS) && defined(CONFIG_HOTPLUG_CPU)
1875 static ssize_t
show_cpuhp_state(struct device
*dev
,
1876 struct device_attribute
*attr
, char *buf
)
1878 struct cpuhp_cpu_state
*st
= per_cpu_ptr(&cpuhp_state
, dev
->id
);
1880 return sprintf(buf
, "%d\n", st
->state
);
1882 static DEVICE_ATTR(state
, 0444, show_cpuhp_state
, NULL
);
1884 static ssize_t
write_cpuhp_target(struct device
*dev
,
1885 struct device_attribute
*attr
,
1886 const char *buf
, size_t count
)
1888 struct cpuhp_cpu_state
*st
= per_cpu_ptr(&cpuhp_state
, dev
->id
);
1889 struct cpuhp_step
*sp
;
1892 ret
= kstrtoint(buf
, 10, &target
);
1896 #ifdef CONFIG_CPU_HOTPLUG_STATE_CONTROL
1897 if (target
< CPUHP_OFFLINE
|| target
> CPUHP_ONLINE
)
1900 if (target
!= CPUHP_OFFLINE
&& target
!= CPUHP_ONLINE
)
1904 ret
= lock_device_hotplug_sysfs();
1908 mutex_lock(&cpuhp_state_mutex
);
1909 sp
= cpuhp_get_step(target
);
1910 ret
= !sp
->name
|| sp
->cant_stop
? -EINVAL
: 0;
1911 mutex_unlock(&cpuhp_state_mutex
);
1915 if (st
->state
< target
)
1916 ret
= do_cpu_up(dev
->id
, target
);
1918 ret
= do_cpu_down(dev
->id
, target
);
1920 unlock_device_hotplug();
1921 return ret
? ret
: count
;
1924 static ssize_t
show_cpuhp_target(struct device
*dev
,
1925 struct device_attribute
*attr
, char *buf
)
1927 struct cpuhp_cpu_state
*st
= per_cpu_ptr(&cpuhp_state
, dev
->id
);
1929 return sprintf(buf
, "%d\n", st
->target
);
1931 static DEVICE_ATTR(target
, 0644, show_cpuhp_target
, write_cpuhp_target
);
1934 static ssize_t
write_cpuhp_fail(struct device
*dev
,
1935 struct device_attribute
*attr
,
1936 const char *buf
, size_t count
)
1938 struct cpuhp_cpu_state
*st
= per_cpu_ptr(&cpuhp_state
, dev
->id
);
1939 struct cpuhp_step
*sp
;
1942 ret
= kstrtoint(buf
, 10, &fail
);
1947 * Cannot fail STARTING/DYING callbacks.
1949 if (cpuhp_is_atomic_state(fail
))
1953 * Cannot fail anything that doesn't have callbacks.
1955 mutex_lock(&cpuhp_state_mutex
);
1956 sp
= cpuhp_get_step(fail
);
1957 if (!sp
->startup
.single
&& !sp
->teardown
.single
)
1959 mutex_unlock(&cpuhp_state_mutex
);
1968 static ssize_t
show_cpuhp_fail(struct device
*dev
,
1969 struct device_attribute
*attr
, char *buf
)
1971 struct cpuhp_cpu_state
*st
= per_cpu_ptr(&cpuhp_state
, dev
->id
);
1973 return sprintf(buf
, "%d\n", st
->fail
);
1976 static DEVICE_ATTR(fail
, 0644, show_cpuhp_fail
, write_cpuhp_fail
);
1978 static struct attribute
*cpuhp_cpu_attrs
[] = {
1979 &dev_attr_state
.attr
,
1980 &dev_attr_target
.attr
,
1981 &dev_attr_fail
.attr
,
1985 static const struct attribute_group cpuhp_cpu_attr_group
= {
1986 .attrs
= cpuhp_cpu_attrs
,
1991 static ssize_t
show_cpuhp_states(struct device
*dev
,
1992 struct device_attribute
*attr
, char *buf
)
1994 ssize_t cur
, res
= 0;
1997 mutex_lock(&cpuhp_state_mutex
);
1998 for (i
= CPUHP_OFFLINE
; i
<= CPUHP_ONLINE
; i
++) {
1999 struct cpuhp_step
*sp
= cpuhp_get_step(i
);
2002 cur
= sprintf(buf
, "%3d: %s\n", i
, sp
->name
);
2007 mutex_unlock(&cpuhp_state_mutex
);
2010 static DEVICE_ATTR(states
, 0444, show_cpuhp_states
, NULL
);
2012 static struct attribute
*cpuhp_cpu_root_attrs
[] = {
2013 &dev_attr_states
.attr
,
2017 static const struct attribute_group cpuhp_cpu_root_attr_group
= {
2018 .attrs
= cpuhp_cpu_root_attrs
,
2023 #ifdef CONFIG_HOTPLUG_SMT
2025 static const char *smt_states
[] = {
2026 [CPU_SMT_ENABLED
] = "on",
2027 [CPU_SMT_DISABLED
] = "off",
2028 [CPU_SMT_FORCE_DISABLED
] = "forceoff",
2029 [CPU_SMT_NOT_SUPPORTED
] = "notsupported",
2033 show_smt_control(struct device
*dev
, struct device_attribute
*attr
, char *buf
)
2035 return snprintf(buf
, PAGE_SIZE
- 2, "%s\n", smt_states
[cpu_smt_control
]);
2038 static void cpuhp_offline_cpu_device(unsigned int cpu
)
2040 struct device
*dev
= get_cpu_device(cpu
);
2042 dev
->offline
= true;
2043 /* Tell user space about the state change */
2044 kobject_uevent(&dev
->kobj
, KOBJ_OFFLINE
);
2047 static void cpuhp_online_cpu_device(unsigned int cpu
)
2049 struct device
*dev
= get_cpu_device(cpu
);
2051 dev
->offline
= false;
2052 /* Tell user space about the state change */
2053 kobject_uevent(&dev
->kobj
, KOBJ_ONLINE
);
2056 static int cpuhp_smt_disable(enum cpuhp_smt_control ctrlval
)
2060 cpu_maps_update_begin();
2061 for_each_online_cpu(cpu
) {
2062 if (topology_is_primary_thread(cpu
))
2064 ret
= cpu_down_maps_locked(cpu
, CPUHP_OFFLINE
);
2068 * As this needs to hold the cpu maps lock it's impossible
2069 * to call device_offline() because that ends up calling
2070 * cpu_down() which takes cpu maps lock. cpu maps lock
2071 * needs to be held as this might race against in kernel
2072 * abusers of the hotplug machinery (thermal management).
2074 * So nothing would update device:offline state. That would
2075 * leave the sysfs entry stale and prevent onlining after
2076 * smt control has been changed to 'off' again. This is
2077 * called under the sysfs hotplug lock, so it is properly
2078 * serialized against the regular offline usage.
2080 cpuhp_offline_cpu_device(cpu
);
2083 cpu_smt_control
= ctrlval
;
2086 cpu_maps_update_done();
2090 static int cpuhp_smt_enable(void)
2094 cpu_maps_update_begin();
2095 cpu_smt_control
= CPU_SMT_ENABLED
;
2097 for_each_present_cpu(cpu
) {
2098 /* Skip online CPUs and CPUs on offline nodes */
2099 if (cpu_online(cpu
) || !node_online(cpu_to_node(cpu
)))
2101 ret
= _cpu_up(cpu
, 0, CPUHP_ONLINE
);
2104 /* See comment in cpuhp_smt_disable() */
2105 cpuhp_online_cpu_device(cpu
);
2107 cpu_maps_update_done();
2112 store_smt_control(struct device
*dev
, struct device_attribute
*attr
,
2113 const char *buf
, size_t count
)
2117 if (sysfs_streq(buf
, "on"))
2118 ctrlval
= CPU_SMT_ENABLED
;
2119 else if (sysfs_streq(buf
, "off"))
2120 ctrlval
= CPU_SMT_DISABLED
;
2121 else if (sysfs_streq(buf
, "forceoff"))
2122 ctrlval
= CPU_SMT_FORCE_DISABLED
;
2126 if (cpu_smt_control
== CPU_SMT_FORCE_DISABLED
)
2129 if (cpu_smt_control
== CPU_SMT_NOT_SUPPORTED
)
2132 ret
= lock_device_hotplug_sysfs();
2136 if (ctrlval
!= cpu_smt_control
) {
2138 case CPU_SMT_ENABLED
:
2139 ret
= cpuhp_smt_enable();
2141 case CPU_SMT_DISABLED
:
2142 case CPU_SMT_FORCE_DISABLED
:
2143 ret
= cpuhp_smt_disable(ctrlval
);
2148 unlock_device_hotplug();
2149 return ret
? ret
: count
;
2151 static DEVICE_ATTR(control
, 0644, show_smt_control
, store_smt_control
);
2154 show_smt_active(struct device
*dev
, struct device_attribute
*attr
, char *buf
)
2156 bool active
= topology_max_smt_threads() > 1;
2158 return snprintf(buf
, PAGE_SIZE
- 2, "%d\n", active
);
2160 static DEVICE_ATTR(active
, 0444, show_smt_active
, NULL
);
2162 static struct attribute
*cpuhp_smt_attrs
[] = {
2163 &dev_attr_control
.attr
,
2164 &dev_attr_active
.attr
,
2168 static const struct attribute_group cpuhp_smt_attr_group
= {
2169 .attrs
= cpuhp_smt_attrs
,
2174 static int __init
cpu_smt_state_init(void)
2176 return sysfs_create_group(&cpu_subsys
.dev_root
->kobj
,
2177 &cpuhp_smt_attr_group
);
2181 static inline int cpu_smt_state_init(void) { return 0; }
2184 static int __init
cpuhp_sysfs_init(void)
2188 ret
= cpu_smt_state_init();
2192 ret
= sysfs_create_group(&cpu_subsys
.dev_root
->kobj
,
2193 &cpuhp_cpu_root_attr_group
);
2197 for_each_possible_cpu(cpu
) {
2198 struct device
*dev
= get_cpu_device(cpu
);
2202 ret
= sysfs_create_group(&dev
->kobj
, &cpuhp_cpu_attr_group
);
2208 device_initcall(cpuhp_sysfs_init
);
2212 * cpu_bit_bitmap[] is a special, "compressed" data structure that
2213 * represents all NR_CPUS bits binary values of 1<<nr.
2215 * It is used by cpumask_of() to get a constant address to a CPU
2216 * mask value that has a single bit set only.
2219 /* cpu_bit_bitmap[0] is empty - so we can back into it */
2220 #define MASK_DECLARE_1(x) [x+1][0] = (1UL << (x))
2221 #define MASK_DECLARE_2(x) MASK_DECLARE_1(x), MASK_DECLARE_1(x+1)
2222 #define MASK_DECLARE_4(x) MASK_DECLARE_2(x), MASK_DECLARE_2(x+2)
2223 #define MASK_DECLARE_8(x) MASK_DECLARE_4(x), MASK_DECLARE_4(x+4)
2225 const unsigned long cpu_bit_bitmap
[BITS_PER_LONG
+1][BITS_TO_LONGS(NR_CPUS
)] = {
2227 MASK_DECLARE_8(0), MASK_DECLARE_8(8),
2228 MASK_DECLARE_8(16), MASK_DECLARE_8(24),
2229 #if BITS_PER_LONG > 32
2230 MASK_DECLARE_8(32), MASK_DECLARE_8(40),
2231 MASK_DECLARE_8(48), MASK_DECLARE_8(56),
2234 EXPORT_SYMBOL_GPL(cpu_bit_bitmap
);
2236 const DECLARE_BITMAP(cpu_all_bits
, NR_CPUS
) = CPU_BITS_ALL
;
2237 EXPORT_SYMBOL(cpu_all_bits
);
2239 #ifdef CONFIG_INIT_ALL_POSSIBLE
2240 struct cpumask __cpu_possible_mask __read_mostly
2243 struct cpumask __cpu_possible_mask __read_mostly
;
2245 EXPORT_SYMBOL(__cpu_possible_mask
);
2247 struct cpumask __cpu_online_mask __read_mostly
;
2248 EXPORT_SYMBOL(__cpu_online_mask
);
2250 struct cpumask __cpu_present_mask __read_mostly
;
2251 EXPORT_SYMBOL(__cpu_present_mask
);
2253 struct cpumask __cpu_active_mask __read_mostly
;
2254 EXPORT_SYMBOL(__cpu_active_mask
);
2256 void init_cpu_present(const struct cpumask
*src
)
2258 cpumask_copy(&__cpu_present_mask
, src
);
2261 void init_cpu_possible(const struct cpumask
*src
)
2263 cpumask_copy(&__cpu_possible_mask
, src
);
2266 void init_cpu_online(const struct cpumask
*src
)
2268 cpumask_copy(&__cpu_online_mask
, src
);
2272 * Activate the first processor.
2274 void __init
boot_cpu_init(void)
2276 int cpu
= smp_processor_id();
2278 /* Mark the boot cpu "present", "online" etc for SMP and UP case */
2279 set_cpu_online(cpu
, true);
2280 set_cpu_active(cpu
, true);
2281 set_cpu_present(cpu
, true);
2282 set_cpu_possible(cpu
, true);
2285 __boot_cpu_id
= cpu
;
2290 * Must be called _AFTER_ setting up the per_cpu areas
2292 void __init
boot_cpu_hotplug_init(void)
2294 per_cpu_ptr(&cpuhp_state
, smp_processor_id())->booted_once
= true;
2295 per_cpu_ptr(&cpuhp_state
, smp_processor_id())->state
= CPUHP_ONLINE
;
2298 enum cpu_mitigations cpu_mitigations __ro_after_init
= CPU_MITIGATIONS_AUTO
;
2300 static int __init
mitigations_parse_cmdline(char *arg
)
2302 if (!strcmp(arg
, "off"))
2303 cpu_mitigations
= CPU_MITIGATIONS_OFF
;
2304 else if (!strcmp(arg
, "auto"))
2305 cpu_mitigations
= CPU_MITIGATIONS_AUTO
;
2306 else if (!strcmp(arg
, "auto,nosmt"))
2307 cpu_mitigations
= CPU_MITIGATIONS_AUTO_NOSMT
;
2311 early_param("mitigations", mitigations_parse_cmdline
);