2 * (C) 2001, 2002, 2003, 2004 Rusty Russell
4 * This code is licenced under the GPL.
6 #include <linux/sched/mm.h>
7 #include <linux/proc_fs.h>
9 #include <linux/init.h>
10 #include <linux/notifier.h>
11 #include <linux/sched/signal.h>
12 #include <linux/sched/hotplug.h>
13 #include <linux/sched/isolation.h>
14 #include <linux/sched/task.h>
15 #include <linux/sched/smt.h>
16 #include <linux/unistd.h>
17 #include <linux/cpu.h>
18 #include <linux/oom.h>
19 #include <linux/rcupdate.h>
20 #include <linux/export.h>
21 #include <linux/bug.h>
22 #include <linux/kthread.h>
23 #include <linux/stop_machine.h>
24 #include <linux/mutex.h>
25 #include <linux/gfp.h>
26 #include <linux/suspend.h>
27 #include <linux/lockdep.h>
28 #include <linux/tick.h>
29 #include <linux/irq.h>
30 #include <linux/nmi.h>
31 #include <linux/smpboot.h>
32 #include <linux/relay.h>
33 #include <linux/slab.h>
34 #include <linux/scs.h>
35 #include <linux/percpu-rwsem.h>
36 #include <linux/cpuset.h>
37 #include <linux/random.h>
38 #include <linux/cc_platform.h>
40 #include <trace/events/power.h>
41 #define CREATE_TRACE_POINTS
42 #include <trace/events/cpuhp.h>
47 * struct cpuhp_cpu_state - Per cpu hotplug state storage
48 * @state: The current cpu state
49 * @target: The target state
50 * @fail: Current CPU hotplug callback state
51 * @thread: Pointer to the hotplug thread
52 * @should_run: Thread should execute
53 * @rollback: Perform a rollback
54 * @single: Single callback invocation
55 * @bringup: Single callback bringup or teardown selector
57 * @node: Remote CPU node; for multi-instance, do a
58 * single entry callback for install/remove
59 * @last: For multi-instance rollback, remember how far we got
60 * @cb_state: The state for a single callback (install/uninstall)
61 * @result: Result of the operation
62 * @done_up: Signal completion to the issuer of the task for cpu-up
63 * @done_down: Signal completion to the issuer of the task for cpu-down
65 struct cpuhp_cpu_state
{
66 enum cpuhp_state state
;
67 enum cpuhp_state target
;
68 enum cpuhp_state fail
;
70 struct task_struct
*thread
;
75 struct hlist_node
*node
;
76 struct hlist_node
*last
;
77 enum cpuhp_state cb_state
;
79 struct completion done_up
;
80 struct completion done_down
;
84 static DEFINE_PER_CPU(struct cpuhp_cpu_state
, cpuhp_state
) = {
85 .fail
= CPUHP_INVALID
,
89 cpumask_t cpus_booted_once_mask
;
92 #if defined(CONFIG_LOCKDEP) && defined(CONFIG_SMP)
93 static struct lockdep_map cpuhp_state_up_map
=
94 STATIC_LOCKDEP_MAP_INIT("cpuhp_state-up", &cpuhp_state_up_map
);
95 static struct lockdep_map cpuhp_state_down_map
=
96 STATIC_LOCKDEP_MAP_INIT("cpuhp_state-down", &cpuhp_state_down_map
);
99 static inline void cpuhp_lock_acquire(bool bringup
)
101 lock_map_acquire(bringup
? &cpuhp_state_up_map
: &cpuhp_state_down_map
);
104 static inline void cpuhp_lock_release(bool bringup
)
106 lock_map_release(bringup
? &cpuhp_state_up_map
: &cpuhp_state_down_map
);
110 static inline void cpuhp_lock_acquire(bool bringup
) { }
111 static inline void cpuhp_lock_release(bool bringup
) { }
116 * struct cpuhp_step - Hotplug state machine step
117 * @name: Name of the step
118 * @startup: Startup function of the step
119 * @teardown: Teardown function of the step
120 * @cant_stop: Bringup/teardown can't be stopped at this step
121 * @multi_instance: State has multiple instances which get added afterwards
126 int (*single
)(unsigned int cpu
);
127 int (*multi
)(unsigned int cpu
,
128 struct hlist_node
*node
);
131 int (*single
)(unsigned int cpu
);
132 int (*multi
)(unsigned int cpu
,
133 struct hlist_node
*node
);
136 struct hlist_head list
;
142 static DEFINE_MUTEX(cpuhp_state_mutex
);
143 static struct cpuhp_step cpuhp_hp_states
[];
145 static struct cpuhp_step
*cpuhp_get_step(enum cpuhp_state state
)
147 return cpuhp_hp_states
+ state
;
150 static bool cpuhp_step_empty(bool bringup
, struct cpuhp_step
*step
)
152 return bringup
? !step
->startup
.single
: !step
->teardown
.single
;
156 * cpuhp_invoke_callback - Invoke the callbacks for a given state
157 * @cpu: The cpu for which the callback should be invoked
158 * @state: The state to do callbacks for
159 * @bringup: True if the bringup callback should be invoked
160 * @node: For multi-instance, do a single entry callback for install/remove
161 * @lastp: For multi-instance rollback, remember how far we got
163 * Called from cpu hotplug and from the state register machinery.
165 * Return: %0 on success or a negative errno code
167 static int cpuhp_invoke_callback(unsigned int cpu
, enum cpuhp_state state
,
168 bool bringup
, struct hlist_node
*node
,
169 struct hlist_node
**lastp
)
171 struct cpuhp_cpu_state
*st
= per_cpu_ptr(&cpuhp_state
, cpu
);
172 struct cpuhp_step
*step
= cpuhp_get_step(state
);
173 int (*cbm
)(unsigned int cpu
, struct hlist_node
*node
);
174 int (*cb
)(unsigned int cpu
);
177 if (st
->fail
== state
) {
178 st
->fail
= CPUHP_INVALID
;
182 if (cpuhp_step_empty(bringup
, step
)) {
187 if (!step
->multi_instance
) {
188 WARN_ON_ONCE(lastp
&& *lastp
);
189 cb
= bringup
? step
->startup
.single
: step
->teardown
.single
;
191 trace_cpuhp_enter(cpu
, st
->target
, state
, cb
);
193 trace_cpuhp_exit(cpu
, st
->state
, state
, ret
);
196 cbm
= bringup
? step
->startup
.multi
: step
->teardown
.multi
;
198 /* Single invocation for instance add/remove */
200 WARN_ON_ONCE(lastp
&& *lastp
);
201 trace_cpuhp_multi_enter(cpu
, st
->target
, state
, cbm
, node
);
202 ret
= cbm(cpu
, node
);
203 trace_cpuhp_exit(cpu
, st
->state
, state
, ret
);
207 /* State transition. Invoke on all instances */
209 hlist_for_each(node
, &step
->list
) {
210 if (lastp
&& node
== *lastp
)
213 trace_cpuhp_multi_enter(cpu
, st
->target
, state
, cbm
, node
);
214 ret
= cbm(cpu
, node
);
215 trace_cpuhp_exit(cpu
, st
->state
, state
, ret
);
229 /* Rollback the instances if one failed */
230 cbm
= !bringup
? step
->startup
.multi
: step
->teardown
.multi
;
234 hlist_for_each(node
, &step
->list
) {
238 trace_cpuhp_multi_enter(cpu
, st
->target
, state
, cbm
, node
);
239 ret
= cbm(cpu
, node
);
240 trace_cpuhp_exit(cpu
, st
->state
, state
, ret
);
242 * Rollback must not fail,
250 static bool cpuhp_is_ap_state(enum cpuhp_state state
)
253 * The extra check for CPUHP_TEARDOWN_CPU is only for documentation
254 * purposes as that state is handled explicitly in cpu_down.
256 return state
> CPUHP_BRINGUP_CPU
&& state
!= CPUHP_TEARDOWN_CPU
;
259 static inline void wait_for_ap_thread(struct cpuhp_cpu_state
*st
, bool bringup
)
261 struct completion
*done
= bringup
? &st
->done_up
: &st
->done_down
;
262 wait_for_completion(done
);
265 static inline void complete_ap_thread(struct cpuhp_cpu_state
*st
, bool bringup
)
267 struct completion
*done
= bringup
? &st
->done_up
: &st
->done_down
;
272 * The former STARTING/DYING states, ran with IRQs disabled and must not fail.
274 static bool cpuhp_is_atomic_state(enum cpuhp_state state
)
276 return CPUHP_AP_IDLE_DEAD
<= state
&& state
< CPUHP_AP_ONLINE
;
279 /* Serializes the updates to cpu_online_mask, cpu_present_mask */
280 static DEFINE_MUTEX(cpu_add_remove_lock
);
281 bool cpuhp_tasks_frozen
;
282 EXPORT_SYMBOL_GPL(cpuhp_tasks_frozen
);
285 * The following two APIs (cpu_maps_update_begin/done) must be used when
286 * attempting to serialize the updates to cpu_online_mask & cpu_present_mask.
288 void cpu_maps_update_begin(void)
290 mutex_lock(&cpu_add_remove_lock
);
293 void cpu_maps_update_done(void)
295 mutex_unlock(&cpu_add_remove_lock
);
299 * If set, cpu_up and cpu_down will return -EBUSY and do nothing.
300 * Should always be manipulated under cpu_add_remove_lock
302 static int cpu_hotplug_disabled
;
304 #ifdef CONFIG_HOTPLUG_CPU
306 DEFINE_STATIC_PERCPU_RWSEM(cpu_hotplug_lock
);
308 void cpus_read_lock(void)
310 percpu_down_read(&cpu_hotplug_lock
);
312 EXPORT_SYMBOL_GPL(cpus_read_lock
);
314 int cpus_read_trylock(void)
316 return percpu_down_read_trylock(&cpu_hotplug_lock
);
318 EXPORT_SYMBOL_GPL(cpus_read_trylock
);
320 void cpus_read_unlock(void)
322 percpu_up_read(&cpu_hotplug_lock
);
324 EXPORT_SYMBOL_GPL(cpus_read_unlock
);
326 void cpus_write_lock(void)
328 percpu_down_write(&cpu_hotplug_lock
);
331 void cpus_write_unlock(void)
333 percpu_up_write(&cpu_hotplug_lock
);
336 void lockdep_assert_cpus_held(void)
339 * We can't have hotplug operations before userspace starts running,
340 * and some init codepaths will knowingly not take the hotplug lock.
341 * This is all valid, so mute lockdep until it makes sense to report
344 if (system_state
< SYSTEM_RUNNING
)
347 percpu_rwsem_assert_held(&cpu_hotplug_lock
);
350 #ifdef CONFIG_LOCKDEP
351 int lockdep_is_cpus_held(void)
353 return percpu_rwsem_is_held(&cpu_hotplug_lock
);
357 static void lockdep_acquire_cpus_lock(void)
359 rwsem_acquire(&cpu_hotplug_lock
.dep_map
, 0, 0, _THIS_IP_
);
362 static void lockdep_release_cpus_lock(void)
364 rwsem_release(&cpu_hotplug_lock
.dep_map
, _THIS_IP_
);
368 * Wait for currently running CPU hotplug operations to complete (if any) and
369 * disable future CPU hotplug (from sysfs). The 'cpu_add_remove_lock' protects
370 * the 'cpu_hotplug_disabled' flag. The same lock is also acquired by the
371 * hotplug path before performing hotplug operations. So acquiring that lock
372 * guarantees mutual exclusion from any currently running hotplug operations.
374 void cpu_hotplug_disable(void)
376 cpu_maps_update_begin();
377 cpu_hotplug_disabled
++;
378 cpu_maps_update_done();
380 EXPORT_SYMBOL_GPL(cpu_hotplug_disable
);
382 static void __cpu_hotplug_enable(void)
384 if (WARN_ONCE(!cpu_hotplug_disabled
, "Unbalanced cpu hotplug enable\n"))
386 cpu_hotplug_disabled
--;
389 void cpu_hotplug_enable(void)
391 cpu_maps_update_begin();
392 __cpu_hotplug_enable();
393 cpu_maps_update_done();
395 EXPORT_SYMBOL_GPL(cpu_hotplug_enable
);
399 static void lockdep_acquire_cpus_lock(void)
403 static void lockdep_release_cpus_lock(void)
407 #endif /* CONFIG_HOTPLUG_CPU */
410 * Architectures that need SMT-specific errata handling during SMT hotplug
411 * should override this.
413 void __weak
arch_smt_update(void) { }
415 #ifdef CONFIG_HOTPLUG_SMT
416 enum cpuhp_smt_control cpu_smt_control __read_mostly
= CPU_SMT_ENABLED
;
418 void __init
cpu_smt_disable(bool force
)
420 if (!cpu_smt_possible())
424 pr_info("SMT: Force disabled\n");
425 cpu_smt_control
= CPU_SMT_FORCE_DISABLED
;
427 pr_info("SMT: disabled\n");
428 cpu_smt_control
= CPU_SMT_DISABLED
;
433 * The decision whether SMT is supported can only be done after the full
434 * CPU identification. Called from architecture code.
436 void __init
cpu_smt_check_topology(void)
438 if (!topology_smt_supported())
439 cpu_smt_control
= CPU_SMT_NOT_SUPPORTED
;
442 static int __init
smt_cmdline_disable(char *str
)
444 cpu_smt_disable(str
&& !strcmp(str
, "force"));
447 early_param("nosmt", smt_cmdline_disable
);
449 static inline bool cpu_smt_allowed(unsigned int cpu
)
451 if (cpu_smt_control
== CPU_SMT_ENABLED
)
454 if (topology_is_primary_thread(cpu
))
458 * On x86 it's required to boot all logical CPUs at least once so
459 * that the init code can get a chance to set CR4.MCE on each
460 * CPU. Otherwise, a broadcasted MCE observing CR4.MCE=0b on any
461 * core will shutdown the machine.
463 return !cpumask_test_cpu(cpu
, &cpus_booted_once_mask
);
466 /* Returns true if SMT is not supported of forcefully (irreversibly) disabled */
467 bool cpu_smt_possible(void)
469 return cpu_smt_control
!= CPU_SMT_FORCE_DISABLED
&&
470 cpu_smt_control
!= CPU_SMT_NOT_SUPPORTED
;
472 EXPORT_SYMBOL_GPL(cpu_smt_possible
);
474 static inline bool cpu_smt_allowed(unsigned int cpu
) { return true; }
477 static inline enum cpuhp_state
478 cpuhp_set_state(int cpu
, struct cpuhp_cpu_state
*st
, enum cpuhp_state target
)
480 enum cpuhp_state prev_state
= st
->state
;
481 bool bringup
= st
->state
< target
;
483 st
->rollback
= false;
488 st
->bringup
= bringup
;
489 if (cpu_dying(cpu
) != !bringup
)
490 set_cpu_dying(cpu
, !bringup
);
496 cpuhp_reset_state(int cpu
, struct cpuhp_cpu_state
*st
,
497 enum cpuhp_state prev_state
)
499 bool bringup
= !st
->bringup
;
501 st
->target
= prev_state
;
504 * Already rolling back. No need invert the bringup value or to change
513 * If we have st->last we need to undo partial multi_instance of this
514 * state first. Otherwise start undo at the previous state.
523 st
->bringup
= bringup
;
524 if (cpu_dying(cpu
) != !bringup
)
525 set_cpu_dying(cpu
, !bringup
);
528 /* Regular hotplug invocation of the AP hotplug thread */
529 static void __cpuhp_kick_ap(struct cpuhp_cpu_state
*st
)
531 if (!st
->single
&& st
->state
== st
->target
)
536 * Make sure the above stores are visible before should_run becomes
537 * true. Paired with the mb() above in cpuhp_thread_fun()
540 st
->should_run
= true;
541 wake_up_process(st
->thread
);
542 wait_for_ap_thread(st
, st
->bringup
);
545 static int cpuhp_kick_ap(int cpu
, struct cpuhp_cpu_state
*st
,
546 enum cpuhp_state target
)
548 enum cpuhp_state prev_state
;
551 prev_state
= cpuhp_set_state(cpu
, st
, target
);
553 if ((ret
= st
->result
)) {
554 cpuhp_reset_state(cpu
, st
, prev_state
);
561 static int bringup_wait_for_ap(unsigned int cpu
)
563 struct cpuhp_cpu_state
*st
= per_cpu_ptr(&cpuhp_state
, cpu
);
565 /* Wait for the CPU to reach CPUHP_AP_ONLINE_IDLE */
566 wait_for_ap_thread(st
, true);
567 if (WARN_ON_ONCE((!cpu_online(cpu
))))
570 /* Unpark the hotplug thread of the target cpu */
571 kthread_unpark(st
->thread
);
574 * SMT soft disabling on X86 requires to bring the CPU out of the
575 * BIOS 'wait for SIPI' state in order to set the CR4.MCE bit. The
576 * CPU marked itself as booted_once in notify_cpu_starting() so the
577 * cpu_smt_allowed() check will now return false if this is not the
580 if (!cpu_smt_allowed(cpu
))
583 if (st
->target
<= CPUHP_AP_ONLINE_IDLE
)
586 return cpuhp_kick_ap(cpu
, st
, st
->target
);
589 static int bringup_cpu(unsigned int cpu
)
591 struct task_struct
*idle
= idle_thread_get(cpu
);
595 * Reset stale stack state from the last time this CPU was online.
597 scs_task_reset(idle
);
598 kasan_unpoison_task_stack(idle
);
601 * Some architectures have to walk the irq descriptors to
602 * setup the vector space for the cpu which comes online.
603 * Prevent irq alloc/free across the bringup.
607 /* Arch-specific enabling code. */
608 ret
= __cpu_up(cpu
, idle
);
612 return bringup_wait_for_ap(cpu
);
615 static int finish_cpu(unsigned int cpu
)
617 struct task_struct
*idle
= idle_thread_get(cpu
);
618 struct mm_struct
*mm
= idle
->active_mm
;
621 * idle_task_exit() will have switched to &init_mm, now
622 * clean up any remaining active_mm state.
625 idle
->active_mm
= &init_mm
;
631 * Hotplug state machine related functions
635 * Get the next state to run. Empty ones will be skipped. Returns true if a
638 * st->state will be modified ahead of time, to match state_to_run, as if it
641 static bool cpuhp_next_state(bool bringup
,
642 enum cpuhp_state
*state_to_run
,
643 struct cpuhp_cpu_state
*st
,
644 enum cpuhp_state target
)
648 if (st
->state
>= target
)
651 *state_to_run
= ++st
->state
;
653 if (st
->state
<= target
)
656 *state_to_run
= st
->state
--;
659 if (!cpuhp_step_empty(bringup
, cpuhp_get_step(*state_to_run
)))
666 static int __cpuhp_invoke_callback_range(bool bringup
,
668 struct cpuhp_cpu_state
*st
,
669 enum cpuhp_state target
,
672 enum cpuhp_state state
;
675 while (cpuhp_next_state(bringup
, &state
, st
, target
)) {
678 err
= cpuhp_invoke_callback(cpu
, state
, bringup
, NULL
, NULL
);
683 pr_warn("CPU %u %s state %s (%d) failed (%d)\n",
684 cpu
, bringup
? "UP" : "DOWN",
685 cpuhp_get_step(st
->state
)->name
,
697 static inline int cpuhp_invoke_callback_range(bool bringup
,
699 struct cpuhp_cpu_state
*st
,
700 enum cpuhp_state target
)
702 return __cpuhp_invoke_callback_range(bringup
, cpu
, st
, target
, false);
705 static inline void cpuhp_invoke_callback_range_nofail(bool bringup
,
707 struct cpuhp_cpu_state
*st
,
708 enum cpuhp_state target
)
710 __cpuhp_invoke_callback_range(bringup
, cpu
, st
, target
, true);
713 static inline bool can_rollback_cpu(struct cpuhp_cpu_state
*st
)
715 if (IS_ENABLED(CONFIG_HOTPLUG_CPU
))
718 * When CPU hotplug is disabled, then taking the CPU down is not
719 * possible because takedown_cpu() and the architecture and
720 * subsystem specific mechanisms are not available. So the CPU
721 * which would be completely unplugged again needs to stay around
722 * in the current state.
724 return st
->state
<= CPUHP_BRINGUP_CPU
;
727 static int cpuhp_up_callbacks(unsigned int cpu
, struct cpuhp_cpu_state
*st
,
728 enum cpuhp_state target
)
730 enum cpuhp_state prev_state
= st
->state
;
733 ret
= cpuhp_invoke_callback_range(true, cpu
, st
, target
);
735 pr_debug("CPU UP failed (%d) CPU %u state %s (%d)\n",
736 ret
, cpu
, cpuhp_get_step(st
->state
)->name
,
739 cpuhp_reset_state(cpu
, st
, prev_state
);
740 if (can_rollback_cpu(st
))
741 WARN_ON(cpuhp_invoke_callback_range(false, cpu
, st
,
748 * The cpu hotplug threads manage the bringup and teardown of the cpus
750 static int cpuhp_should_run(unsigned int cpu
)
752 struct cpuhp_cpu_state
*st
= this_cpu_ptr(&cpuhp_state
);
754 return st
->should_run
;
758 * Execute teardown/startup callbacks on the plugged cpu. Also used to invoke
759 * callbacks when a state gets [un]installed at runtime.
761 * Each invocation of this function by the smpboot thread does a single AP
764 * It has 3 modes of operation:
765 * - single: runs st->cb_state
766 * - up: runs ++st->state, while st->state < st->target
767 * - down: runs st->state--, while st->state > st->target
769 * When complete or on error, should_run is cleared and the completion is fired.
771 static void cpuhp_thread_fun(unsigned int cpu
)
773 struct cpuhp_cpu_state
*st
= this_cpu_ptr(&cpuhp_state
);
774 bool bringup
= st
->bringup
;
775 enum cpuhp_state state
;
777 if (WARN_ON_ONCE(!st
->should_run
))
781 * ACQUIRE for the cpuhp_should_run() load of ->should_run. Ensures
782 * that if we see ->should_run we also see the rest of the state.
787 * The BP holds the hotplug lock, but we're now running on the AP,
788 * ensure that anybody asserting the lock is held, will actually find
791 lockdep_acquire_cpus_lock();
792 cpuhp_lock_acquire(bringup
);
795 state
= st
->cb_state
;
796 st
->should_run
= false;
798 st
->should_run
= cpuhp_next_state(bringup
, &state
, st
, st
->target
);
803 WARN_ON_ONCE(!cpuhp_is_ap_state(state
));
805 if (cpuhp_is_atomic_state(state
)) {
807 st
->result
= cpuhp_invoke_callback(cpu
, state
, bringup
, st
->node
, &st
->last
);
811 * STARTING/DYING must not fail!
813 WARN_ON_ONCE(st
->result
);
815 st
->result
= cpuhp_invoke_callback(cpu
, state
, bringup
, st
->node
, &st
->last
);
820 * If we fail on a rollback, we're up a creek without no
821 * paddle, no way forward, no way back. We loose, thanks for
824 WARN_ON_ONCE(st
->rollback
);
825 st
->should_run
= false;
829 cpuhp_lock_release(bringup
);
830 lockdep_release_cpus_lock();
833 complete_ap_thread(st
, bringup
);
836 /* Invoke a single callback on a remote cpu */
838 cpuhp_invoke_ap_callback(int cpu
, enum cpuhp_state state
, bool bringup
,
839 struct hlist_node
*node
)
841 struct cpuhp_cpu_state
*st
= per_cpu_ptr(&cpuhp_state
, cpu
);
844 if (!cpu_online(cpu
))
847 cpuhp_lock_acquire(false);
848 cpuhp_lock_release(false);
850 cpuhp_lock_acquire(true);
851 cpuhp_lock_release(true);
854 * If we are up and running, use the hotplug thread. For early calls
855 * we invoke the thread function directly.
858 return cpuhp_invoke_callback(cpu
, state
, bringup
, node
, NULL
);
860 st
->rollback
= false;
864 st
->bringup
= bringup
;
865 st
->cb_state
= state
;
871 * If we failed and did a partial, do a rollback.
873 if ((ret
= st
->result
) && st
->last
) {
875 st
->bringup
= !bringup
;
881 * Clean up the leftovers so the next hotplug operation wont use stale
884 st
->node
= st
->last
= NULL
;
888 static int cpuhp_kick_ap_work(unsigned int cpu
)
890 struct cpuhp_cpu_state
*st
= per_cpu_ptr(&cpuhp_state
, cpu
);
891 enum cpuhp_state prev_state
= st
->state
;
894 cpuhp_lock_acquire(false);
895 cpuhp_lock_release(false);
897 cpuhp_lock_acquire(true);
898 cpuhp_lock_release(true);
900 trace_cpuhp_enter(cpu
, st
->target
, prev_state
, cpuhp_kick_ap_work
);
901 ret
= cpuhp_kick_ap(cpu
, st
, st
->target
);
902 trace_cpuhp_exit(cpu
, st
->state
, prev_state
, ret
);
907 static struct smp_hotplug_thread cpuhp_threads
= {
908 .store
= &cpuhp_state
.thread
,
909 .thread_should_run
= cpuhp_should_run
,
910 .thread_fn
= cpuhp_thread_fun
,
911 .thread_comm
= "cpuhp/%u",
915 static __init
void cpuhp_init_state(void)
917 struct cpuhp_cpu_state
*st
;
920 for_each_possible_cpu(cpu
) {
921 st
= per_cpu_ptr(&cpuhp_state
, cpu
);
922 init_completion(&st
->done_up
);
923 init_completion(&st
->done_down
);
927 void __init
cpuhp_threads_init(void)
930 BUG_ON(smpboot_register_percpu_thread(&cpuhp_threads
));
931 kthread_unpark(this_cpu_read(cpuhp_state
.thread
));
936 * Serialize hotplug trainwrecks outside of the cpu_hotplug_lock
939 * The operation is still serialized against concurrent CPU hotplug via
940 * cpu_add_remove_lock, i.e. CPU map protection. But it is _not_
941 * serialized against other hotplug related activity like adding or
942 * removing of state callbacks and state instances, which invoke either the
943 * startup or the teardown callback of the affected state.
945 * This is required for subsystems which are unfixable vs. CPU hotplug and
946 * evade lock inversion problems by scheduling work which has to be
947 * completed _before_ cpu_up()/_cpu_down() returns.
949 * Don't even think about adding anything to this for any new code or even
950 * drivers. It's only purpose is to keep existing lock order trainwrecks
953 * For cpu_down() there might be valid reasons to finish cleanups which are
954 * not required to be done under cpu_hotplug_lock, but that's a different
955 * story and would be not invoked via this.
957 static void cpu_up_down_serialize_trainwrecks(bool tasks_frozen
)
960 * cpusets delegate hotplug operations to a worker to "solve" the
961 * lock order problems. Wait for the worker, but only if tasks are
962 * _not_ frozen (suspend, hibernate) as that would wait forever.
964 * The wait is required because otherwise the hotplug operation
965 * returns with inconsistent state, which could even be observed in
966 * user space when a new CPU is brought up. The CPU plug uevent
967 * would be delivered and user space reacting on it would fail to
968 * move tasks to the newly plugged CPU up to the point where the
969 * work has finished because up to that point the newly plugged CPU
970 * is not assignable in cpusets/cgroups. On unplug that's not
971 * necessarily a visible issue, but it is still inconsistent state,
972 * which is the real problem which needs to be "fixed". This can't
973 * prevent the transient state between scheduling the work and
974 * returning from waiting for it.
977 cpuset_wait_for_hotplug();
980 #ifdef CONFIG_HOTPLUG_CPU
981 #ifndef arch_clear_mm_cpumask_cpu
982 #define arch_clear_mm_cpumask_cpu(cpu, mm) cpumask_clear_cpu(cpu, mm_cpumask(mm))
986 * clear_tasks_mm_cpumask - Safely clear tasks' mm_cpumask for a CPU
989 * This function walks all processes, finds a valid mm struct for each one and
990 * then clears a corresponding bit in mm's cpumask. While this all sounds
991 * trivial, there are various non-obvious corner cases, which this function
992 * tries to solve in a safe manner.
994 * Also note that the function uses a somewhat relaxed locking scheme, so it may
995 * be called only for an already offlined CPU.
997 void clear_tasks_mm_cpumask(int cpu
)
999 struct task_struct
*p
;
1002 * This function is called after the cpu is taken down and marked
1003 * offline, so its not like new tasks will ever get this cpu set in
1004 * their mm mask. -- Peter Zijlstra
1005 * Thus, we may use rcu_read_lock() here, instead of grabbing
1006 * full-fledged tasklist_lock.
1008 WARN_ON(cpu_online(cpu
));
1010 for_each_process(p
) {
1011 struct task_struct
*t
;
1014 * Main thread might exit, but other threads may still have
1015 * a valid mm. Find one.
1017 t
= find_lock_task_mm(p
);
1020 arch_clear_mm_cpumask_cpu(cpu
, t
->mm
);
1026 /* Take this CPU down. */
1027 static int take_cpu_down(void *_param
)
1029 struct cpuhp_cpu_state
*st
= this_cpu_ptr(&cpuhp_state
);
1030 enum cpuhp_state target
= max((int)st
->target
, CPUHP_AP_OFFLINE
);
1031 int err
, cpu
= smp_processor_id();
1033 /* Ensure this CPU doesn't handle any more interrupts. */
1034 err
= __cpu_disable();
1039 * Must be called from CPUHP_TEARDOWN_CPU, which means, as we are going
1040 * down, that the current state is CPUHP_TEARDOWN_CPU - 1.
1042 WARN_ON(st
->state
!= (CPUHP_TEARDOWN_CPU
- 1));
1045 * Invoke the former CPU_DYING callbacks. DYING must not fail!
1047 cpuhp_invoke_callback_range_nofail(false, cpu
, st
, target
);
1049 /* Give up timekeeping duties */
1050 tick_handover_do_timer();
1051 /* Remove CPU from timer broadcasting */
1052 tick_offline_cpu(cpu
);
1053 /* Park the stopper thread */
1054 stop_machine_park(cpu
);
1058 static int takedown_cpu(unsigned int cpu
)
1060 struct cpuhp_cpu_state
*st
= per_cpu_ptr(&cpuhp_state
, cpu
);
1063 /* Park the smpboot threads */
1064 kthread_park(st
->thread
);
1067 * Prevent irq alloc/free while the dying cpu reorganizes the
1068 * interrupt affinities.
1073 * So now all preempt/rcu users must observe !cpu_active().
1075 err
= stop_machine_cpuslocked(take_cpu_down
, NULL
, cpumask_of(cpu
));
1077 /* CPU refused to die */
1078 irq_unlock_sparse();
1079 /* Unpark the hotplug thread so we can rollback there */
1080 kthread_unpark(st
->thread
);
1083 BUG_ON(cpu_online(cpu
));
1086 * The teardown callback for CPUHP_AP_SCHED_STARTING will have removed
1087 * all runnable tasks from the CPU, there's only the idle task left now
1088 * that the migration thread is done doing the stop_machine thing.
1090 * Wait for the stop thread to go away.
1092 wait_for_ap_thread(st
, false);
1093 BUG_ON(st
->state
!= CPUHP_AP_IDLE_DEAD
);
1095 /* Interrupts are moved away from the dying cpu, reenable alloc/free */
1096 irq_unlock_sparse();
1098 hotplug_cpu__broadcast_tick_pull(cpu
);
1099 /* This actually kills the CPU. */
1102 tick_cleanup_dead_cpu(cpu
);
1103 rcutree_migrate_callbacks(cpu
);
1107 static void cpuhp_complete_idle_dead(void *arg
)
1109 struct cpuhp_cpu_state
*st
= arg
;
1111 complete_ap_thread(st
, false);
1114 void cpuhp_report_idle_dead(void)
1116 struct cpuhp_cpu_state
*st
= this_cpu_ptr(&cpuhp_state
);
1118 BUG_ON(st
->state
!= CPUHP_AP_OFFLINE
);
1119 rcu_report_dead(smp_processor_id());
1120 st
->state
= CPUHP_AP_IDLE_DEAD
;
1122 * We cannot call complete after rcu_report_dead() so we delegate it
1125 smp_call_function_single(cpumask_first(cpu_online_mask
),
1126 cpuhp_complete_idle_dead
, st
, 0);
1129 static int cpuhp_down_callbacks(unsigned int cpu
, struct cpuhp_cpu_state
*st
,
1130 enum cpuhp_state target
)
1132 enum cpuhp_state prev_state
= st
->state
;
1135 ret
= cpuhp_invoke_callback_range(false, cpu
, st
, target
);
1137 pr_debug("CPU DOWN failed (%d) CPU %u state %s (%d)\n",
1138 ret
, cpu
, cpuhp_get_step(st
->state
)->name
,
1141 cpuhp_reset_state(cpu
, st
, prev_state
);
1143 if (st
->state
< prev_state
)
1144 WARN_ON(cpuhp_invoke_callback_range(true, cpu
, st
,
1151 /* Requires cpu_add_remove_lock to be held */
1152 static int __ref
_cpu_down(unsigned int cpu
, int tasks_frozen
,
1153 enum cpuhp_state target
)
1155 struct cpuhp_cpu_state
*st
= per_cpu_ptr(&cpuhp_state
, cpu
);
1156 int prev_state
, ret
= 0;
1158 if (num_online_cpus() == 1)
1161 if (!cpu_present(cpu
))
1166 cpuhp_tasks_frozen
= tasks_frozen
;
1168 prev_state
= cpuhp_set_state(cpu
, st
, target
);
1170 * If the current CPU state is in the range of the AP hotplug thread,
1171 * then we need to kick the thread.
1173 if (st
->state
> CPUHP_TEARDOWN_CPU
) {
1174 st
->target
= max((int)target
, CPUHP_TEARDOWN_CPU
);
1175 ret
= cpuhp_kick_ap_work(cpu
);
1177 * The AP side has done the error rollback already. Just
1178 * return the error code..
1184 * We might have stopped still in the range of the AP hotplug
1185 * thread. Nothing to do anymore.
1187 if (st
->state
> CPUHP_TEARDOWN_CPU
)
1190 st
->target
= target
;
1193 * The AP brought itself down to CPUHP_TEARDOWN_CPU. So we need
1194 * to do the further cleanups.
1196 ret
= cpuhp_down_callbacks(cpu
, st
, target
);
1197 if (ret
&& st
->state
< prev_state
) {
1198 if (st
->state
== CPUHP_TEARDOWN_CPU
) {
1199 cpuhp_reset_state(cpu
, st
, prev_state
);
1200 __cpuhp_kick_ap(st
);
1202 WARN(1, "DEAD callback error for CPU%d", cpu
);
1207 cpus_write_unlock();
1209 * Do post unplug cleanup. This is still protected against
1210 * concurrent CPU hotplug via cpu_add_remove_lock.
1212 lockup_detector_cleanup();
1214 cpu_up_down_serialize_trainwrecks(tasks_frozen
);
1218 static int cpu_down_maps_locked(unsigned int cpu
, enum cpuhp_state target
)
1221 * If the platform does not support hotplug, report it explicitly to
1222 * differentiate it from a transient offlining failure.
1224 if (cc_platform_has(CC_ATTR_HOTPLUG_DISABLED
))
1226 if (cpu_hotplug_disabled
)
1228 return _cpu_down(cpu
, 0, target
);
1231 static int cpu_down(unsigned int cpu
, enum cpuhp_state target
)
1235 cpu_maps_update_begin();
1236 err
= cpu_down_maps_locked(cpu
, target
);
1237 cpu_maps_update_done();
1242 * cpu_device_down - Bring down a cpu device
1243 * @dev: Pointer to the cpu device to offline
1245 * This function is meant to be used by device core cpu subsystem only.
1247 * Other subsystems should use remove_cpu() instead.
1249 * Return: %0 on success or a negative errno code
1251 int cpu_device_down(struct device
*dev
)
1253 return cpu_down(dev
->id
, CPUHP_OFFLINE
);
1256 int remove_cpu(unsigned int cpu
)
1260 lock_device_hotplug();
1261 ret
= device_offline(get_cpu_device(cpu
));
1262 unlock_device_hotplug();
1266 EXPORT_SYMBOL_GPL(remove_cpu
);
1268 void smp_shutdown_nonboot_cpus(unsigned int primary_cpu
)
1273 cpu_maps_update_begin();
1276 * Make certain the cpu I'm about to reboot on is online.
1278 * This is inline to what migrate_to_reboot_cpu() already do.
1280 if (!cpu_online(primary_cpu
))
1281 primary_cpu
= cpumask_first(cpu_online_mask
);
1283 for_each_online_cpu(cpu
) {
1284 if (cpu
== primary_cpu
)
1287 error
= cpu_down_maps_locked(cpu
, CPUHP_OFFLINE
);
1289 pr_err("Failed to offline CPU%d - error=%d",
1296 * Ensure all but the reboot CPU are offline.
1298 BUG_ON(num_online_cpus() > 1);
1301 * Make sure the CPUs won't be enabled by someone else after this
1302 * point. Kexec will reboot to a new kernel shortly resetting
1303 * everything along the way.
1305 cpu_hotplug_disabled
++;
1307 cpu_maps_update_done();
1311 #define takedown_cpu NULL
1312 #endif /*CONFIG_HOTPLUG_CPU*/
1315 * notify_cpu_starting(cpu) - Invoke the callbacks on the starting CPU
1316 * @cpu: cpu that just started
1318 * It must be called by the arch code on the new cpu, before the new cpu
1319 * enables interrupts and before the "boot" cpu returns from __cpu_up().
1321 void notify_cpu_starting(unsigned int cpu
)
1323 struct cpuhp_cpu_state
*st
= per_cpu_ptr(&cpuhp_state
, cpu
);
1324 enum cpuhp_state target
= min((int)st
->target
, CPUHP_AP_ONLINE
);
1326 rcu_cpu_starting(cpu
); /* Enables RCU usage on this CPU. */
1327 cpumask_set_cpu(cpu
, &cpus_booted_once_mask
);
1330 * STARTING must not fail!
1332 cpuhp_invoke_callback_range_nofail(true, cpu
, st
, target
);
1336 * Called from the idle task. Wake up the controlling task which brings the
1337 * hotplug thread of the upcoming CPU up and then delegates the rest of the
1338 * online bringup to the hotplug thread.
1340 void cpuhp_online_idle(enum cpuhp_state state
)
1342 struct cpuhp_cpu_state
*st
= this_cpu_ptr(&cpuhp_state
);
1344 /* Happens for the boot cpu */
1345 if (state
!= CPUHP_AP_ONLINE_IDLE
)
1349 * Unpart the stopper thread before we start the idle loop (and start
1350 * scheduling); this ensures the stopper task is always available.
1352 stop_machine_unpark(smp_processor_id());
1354 st
->state
= CPUHP_AP_ONLINE_IDLE
;
1355 complete_ap_thread(st
, true);
1358 /* Requires cpu_add_remove_lock to be held */
1359 static int _cpu_up(unsigned int cpu
, int tasks_frozen
, enum cpuhp_state target
)
1361 struct cpuhp_cpu_state
*st
= per_cpu_ptr(&cpuhp_state
, cpu
);
1362 struct task_struct
*idle
;
1367 if (!cpu_present(cpu
)) {
1373 * The caller of cpu_up() might have raced with another
1374 * caller. Nothing to do.
1376 if (st
->state
>= target
)
1379 if (st
->state
== CPUHP_OFFLINE
) {
1380 /* Let it fail before we try to bring the cpu up */
1381 idle
= idle_thread_get(cpu
);
1383 ret
= PTR_ERR(idle
);
1388 cpuhp_tasks_frozen
= tasks_frozen
;
1390 cpuhp_set_state(cpu
, st
, target
);
1392 * If the current CPU state is in the range of the AP hotplug thread,
1393 * then we need to kick the thread once more.
1395 if (st
->state
> CPUHP_BRINGUP_CPU
) {
1396 ret
= cpuhp_kick_ap_work(cpu
);
1398 * The AP side has done the error rollback already. Just
1399 * return the error code..
1406 * Try to reach the target state. We max out on the BP at
1407 * CPUHP_BRINGUP_CPU. After that the AP hotplug thread is
1408 * responsible for bringing it up to the target state.
1410 target
= min((int)target
, CPUHP_BRINGUP_CPU
);
1411 ret
= cpuhp_up_callbacks(cpu
, st
, target
);
1413 cpus_write_unlock();
1415 cpu_up_down_serialize_trainwrecks(tasks_frozen
);
1419 static int cpu_up(unsigned int cpu
, enum cpuhp_state target
)
1423 if (!cpu_possible(cpu
)) {
1424 pr_err("can't online cpu %d because it is not configured as may-hotadd at boot time\n",
1426 #if defined(CONFIG_IA64)
1427 pr_err("please check additional_cpus= boot parameter\n");
1432 err
= try_online_node(cpu_to_node(cpu
));
1436 cpu_maps_update_begin();
1438 if (cpu_hotplug_disabled
) {
1442 if (!cpu_smt_allowed(cpu
)) {
1447 err
= _cpu_up(cpu
, 0, target
);
1449 cpu_maps_update_done();
1454 * cpu_device_up - Bring up a cpu device
1455 * @dev: Pointer to the cpu device to online
1457 * This function is meant to be used by device core cpu subsystem only.
1459 * Other subsystems should use add_cpu() instead.
1461 * Return: %0 on success or a negative errno code
1463 int cpu_device_up(struct device
*dev
)
1465 return cpu_up(dev
->id
, CPUHP_ONLINE
);
1468 int add_cpu(unsigned int cpu
)
1472 lock_device_hotplug();
1473 ret
= device_online(get_cpu_device(cpu
));
1474 unlock_device_hotplug();
1478 EXPORT_SYMBOL_GPL(add_cpu
);
1481 * bringup_hibernate_cpu - Bring up the CPU that we hibernated on
1482 * @sleep_cpu: The cpu we hibernated on and should be brought up.
1484 * On some architectures like arm64, we can hibernate on any CPU, but on
1485 * wake up the CPU we hibernated on might be offline as a side effect of
1486 * using maxcpus= for example.
1488 * Return: %0 on success or a negative errno code
1490 int bringup_hibernate_cpu(unsigned int sleep_cpu
)
1494 if (!cpu_online(sleep_cpu
)) {
1495 pr_info("Hibernated on a CPU that is offline! Bringing CPU up.\n");
1496 ret
= cpu_up(sleep_cpu
, CPUHP_ONLINE
);
1498 pr_err("Failed to bring hibernate-CPU up!\n");
1505 void bringup_nonboot_cpus(unsigned int setup_max_cpus
)
1509 for_each_present_cpu(cpu
) {
1510 if (num_online_cpus() >= setup_max_cpus
)
1512 if (!cpu_online(cpu
))
1513 cpu_up(cpu
, CPUHP_ONLINE
);
1517 #ifdef CONFIG_PM_SLEEP_SMP
1518 static cpumask_var_t frozen_cpus
;
1520 int freeze_secondary_cpus(int primary
)
1524 cpu_maps_update_begin();
1525 if (primary
== -1) {
1526 primary
= cpumask_first(cpu_online_mask
);
1527 if (!housekeeping_cpu(primary
, HK_TYPE_TIMER
))
1528 primary
= housekeeping_any_cpu(HK_TYPE_TIMER
);
1530 if (!cpu_online(primary
))
1531 primary
= cpumask_first(cpu_online_mask
);
1535 * We take down all of the non-boot CPUs in one shot to avoid races
1536 * with the userspace trying to use the CPU hotplug at the same time
1538 cpumask_clear(frozen_cpus
);
1540 pr_info("Disabling non-boot CPUs ...\n");
1541 for_each_online_cpu(cpu
) {
1545 if (pm_wakeup_pending()) {
1546 pr_info("Wakeup pending. Abort CPU freeze\n");
1551 trace_suspend_resume(TPS("CPU_OFF"), cpu
, true);
1552 error
= _cpu_down(cpu
, 1, CPUHP_OFFLINE
);
1553 trace_suspend_resume(TPS("CPU_OFF"), cpu
, false);
1555 cpumask_set_cpu(cpu
, frozen_cpus
);
1557 pr_err("Error taking CPU%d down: %d\n", cpu
, error
);
1563 BUG_ON(num_online_cpus() > 1);
1565 pr_err("Non-boot CPUs are not disabled\n");
1568 * Make sure the CPUs won't be enabled by someone else. We need to do
1569 * this even in case of failure as all freeze_secondary_cpus() users are
1570 * supposed to do thaw_secondary_cpus() on the failure path.
1572 cpu_hotplug_disabled
++;
1574 cpu_maps_update_done();
1578 void __weak
arch_thaw_secondary_cpus_begin(void)
1582 void __weak
arch_thaw_secondary_cpus_end(void)
1586 void thaw_secondary_cpus(void)
1590 /* Allow everyone to use the CPU hotplug again */
1591 cpu_maps_update_begin();
1592 __cpu_hotplug_enable();
1593 if (cpumask_empty(frozen_cpus
))
1596 pr_info("Enabling non-boot CPUs ...\n");
1598 arch_thaw_secondary_cpus_begin();
1600 for_each_cpu(cpu
, frozen_cpus
) {
1601 trace_suspend_resume(TPS("CPU_ON"), cpu
, true);
1602 error
= _cpu_up(cpu
, 1, CPUHP_ONLINE
);
1603 trace_suspend_resume(TPS("CPU_ON"), cpu
, false);
1605 pr_info("CPU%d is up\n", cpu
);
1608 pr_warn("Error taking CPU%d up: %d\n", cpu
, error
);
1611 arch_thaw_secondary_cpus_end();
1613 cpumask_clear(frozen_cpus
);
1615 cpu_maps_update_done();
1618 static int __init
alloc_frozen_cpus(void)
1620 if (!alloc_cpumask_var(&frozen_cpus
, GFP_KERNEL
|__GFP_ZERO
))
1624 core_initcall(alloc_frozen_cpus
);
1627 * When callbacks for CPU hotplug notifications are being executed, we must
1628 * ensure that the state of the system with respect to the tasks being frozen
1629 * or not, as reported by the notification, remains unchanged *throughout the
1630 * duration* of the execution of the callbacks.
1631 * Hence we need to prevent the freezer from racing with regular CPU hotplug.
1633 * This synchronization is implemented by mutually excluding regular CPU
1634 * hotplug and Suspend/Hibernate call paths by hooking onto the Suspend/
1635 * Hibernate notifications.
1638 cpu_hotplug_pm_callback(struct notifier_block
*nb
,
1639 unsigned long action
, void *ptr
)
1643 case PM_SUSPEND_PREPARE
:
1644 case PM_HIBERNATION_PREPARE
:
1645 cpu_hotplug_disable();
1648 case PM_POST_SUSPEND
:
1649 case PM_POST_HIBERNATION
:
1650 cpu_hotplug_enable();
1661 static int __init
cpu_hotplug_pm_sync_init(void)
1664 * cpu_hotplug_pm_callback has higher priority than x86
1665 * bsp_pm_callback which depends on cpu_hotplug_pm_callback
1666 * to disable cpu hotplug to avoid cpu hotplug race.
1668 pm_notifier(cpu_hotplug_pm_callback
, 0);
1671 core_initcall(cpu_hotplug_pm_sync_init
);
1673 #endif /* CONFIG_PM_SLEEP_SMP */
1677 #endif /* CONFIG_SMP */
1679 /* Boot processor state steps */
1680 static struct cpuhp_step cpuhp_hp_states
[] = {
1683 .startup
.single
= NULL
,
1684 .teardown
.single
= NULL
,
1687 [CPUHP_CREATE_THREADS
]= {
1688 .name
= "threads:prepare",
1689 .startup
.single
= smpboot_create_threads
,
1690 .teardown
.single
= NULL
,
1693 [CPUHP_PERF_PREPARE
] = {
1694 .name
= "perf:prepare",
1695 .startup
.single
= perf_event_init_cpu
,
1696 .teardown
.single
= perf_event_exit_cpu
,
1698 [CPUHP_RANDOM_PREPARE
] = {
1699 .name
= "random:prepare",
1700 .startup
.single
= random_prepare_cpu
,
1701 .teardown
.single
= NULL
,
1703 [CPUHP_WORKQUEUE_PREP
] = {
1704 .name
= "workqueue:prepare",
1705 .startup
.single
= workqueue_prepare_cpu
,
1706 .teardown
.single
= NULL
,
1708 [CPUHP_HRTIMERS_PREPARE
] = {
1709 .name
= "hrtimers:prepare",
1710 .startup
.single
= hrtimers_prepare_cpu
,
1711 .teardown
.single
= hrtimers_dead_cpu
,
1713 [CPUHP_SMPCFD_PREPARE
] = {
1714 .name
= "smpcfd:prepare",
1715 .startup
.single
= smpcfd_prepare_cpu
,
1716 .teardown
.single
= smpcfd_dead_cpu
,
1718 [CPUHP_RELAY_PREPARE
] = {
1719 .name
= "relay:prepare",
1720 .startup
.single
= relay_prepare_cpu
,
1721 .teardown
.single
= NULL
,
1723 [CPUHP_SLAB_PREPARE
] = {
1724 .name
= "slab:prepare",
1725 .startup
.single
= slab_prepare_cpu
,
1726 .teardown
.single
= slab_dead_cpu
,
1728 [CPUHP_RCUTREE_PREP
] = {
1729 .name
= "RCU/tree:prepare",
1730 .startup
.single
= rcutree_prepare_cpu
,
1731 .teardown
.single
= rcutree_dead_cpu
,
1734 * On the tear-down path, timers_dead_cpu() must be invoked
1735 * before blk_mq_queue_reinit_notify() from notify_dead(),
1736 * otherwise a RCU stall occurs.
1738 [CPUHP_TIMERS_PREPARE
] = {
1739 .name
= "timers:prepare",
1740 .startup
.single
= timers_prepare_cpu
,
1741 .teardown
.single
= timers_dead_cpu
,
1743 /* Kicks the plugged cpu into life */
1744 [CPUHP_BRINGUP_CPU
] = {
1745 .name
= "cpu:bringup",
1746 .startup
.single
= bringup_cpu
,
1747 .teardown
.single
= finish_cpu
,
1750 /* Final state before CPU kills itself */
1751 [CPUHP_AP_IDLE_DEAD
] = {
1752 .name
= "idle:dead",
1755 * Last state before CPU enters the idle loop to die. Transient state
1756 * for synchronization.
1758 [CPUHP_AP_OFFLINE
] = {
1759 .name
= "ap:offline",
1762 /* First state is scheduler control. Interrupts are disabled */
1763 [CPUHP_AP_SCHED_STARTING
] = {
1764 .name
= "sched:starting",
1765 .startup
.single
= sched_cpu_starting
,
1766 .teardown
.single
= sched_cpu_dying
,
1768 [CPUHP_AP_RCUTREE_DYING
] = {
1769 .name
= "RCU/tree:dying",
1770 .startup
.single
= NULL
,
1771 .teardown
.single
= rcutree_dying_cpu
,
1773 [CPUHP_AP_SMPCFD_DYING
] = {
1774 .name
= "smpcfd:dying",
1775 .startup
.single
= NULL
,
1776 .teardown
.single
= smpcfd_dying_cpu
,
1778 /* Entry state on starting. Interrupts enabled from here on. Transient
1779 * state for synchronsization */
1780 [CPUHP_AP_ONLINE
] = {
1781 .name
= "ap:online",
1784 * Handled on control processor until the plugged processor manages
1787 [CPUHP_TEARDOWN_CPU
] = {
1788 .name
= "cpu:teardown",
1789 .startup
.single
= NULL
,
1790 .teardown
.single
= takedown_cpu
,
1794 [CPUHP_AP_SCHED_WAIT_EMPTY
] = {
1795 .name
= "sched:waitempty",
1796 .startup
.single
= NULL
,
1797 .teardown
.single
= sched_cpu_wait_empty
,
1800 /* Handle smpboot threads park/unpark */
1801 [CPUHP_AP_SMPBOOT_THREADS
] = {
1802 .name
= "smpboot/threads:online",
1803 .startup
.single
= smpboot_unpark_threads
,
1804 .teardown
.single
= smpboot_park_threads
,
1806 [CPUHP_AP_IRQ_AFFINITY_ONLINE
] = {
1807 .name
= "irq/affinity:online",
1808 .startup
.single
= irq_affinity_online_cpu
,
1809 .teardown
.single
= NULL
,
1811 [CPUHP_AP_PERF_ONLINE
] = {
1812 .name
= "perf:online",
1813 .startup
.single
= perf_event_init_cpu
,
1814 .teardown
.single
= perf_event_exit_cpu
,
1816 [CPUHP_AP_WATCHDOG_ONLINE
] = {
1817 .name
= "lockup_detector:online",
1818 .startup
.single
= lockup_detector_online_cpu
,
1819 .teardown
.single
= lockup_detector_offline_cpu
,
1821 [CPUHP_AP_WORKQUEUE_ONLINE
] = {
1822 .name
= "workqueue:online",
1823 .startup
.single
= workqueue_online_cpu
,
1824 .teardown
.single
= workqueue_offline_cpu
,
1826 [CPUHP_AP_RANDOM_ONLINE
] = {
1827 .name
= "random:online",
1828 .startup
.single
= random_online_cpu
,
1829 .teardown
.single
= NULL
,
1831 [CPUHP_AP_RCUTREE_ONLINE
] = {
1832 .name
= "RCU/tree:online",
1833 .startup
.single
= rcutree_online_cpu
,
1834 .teardown
.single
= rcutree_offline_cpu
,
1838 * The dynamically registered state space is here
1842 /* Last state is scheduler control setting the cpu active */
1843 [CPUHP_AP_ACTIVE
] = {
1844 .name
= "sched:active",
1845 .startup
.single
= sched_cpu_activate
,
1846 .teardown
.single
= sched_cpu_deactivate
,
1850 /* CPU is fully up and running. */
1853 .startup
.single
= NULL
,
1854 .teardown
.single
= NULL
,
1858 /* Sanity check for callbacks */
1859 static int cpuhp_cb_check(enum cpuhp_state state
)
1861 if (state
<= CPUHP_OFFLINE
|| state
>= CPUHP_ONLINE
)
1867 * Returns a free for dynamic slot assignment of the Online state. The states
1868 * are protected by the cpuhp_slot_states mutex and an empty slot is identified
1869 * by having no name assigned.
1871 static int cpuhp_reserve_state(enum cpuhp_state state
)
1873 enum cpuhp_state i
, end
;
1874 struct cpuhp_step
*step
;
1877 case CPUHP_AP_ONLINE_DYN
:
1878 step
= cpuhp_hp_states
+ CPUHP_AP_ONLINE_DYN
;
1879 end
= CPUHP_AP_ONLINE_DYN_END
;
1881 case CPUHP_BP_PREPARE_DYN
:
1882 step
= cpuhp_hp_states
+ CPUHP_BP_PREPARE_DYN
;
1883 end
= CPUHP_BP_PREPARE_DYN_END
;
1889 for (i
= state
; i
<= end
; i
++, step
++) {
1893 WARN(1, "No more dynamic states available for CPU hotplug\n");
1897 static int cpuhp_store_callbacks(enum cpuhp_state state
, const char *name
,
1898 int (*startup
)(unsigned int cpu
),
1899 int (*teardown
)(unsigned int cpu
),
1900 bool multi_instance
)
1902 /* (Un)Install the callbacks for further cpu hotplug operations */
1903 struct cpuhp_step
*sp
;
1907 * If name is NULL, then the state gets removed.
1909 * CPUHP_AP_ONLINE_DYN and CPUHP_BP_PREPARE_DYN are handed out on
1910 * the first allocation from these dynamic ranges, so the removal
1911 * would trigger a new allocation and clear the wrong (already
1912 * empty) state, leaving the callbacks of the to be cleared state
1913 * dangling, which causes wreckage on the next hotplug operation.
1915 if (name
&& (state
== CPUHP_AP_ONLINE_DYN
||
1916 state
== CPUHP_BP_PREPARE_DYN
)) {
1917 ret
= cpuhp_reserve_state(state
);
1922 sp
= cpuhp_get_step(state
);
1923 if (name
&& sp
->name
)
1926 sp
->startup
.single
= startup
;
1927 sp
->teardown
.single
= teardown
;
1929 sp
->multi_instance
= multi_instance
;
1930 INIT_HLIST_HEAD(&sp
->list
);
1934 static void *cpuhp_get_teardown_cb(enum cpuhp_state state
)
1936 return cpuhp_get_step(state
)->teardown
.single
;
1940 * Call the startup/teardown function for a step either on the AP or
1941 * on the current CPU.
1943 static int cpuhp_issue_call(int cpu
, enum cpuhp_state state
, bool bringup
,
1944 struct hlist_node
*node
)
1946 struct cpuhp_step
*sp
= cpuhp_get_step(state
);
1950 * If there's nothing to do, we done.
1951 * Relies on the union for multi_instance.
1953 if (cpuhp_step_empty(bringup
, sp
))
1956 * The non AP bound callbacks can fail on bringup. On teardown
1957 * e.g. module removal we crash for now.
1960 if (cpuhp_is_ap_state(state
))
1961 ret
= cpuhp_invoke_ap_callback(cpu
, state
, bringup
, node
);
1963 ret
= cpuhp_invoke_callback(cpu
, state
, bringup
, node
, NULL
);
1965 ret
= cpuhp_invoke_callback(cpu
, state
, bringup
, node
, NULL
);
1967 BUG_ON(ret
&& !bringup
);
1972 * Called from __cpuhp_setup_state on a recoverable failure.
1974 * Note: The teardown callbacks for rollback are not allowed to fail!
1976 static void cpuhp_rollback_install(int failedcpu
, enum cpuhp_state state
,
1977 struct hlist_node
*node
)
1981 /* Roll back the already executed steps on the other cpus */
1982 for_each_present_cpu(cpu
) {
1983 struct cpuhp_cpu_state
*st
= per_cpu_ptr(&cpuhp_state
, cpu
);
1984 int cpustate
= st
->state
;
1986 if (cpu
>= failedcpu
)
1989 /* Did we invoke the startup call on that cpu ? */
1990 if (cpustate
>= state
)
1991 cpuhp_issue_call(cpu
, state
, false, node
);
1995 int __cpuhp_state_add_instance_cpuslocked(enum cpuhp_state state
,
1996 struct hlist_node
*node
,
1999 struct cpuhp_step
*sp
;
2003 lockdep_assert_cpus_held();
2005 sp
= cpuhp_get_step(state
);
2006 if (sp
->multi_instance
== false)
2009 mutex_lock(&cpuhp_state_mutex
);
2011 if (!invoke
|| !sp
->startup
.multi
)
2015 * Try to call the startup callback for each present cpu
2016 * depending on the hotplug state of the cpu.
2018 for_each_present_cpu(cpu
) {
2019 struct cpuhp_cpu_state
*st
= per_cpu_ptr(&cpuhp_state
, cpu
);
2020 int cpustate
= st
->state
;
2022 if (cpustate
< state
)
2025 ret
= cpuhp_issue_call(cpu
, state
, true, node
);
2027 if (sp
->teardown
.multi
)
2028 cpuhp_rollback_install(cpu
, state
, node
);
2034 hlist_add_head(node
, &sp
->list
);
2036 mutex_unlock(&cpuhp_state_mutex
);
2040 int __cpuhp_state_add_instance(enum cpuhp_state state
, struct hlist_node
*node
,
2046 ret
= __cpuhp_state_add_instance_cpuslocked(state
, node
, invoke
);
2050 EXPORT_SYMBOL_GPL(__cpuhp_state_add_instance
);
2053 * __cpuhp_setup_state_cpuslocked - Setup the callbacks for an hotplug machine state
2054 * @state: The state to setup
2055 * @name: Name of the step
2056 * @invoke: If true, the startup function is invoked for cpus where
2057 * cpu state >= @state
2058 * @startup: startup callback function
2059 * @teardown: teardown callback function
2060 * @multi_instance: State is set up for multiple instances which get
2063 * The caller needs to hold cpus read locked while calling this function.
2066 * Positive state number if @state is CPUHP_AP_ONLINE_DYN;
2067 * 0 for all other states
2068 * On failure: proper (negative) error code
2070 int __cpuhp_setup_state_cpuslocked(enum cpuhp_state state
,
2071 const char *name
, bool invoke
,
2072 int (*startup
)(unsigned int cpu
),
2073 int (*teardown
)(unsigned int cpu
),
2074 bool multi_instance
)
2079 lockdep_assert_cpus_held();
2081 if (cpuhp_cb_check(state
) || !name
)
2084 mutex_lock(&cpuhp_state_mutex
);
2086 ret
= cpuhp_store_callbacks(state
, name
, startup
, teardown
,
2089 dynstate
= state
== CPUHP_AP_ONLINE_DYN
;
2090 if (ret
> 0 && dynstate
) {
2095 if (ret
|| !invoke
|| !startup
)
2099 * Try to call the startup callback for each present cpu
2100 * depending on the hotplug state of the cpu.
2102 for_each_present_cpu(cpu
) {
2103 struct cpuhp_cpu_state
*st
= per_cpu_ptr(&cpuhp_state
, cpu
);
2104 int cpustate
= st
->state
;
2106 if (cpustate
< state
)
2109 ret
= cpuhp_issue_call(cpu
, state
, true, NULL
);
2112 cpuhp_rollback_install(cpu
, state
, NULL
);
2113 cpuhp_store_callbacks(state
, NULL
, NULL
, NULL
, false);
2118 mutex_unlock(&cpuhp_state_mutex
);
2120 * If the requested state is CPUHP_AP_ONLINE_DYN, return the
2121 * dynamically allocated state in case of success.
2123 if (!ret
&& dynstate
)
2127 EXPORT_SYMBOL(__cpuhp_setup_state_cpuslocked
);
2129 int __cpuhp_setup_state(enum cpuhp_state state
,
2130 const char *name
, bool invoke
,
2131 int (*startup
)(unsigned int cpu
),
2132 int (*teardown
)(unsigned int cpu
),
2133 bool multi_instance
)
2138 ret
= __cpuhp_setup_state_cpuslocked(state
, name
, invoke
, startup
,
2139 teardown
, multi_instance
);
2143 EXPORT_SYMBOL(__cpuhp_setup_state
);
2145 int __cpuhp_state_remove_instance(enum cpuhp_state state
,
2146 struct hlist_node
*node
, bool invoke
)
2148 struct cpuhp_step
*sp
= cpuhp_get_step(state
);
2151 BUG_ON(cpuhp_cb_check(state
));
2153 if (!sp
->multi_instance
)
2157 mutex_lock(&cpuhp_state_mutex
);
2159 if (!invoke
|| !cpuhp_get_teardown_cb(state
))
2162 * Call the teardown callback for each present cpu depending
2163 * on the hotplug state of the cpu. This function is not
2164 * allowed to fail currently!
2166 for_each_present_cpu(cpu
) {
2167 struct cpuhp_cpu_state
*st
= per_cpu_ptr(&cpuhp_state
, cpu
);
2168 int cpustate
= st
->state
;
2170 if (cpustate
>= state
)
2171 cpuhp_issue_call(cpu
, state
, false, node
);
2176 mutex_unlock(&cpuhp_state_mutex
);
2181 EXPORT_SYMBOL_GPL(__cpuhp_state_remove_instance
);
2184 * __cpuhp_remove_state_cpuslocked - Remove the callbacks for an hotplug machine state
2185 * @state: The state to remove
2186 * @invoke: If true, the teardown function is invoked for cpus where
2187 * cpu state >= @state
2189 * The caller needs to hold cpus read locked while calling this function.
2190 * The teardown callback is currently not allowed to fail. Think
2191 * about module removal!
2193 void __cpuhp_remove_state_cpuslocked(enum cpuhp_state state
, bool invoke
)
2195 struct cpuhp_step
*sp
= cpuhp_get_step(state
);
2198 BUG_ON(cpuhp_cb_check(state
));
2200 lockdep_assert_cpus_held();
2202 mutex_lock(&cpuhp_state_mutex
);
2203 if (sp
->multi_instance
) {
2204 WARN(!hlist_empty(&sp
->list
),
2205 "Error: Removing state %d which has instances left.\n",
2210 if (!invoke
|| !cpuhp_get_teardown_cb(state
))
2214 * Call the teardown callback for each present cpu depending
2215 * on the hotplug state of the cpu. This function is not
2216 * allowed to fail currently!
2218 for_each_present_cpu(cpu
) {
2219 struct cpuhp_cpu_state
*st
= per_cpu_ptr(&cpuhp_state
, cpu
);
2220 int cpustate
= st
->state
;
2222 if (cpustate
>= state
)
2223 cpuhp_issue_call(cpu
, state
, false, NULL
);
2226 cpuhp_store_callbacks(state
, NULL
, NULL
, NULL
, false);
2227 mutex_unlock(&cpuhp_state_mutex
);
2229 EXPORT_SYMBOL(__cpuhp_remove_state_cpuslocked
);
2231 void __cpuhp_remove_state(enum cpuhp_state state
, bool invoke
)
2234 __cpuhp_remove_state_cpuslocked(state
, invoke
);
2237 EXPORT_SYMBOL(__cpuhp_remove_state
);
2239 #ifdef CONFIG_HOTPLUG_SMT
2240 static void cpuhp_offline_cpu_device(unsigned int cpu
)
2242 struct device
*dev
= get_cpu_device(cpu
);
2244 dev
->offline
= true;
2245 /* Tell user space about the state change */
2246 kobject_uevent(&dev
->kobj
, KOBJ_OFFLINE
);
2249 static void cpuhp_online_cpu_device(unsigned int cpu
)
2251 struct device
*dev
= get_cpu_device(cpu
);
2253 dev
->offline
= false;
2254 /* Tell user space about the state change */
2255 kobject_uevent(&dev
->kobj
, KOBJ_ONLINE
);
2258 int cpuhp_smt_disable(enum cpuhp_smt_control ctrlval
)
2262 cpu_maps_update_begin();
2263 for_each_online_cpu(cpu
) {
2264 if (topology_is_primary_thread(cpu
))
2266 ret
= cpu_down_maps_locked(cpu
, CPUHP_OFFLINE
);
2270 * As this needs to hold the cpu maps lock it's impossible
2271 * to call device_offline() because that ends up calling
2272 * cpu_down() which takes cpu maps lock. cpu maps lock
2273 * needs to be held as this might race against in kernel
2274 * abusers of the hotplug machinery (thermal management).
2276 * So nothing would update device:offline state. That would
2277 * leave the sysfs entry stale and prevent onlining after
2278 * smt control has been changed to 'off' again. This is
2279 * called under the sysfs hotplug lock, so it is properly
2280 * serialized against the regular offline usage.
2282 cpuhp_offline_cpu_device(cpu
);
2285 cpu_smt_control
= ctrlval
;
2286 cpu_maps_update_done();
2290 int cpuhp_smt_enable(void)
2294 cpu_maps_update_begin();
2295 cpu_smt_control
= CPU_SMT_ENABLED
;
2296 for_each_present_cpu(cpu
) {
2297 /* Skip online CPUs and CPUs on offline nodes */
2298 if (cpu_online(cpu
) || !node_online(cpu_to_node(cpu
)))
2300 ret
= _cpu_up(cpu
, 0, CPUHP_ONLINE
);
2303 /* See comment in cpuhp_smt_disable() */
2304 cpuhp_online_cpu_device(cpu
);
2306 cpu_maps_update_done();
2311 #if defined(CONFIG_SYSFS) && defined(CONFIG_HOTPLUG_CPU)
2312 static ssize_t
state_show(struct device
*dev
,
2313 struct device_attribute
*attr
, char *buf
)
2315 struct cpuhp_cpu_state
*st
= per_cpu_ptr(&cpuhp_state
, dev
->id
);
2317 return sprintf(buf
, "%d\n", st
->state
);
2319 static DEVICE_ATTR_RO(state
);
2321 static ssize_t
target_store(struct device
*dev
, struct device_attribute
*attr
,
2322 const char *buf
, size_t count
)
2324 struct cpuhp_cpu_state
*st
= per_cpu_ptr(&cpuhp_state
, dev
->id
);
2325 struct cpuhp_step
*sp
;
2328 ret
= kstrtoint(buf
, 10, &target
);
2332 #ifdef CONFIG_CPU_HOTPLUG_STATE_CONTROL
2333 if (target
< CPUHP_OFFLINE
|| target
> CPUHP_ONLINE
)
2336 if (target
!= CPUHP_OFFLINE
&& target
!= CPUHP_ONLINE
)
2340 ret
= lock_device_hotplug_sysfs();
2344 mutex_lock(&cpuhp_state_mutex
);
2345 sp
= cpuhp_get_step(target
);
2346 ret
= !sp
->name
|| sp
->cant_stop
? -EINVAL
: 0;
2347 mutex_unlock(&cpuhp_state_mutex
);
2351 if (st
->state
< target
)
2352 ret
= cpu_up(dev
->id
, target
);
2353 else if (st
->state
> target
)
2354 ret
= cpu_down(dev
->id
, target
);
2355 else if (WARN_ON(st
->target
!= target
))
2356 st
->target
= target
;
2358 unlock_device_hotplug();
2359 return ret
? ret
: count
;
2362 static ssize_t
target_show(struct device
*dev
,
2363 struct device_attribute
*attr
, char *buf
)
2365 struct cpuhp_cpu_state
*st
= per_cpu_ptr(&cpuhp_state
, dev
->id
);
2367 return sprintf(buf
, "%d\n", st
->target
);
2369 static DEVICE_ATTR_RW(target
);
2371 static ssize_t
fail_store(struct device
*dev
, struct device_attribute
*attr
,
2372 const char *buf
, size_t count
)
2374 struct cpuhp_cpu_state
*st
= per_cpu_ptr(&cpuhp_state
, dev
->id
);
2375 struct cpuhp_step
*sp
;
2378 ret
= kstrtoint(buf
, 10, &fail
);
2382 if (fail
== CPUHP_INVALID
) {
2387 if (fail
< CPUHP_OFFLINE
|| fail
> CPUHP_ONLINE
)
2391 * Cannot fail STARTING/DYING callbacks.
2393 if (cpuhp_is_atomic_state(fail
))
2397 * DEAD callbacks cannot fail...
2398 * ... neither can CPUHP_BRINGUP_CPU during hotunplug. The latter
2399 * triggering STARTING callbacks, a failure in this state would
2402 if (fail
<= CPUHP_BRINGUP_CPU
&& st
->state
> CPUHP_BRINGUP_CPU
)
2406 * Cannot fail anything that doesn't have callbacks.
2408 mutex_lock(&cpuhp_state_mutex
);
2409 sp
= cpuhp_get_step(fail
);
2410 if (!sp
->startup
.single
&& !sp
->teardown
.single
)
2412 mutex_unlock(&cpuhp_state_mutex
);
2421 static ssize_t
fail_show(struct device
*dev
,
2422 struct device_attribute
*attr
, char *buf
)
2424 struct cpuhp_cpu_state
*st
= per_cpu_ptr(&cpuhp_state
, dev
->id
);
2426 return sprintf(buf
, "%d\n", st
->fail
);
2429 static DEVICE_ATTR_RW(fail
);
2431 static struct attribute
*cpuhp_cpu_attrs
[] = {
2432 &dev_attr_state
.attr
,
2433 &dev_attr_target
.attr
,
2434 &dev_attr_fail
.attr
,
2438 static const struct attribute_group cpuhp_cpu_attr_group
= {
2439 .attrs
= cpuhp_cpu_attrs
,
2444 static ssize_t
states_show(struct device
*dev
,
2445 struct device_attribute
*attr
, char *buf
)
2447 ssize_t cur
, res
= 0;
2450 mutex_lock(&cpuhp_state_mutex
);
2451 for (i
= CPUHP_OFFLINE
; i
<= CPUHP_ONLINE
; i
++) {
2452 struct cpuhp_step
*sp
= cpuhp_get_step(i
);
2455 cur
= sprintf(buf
, "%3d: %s\n", i
, sp
->name
);
2460 mutex_unlock(&cpuhp_state_mutex
);
2463 static DEVICE_ATTR_RO(states
);
2465 static struct attribute
*cpuhp_cpu_root_attrs
[] = {
2466 &dev_attr_states
.attr
,
2470 static const struct attribute_group cpuhp_cpu_root_attr_group
= {
2471 .attrs
= cpuhp_cpu_root_attrs
,
2476 #ifdef CONFIG_HOTPLUG_SMT
2479 __store_smt_control(struct device
*dev
, struct device_attribute
*attr
,
2480 const char *buf
, size_t count
)
2484 if (sysfs_streq(buf
, "on"))
2485 ctrlval
= CPU_SMT_ENABLED
;
2486 else if (sysfs_streq(buf
, "off"))
2487 ctrlval
= CPU_SMT_DISABLED
;
2488 else if (sysfs_streq(buf
, "forceoff"))
2489 ctrlval
= CPU_SMT_FORCE_DISABLED
;
2493 if (cpu_smt_control
== CPU_SMT_FORCE_DISABLED
)
2496 if (cpu_smt_control
== CPU_SMT_NOT_SUPPORTED
)
2499 ret
= lock_device_hotplug_sysfs();
2503 if (ctrlval
!= cpu_smt_control
) {
2505 case CPU_SMT_ENABLED
:
2506 ret
= cpuhp_smt_enable();
2508 case CPU_SMT_DISABLED
:
2509 case CPU_SMT_FORCE_DISABLED
:
2510 ret
= cpuhp_smt_disable(ctrlval
);
2515 unlock_device_hotplug();
2516 return ret
? ret
: count
;
2519 #else /* !CONFIG_HOTPLUG_SMT */
2521 __store_smt_control(struct device
*dev
, struct device_attribute
*attr
,
2522 const char *buf
, size_t count
)
2526 #endif /* CONFIG_HOTPLUG_SMT */
2528 static const char *smt_states
[] = {
2529 [CPU_SMT_ENABLED
] = "on",
2530 [CPU_SMT_DISABLED
] = "off",
2531 [CPU_SMT_FORCE_DISABLED
] = "forceoff",
2532 [CPU_SMT_NOT_SUPPORTED
] = "notsupported",
2533 [CPU_SMT_NOT_IMPLEMENTED
] = "notimplemented",
2536 static ssize_t
control_show(struct device
*dev
,
2537 struct device_attribute
*attr
, char *buf
)
2539 const char *state
= smt_states
[cpu_smt_control
];
2541 return snprintf(buf
, PAGE_SIZE
- 2, "%s\n", state
);
2544 static ssize_t
control_store(struct device
*dev
, struct device_attribute
*attr
,
2545 const char *buf
, size_t count
)
2547 return __store_smt_control(dev
, attr
, buf
, count
);
2549 static DEVICE_ATTR_RW(control
);
2551 static ssize_t
active_show(struct device
*dev
,
2552 struct device_attribute
*attr
, char *buf
)
2554 return snprintf(buf
, PAGE_SIZE
- 2, "%d\n", sched_smt_active());
2556 static DEVICE_ATTR_RO(active
);
2558 static struct attribute
*cpuhp_smt_attrs
[] = {
2559 &dev_attr_control
.attr
,
2560 &dev_attr_active
.attr
,
2564 static const struct attribute_group cpuhp_smt_attr_group
= {
2565 .attrs
= cpuhp_smt_attrs
,
2570 static int __init
cpu_smt_sysfs_init(void)
2572 return sysfs_create_group(&cpu_subsys
.dev_root
->kobj
,
2573 &cpuhp_smt_attr_group
);
2576 static int __init
cpuhp_sysfs_init(void)
2580 ret
= cpu_smt_sysfs_init();
2584 ret
= sysfs_create_group(&cpu_subsys
.dev_root
->kobj
,
2585 &cpuhp_cpu_root_attr_group
);
2589 for_each_possible_cpu(cpu
) {
2590 struct device
*dev
= get_cpu_device(cpu
);
2594 ret
= sysfs_create_group(&dev
->kobj
, &cpuhp_cpu_attr_group
);
2600 device_initcall(cpuhp_sysfs_init
);
2601 #endif /* CONFIG_SYSFS && CONFIG_HOTPLUG_CPU */
2604 * cpu_bit_bitmap[] is a special, "compressed" data structure that
2605 * represents all NR_CPUS bits binary values of 1<<nr.
2607 * It is used by cpumask_of() to get a constant address to a CPU
2608 * mask value that has a single bit set only.
2611 /* cpu_bit_bitmap[0] is empty - so we can back into it */
2612 #define MASK_DECLARE_1(x) [x+1][0] = (1UL << (x))
2613 #define MASK_DECLARE_2(x) MASK_DECLARE_1(x), MASK_DECLARE_1(x+1)
2614 #define MASK_DECLARE_4(x) MASK_DECLARE_2(x), MASK_DECLARE_2(x+2)
2615 #define MASK_DECLARE_8(x) MASK_DECLARE_4(x), MASK_DECLARE_4(x+4)
2617 const unsigned long cpu_bit_bitmap
[BITS_PER_LONG
+1][BITS_TO_LONGS(NR_CPUS
)] = {
2619 MASK_DECLARE_8(0), MASK_DECLARE_8(8),
2620 MASK_DECLARE_8(16), MASK_DECLARE_8(24),
2621 #if BITS_PER_LONG > 32
2622 MASK_DECLARE_8(32), MASK_DECLARE_8(40),
2623 MASK_DECLARE_8(48), MASK_DECLARE_8(56),
2626 EXPORT_SYMBOL_GPL(cpu_bit_bitmap
);
2628 const DECLARE_BITMAP(cpu_all_bits
, NR_CPUS
) = CPU_BITS_ALL
;
2629 EXPORT_SYMBOL(cpu_all_bits
);
2631 #ifdef CONFIG_INIT_ALL_POSSIBLE
2632 struct cpumask __cpu_possible_mask __read_mostly
2635 struct cpumask __cpu_possible_mask __read_mostly
;
2637 EXPORT_SYMBOL(__cpu_possible_mask
);
2639 struct cpumask __cpu_online_mask __read_mostly
;
2640 EXPORT_SYMBOL(__cpu_online_mask
);
2642 struct cpumask __cpu_present_mask __read_mostly
;
2643 EXPORT_SYMBOL(__cpu_present_mask
);
2645 struct cpumask __cpu_active_mask __read_mostly
;
2646 EXPORT_SYMBOL(__cpu_active_mask
);
2648 struct cpumask __cpu_dying_mask __read_mostly
;
2649 EXPORT_SYMBOL(__cpu_dying_mask
);
2651 atomic_t __num_online_cpus __read_mostly
;
2652 EXPORT_SYMBOL(__num_online_cpus
);
2654 void init_cpu_present(const struct cpumask
*src
)
2656 cpumask_copy(&__cpu_present_mask
, src
);
2659 void init_cpu_possible(const struct cpumask
*src
)
2661 cpumask_copy(&__cpu_possible_mask
, src
);
2664 void init_cpu_online(const struct cpumask
*src
)
2666 cpumask_copy(&__cpu_online_mask
, src
);
2669 void set_cpu_online(unsigned int cpu
, bool online
)
2672 * atomic_inc/dec() is required to handle the horrid abuse of this
2673 * function by the reboot and kexec code which invoke it from
2674 * IPI/NMI broadcasts when shutting down CPUs. Invocation from
2675 * regular CPU hotplug is properly serialized.
2677 * Note, that the fact that __num_online_cpus is of type atomic_t
2678 * does not protect readers which are not serialized against
2679 * concurrent hotplug operations.
2682 if (!cpumask_test_and_set_cpu(cpu
, &__cpu_online_mask
))
2683 atomic_inc(&__num_online_cpus
);
2685 if (cpumask_test_and_clear_cpu(cpu
, &__cpu_online_mask
))
2686 atomic_dec(&__num_online_cpus
);
2691 * Activate the first processor.
2693 void __init
boot_cpu_init(void)
2695 int cpu
= smp_processor_id();
2697 /* Mark the boot cpu "present", "online" etc for SMP and UP case */
2698 set_cpu_online(cpu
, true);
2699 set_cpu_active(cpu
, true);
2700 set_cpu_present(cpu
, true);
2701 set_cpu_possible(cpu
, true);
2704 __boot_cpu_id
= cpu
;
2709 * Must be called _AFTER_ setting up the per_cpu areas
2711 void __init
boot_cpu_hotplug_init(void)
2714 cpumask_set_cpu(smp_processor_id(), &cpus_booted_once_mask
);
2716 this_cpu_write(cpuhp_state
.state
, CPUHP_ONLINE
);
2717 this_cpu_write(cpuhp_state
.target
, CPUHP_ONLINE
);
2721 * These are used for a global "mitigations=" cmdline option for toggling
2722 * optional CPU mitigations.
2724 enum cpu_mitigations
{
2725 CPU_MITIGATIONS_OFF
,
2726 CPU_MITIGATIONS_AUTO
,
2727 CPU_MITIGATIONS_AUTO_NOSMT
,
2730 static enum cpu_mitigations cpu_mitigations __ro_after_init
=
2731 CPU_MITIGATIONS_AUTO
;
2733 static int __init
mitigations_parse_cmdline(char *arg
)
2735 if (!strcmp(arg
, "off"))
2736 cpu_mitigations
= CPU_MITIGATIONS_OFF
;
2737 else if (!strcmp(arg
, "auto"))
2738 cpu_mitigations
= CPU_MITIGATIONS_AUTO
;
2739 else if (!strcmp(arg
, "auto,nosmt"))
2740 cpu_mitigations
= CPU_MITIGATIONS_AUTO_NOSMT
;
2742 pr_crit("Unsupported mitigations=%s, system may still be vulnerable\n",
2747 early_param("mitigations", mitigations_parse_cmdline
);
2749 /* mitigations=off */
2750 bool cpu_mitigations_off(void)
2752 return cpu_mitigations
== CPU_MITIGATIONS_OFF
;
2754 EXPORT_SYMBOL_GPL(cpu_mitigations_off
);
2756 /* mitigations=auto,nosmt */
2757 bool cpu_mitigations_auto_nosmt(void)
2759 return cpu_mitigations
== CPU_MITIGATIONS_AUTO_NOSMT
;
2761 EXPORT_SYMBOL_GPL(cpu_mitigations_auto_nosmt
);