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/unistd.h>
14 #include <linux/cpu.h>
15 #include <linux/oom.h>
16 #include <linux/rcupdate.h>
17 #include <linux/export.h>
18 #include <linux/bug.h>
19 #include <linux/kthread.h>
20 #include <linux/stop_machine.h>
21 #include <linux/mutex.h>
22 #include <linux/gfp.h>
23 #include <linux/suspend.h>
24 #include <linux/lockdep.h>
25 #include <linux/tick.h>
26 #include <linux/irq.h>
27 #include <linux/nmi.h>
28 #include <linux/smpboot.h>
29 #include <linux/relay.h>
30 #include <linux/slab.h>
31 #include <linux/percpu-rwsem.h>
33 #include <trace/events/power.h>
34 #define CREATE_TRACE_POINTS
35 #include <trace/events/cpuhp.h>
40 * cpuhp_cpu_state - Per cpu hotplug state storage
41 * @state: The current cpu state
42 * @target: The target state
43 * @thread: Pointer to the hotplug thread
44 * @should_run: Thread should execute
45 * @rollback: Perform a rollback
46 * @single: Single callback invocation
47 * @bringup: Single callback bringup or teardown selector
48 * @cb_state: The state for a single callback (install/uninstall)
49 * @result: Result of the operation
50 * @done_up: Signal completion to the issuer of the task for cpu-up
51 * @done_down: Signal completion to the issuer of the task for cpu-down
53 struct cpuhp_cpu_state {
54 enum cpuhp_state state;
55 enum cpuhp_state target;
56 enum cpuhp_state fail;
58 struct task_struct *thread;
64 struct hlist_node *node;
65 struct hlist_node *last;
66 enum cpuhp_state cb_state;
68 struct completion done_up;
69 struct completion done_down;
73 static DEFINE_PER_CPU(struct cpuhp_cpu_state, cpuhp_state) = {
74 .fail = CPUHP_INVALID,
77 #if defined(CONFIG_LOCKDEP) && defined(CONFIG_SMP)
78 static struct lockdep_map cpuhp_state_up_map =
79 STATIC_LOCKDEP_MAP_INIT("cpuhp_state-up", &cpuhp_state_up_map);
80 static struct lockdep_map cpuhp_state_down_map =
81 STATIC_LOCKDEP_MAP_INIT("cpuhp_state-down", &cpuhp_state_down_map);
84 static inline void cpuhp_lock_acquire(bool bringup)
86 lock_map_acquire(bringup ? &cpuhp_state_up_map : &cpuhp_state_down_map);
89 static inline void cpuhp_lock_release(bool bringup)
91 lock_map_release(bringup ? &cpuhp_state_up_map : &cpuhp_state_down_map);
95 static inline void cpuhp_lock_acquire(bool bringup) { }
96 static inline void cpuhp_lock_release(bool bringup) { }
101 * cpuhp_step - Hotplug state machine step
102 * @name: Name of the step
103 * @startup: Startup function of the step
104 * @teardown: Teardown function of the step
105 * @skip_onerr: Do not invoke the functions on error rollback
106 * Will go away once the notifiers are gone
107 * @cant_stop: Bringup/teardown can't be stopped at this step
112 int (*single)(unsigned int cpu);
113 int (*multi)(unsigned int cpu,
114 struct hlist_node *node);
117 int (*single)(unsigned int cpu);
118 int (*multi)(unsigned int cpu,
119 struct hlist_node *node);
121 struct hlist_head list;
127 static DEFINE_MUTEX(cpuhp_state_mutex);
128 static struct cpuhp_step cpuhp_hp_states[];
130 static struct cpuhp_step *cpuhp_get_step(enum cpuhp_state state)
132 return cpuhp_hp_states + state;
136 * cpuhp_invoke_callback _ Invoke the callbacks for a given state
137 * @cpu: The cpu for which the callback should be invoked
138 * @state: The state to do callbacks for
139 * @bringup: True if the bringup callback should be invoked
140 * @node: For multi-instance, do a single entry callback for install/remove
141 * @lastp: For multi-instance rollback, remember how far we got
143 * Called from cpu hotplug and from the state register machinery.
145 static int cpuhp_invoke_callback(unsigned int cpu, enum cpuhp_state state,
146 bool bringup, struct hlist_node *node,
147 struct hlist_node **lastp)
149 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
150 struct cpuhp_step *step = cpuhp_get_step(state);
151 int (*cbm)(unsigned int cpu, struct hlist_node *node);
152 int (*cb)(unsigned int cpu);
155 if (st->fail == state) {
156 st->fail = CPUHP_INVALID;
158 if (!(bringup ? step->startup.single : step->teardown.single))
164 if (!step->multi_instance) {
165 WARN_ON_ONCE(lastp && *lastp);
166 cb = bringup ? step->startup.single : step->teardown.single;
169 trace_cpuhp_enter(cpu, st->target, state, cb);
171 trace_cpuhp_exit(cpu, st->state, state, ret);
174 cbm = bringup ? step->startup.multi : step->teardown.multi;
178 /* Single invocation for instance add/remove */
180 WARN_ON_ONCE(lastp && *lastp);
181 trace_cpuhp_multi_enter(cpu, st->target, state, cbm, node);
182 ret = cbm(cpu, node);
183 trace_cpuhp_exit(cpu, st->state, state, ret);
187 /* State transition. Invoke on all instances */
189 hlist_for_each(node, &step->list) {
190 if (lastp && node == *lastp)
193 trace_cpuhp_multi_enter(cpu, st->target, state, cbm, node);
194 ret = cbm(cpu, node);
195 trace_cpuhp_exit(cpu, st->state, state, ret);
209 /* Rollback the instances if one failed */
210 cbm = !bringup ? step->startup.multi : step->teardown.multi;
214 hlist_for_each(node, &step->list) {
218 trace_cpuhp_multi_enter(cpu, st->target, state, cbm, node);
219 ret = cbm(cpu, node);
220 trace_cpuhp_exit(cpu, st->state, state, ret);
222 * Rollback must not fail,
230 static bool cpuhp_is_ap_state(enum cpuhp_state state)
233 * The extra check for CPUHP_TEARDOWN_CPU is only for documentation
234 * purposes as that state is handled explicitly in cpu_down.
236 return state > CPUHP_BRINGUP_CPU && state != CPUHP_TEARDOWN_CPU;
239 static inline void wait_for_ap_thread(struct cpuhp_cpu_state *st, bool bringup)
241 struct completion *done = bringup ? &st->done_up : &st->done_down;
242 wait_for_completion(done);
245 static inline void complete_ap_thread(struct cpuhp_cpu_state *st, bool bringup)
247 struct completion *done = bringup ? &st->done_up : &st->done_down;
252 * The former STARTING/DYING states, ran with IRQs disabled and must not fail.
254 static bool cpuhp_is_atomic_state(enum cpuhp_state state)
256 return CPUHP_AP_IDLE_DEAD <= state && state < CPUHP_AP_ONLINE;
259 /* Serializes the updates to cpu_online_mask, cpu_present_mask */
260 static DEFINE_MUTEX(cpu_add_remove_lock);
261 bool cpuhp_tasks_frozen;
262 EXPORT_SYMBOL_GPL(cpuhp_tasks_frozen);
265 * The following two APIs (cpu_maps_update_begin/done) must be used when
266 * attempting to serialize the updates to cpu_online_mask & cpu_present_mask.
268 void cpu_maps_update_begin(void)
270 mutex_lock(&cpu_add_remove_lock);
273 void cpu_maps_update_done(void)
275 mutex_unlock(&cpu_add_remove_lock);
279 * If set, cpu_up and cpu_down will return -EBUSY and do nothing.
280 * Should always be manipulated under cpu_add_remove_lock
282 static int cpu_hotplug_disabled;
284 #ifdef CONFIG_HOTPLUG_CPU
286 DEFINE_STATIC_PERCPU_RWSEM(cpu_hotplug_lock);
288 void cpus_read_lock(void)
290 percpu_down_read(&cpu_hotplug_lock);
292 EXPORT_SYMBOL_GPL(cpus_read_lock);
294 void cpus_read_unlock(void)
296 percpu_up_read(&cpu_hotplug_lock);
298 EXPORT_SYMBOL_GPL(cpus_read_unlock);
300 void cpus_write_lock(void)
302 percpu_down_write(&cpu_hotplug_lock);
305 void cpus_write_unlock(void)
307 percpu_up_write(&cpu_hotplug_lock);
310 void lockdep_assert_cpus_held(void)
312 percpu_rwsem_assert_held(&cpu_hotplug_lock);
316 * Wait for currently running CPU hotplug operations to complete (if any) and
317 * disable future CPU hotplug (from sysfs). The 'cpu_add_remove_lock' protects
318 * the 'cpu_hotplug_disabled' flag. The same lock is also acquired by the
319 * hotplug path before performing hotplug operations. So acquiring that lock
320 * guarantees mutual exclusion from any currently running hotplug operations.
322 void cpu_hotplug_disable(void)
324 cpu_maps_update_begin();
325 cpu_hotplug_disabled++;
326 cpu_maps_update_done();
328 EXPORT_SYMBOL_GPL(cpu_hotplug_disable);
330 static void __cpu_hotplug_enable(void)
332 if (WARN_ONCE(!cpu_hotplug_disabled, "Unbalanced cpu hotplug enable\n"))
334 cpu_hotplug_disabled--;
337 void cpu_hotplug_enable(void)
339 cpu_maps_update_begin();
340 __cpu_hotplug_enable();
341 cpu_maps_update_done();
343 EXPORT_SYMBOL_GPL(cpu_hotplug_enable);
344 #endif /* CONFIG_HOTPLUG_CPU */
346 #ifdef CONFIG_HOTPLUG_SMT
347 enum cpuhp_smt_control cpu_smt_control __read_mostly = CPU_SMT_ENABLED;
348 EXPORT_SYMBOL_GPL(cpu_smt_control);
350 static bool cpu_smt_available __read_mostly;
352 void __init cpu_smt_disable(bool force)
354 if (cpu_smt_control == CPU_SMT_FORCE_DISABLED ||
355 cpu_smt_control == CPU_SMT_NOT_SUPPORTED)
359 pr_info("SMT: Force disabled\n");
360 cpu_smt_control = CPU_SMT_FORCE_DISABLED;
362 cpu_smt_control = CPU_SMT_DISABLED;
367 * The decision whether SMT is supported can only be done after the full
368 * CPU identification. Called from architecture code before non boot CPUs
371 void __init cpu_smt_check_topology_early(void)
373 if (!topology_smt_supported())
374 cpu_smt_control = CPU_SMT_NOT_SUPPORTED;
378 * If SMT was disabled by BIOS, detect it here, after the CPUs have been
379 * brought online. This ensures the smt/l1tf sysfs entries are consistent
380 * with reality. cpu_smt_available is set to true during the bringup of non
381 * boot CPUs when a SMT sibling is detected. Note, this may overwrite
382 * cpu_smt_control's previous setting.
384 void __init cpu_smt_check_topology(void)
386 if (!cpu_smt_available)
387 cpu_smt_control = CPU_SMT_NOT_SUPPORTED;
390 static int __init smt_cmdline_disable(char *str)
392 cpu_smt_disable(str && !strcmp(str, "force"));
395 early_param("nosmt", smt_cmdline_disable);
397 static inline bool cpu_smt_allowed(unsigned int cpu)
399 if (topology_is_primary_thread(cpu))
403 * If the CPU is not a 'primary' thread and the booted_once bit is
404 * set then the processor has SMT support. Store this information
405 * for the late check of SMT support in cpu_smt_check_topology().
407 if (per_cpu(cpuhp_state, cpu).booted_once)
408 cpu_smt_available = true;
410 if (cpu_smt_control == CPU_SMT_ENABLED)
414 * On x86 it's required to boot all logical CPUs at least once so
415 * that the init code can get a chance to set CR4.MCE on each
416 * CPU. Otherwise, a broadacasted MCE observing CR4.MCE=0b on any
417 * core will shutdown the machine.
419 return !per_cpu(cpuhp_state, cpu).booted_once;
422 static inline bool cpu_smt_allowed(unsigned int cpu) { return true; }
425 static inline enum cpuhp_state
426 cpuhp_set_state(struct cpuhp_cpu_state *st, enum cpuhp_state target)
428 enum cpuhp_state prev_state = st->state;
430 st->rollback = false;
435 st->bringup = st->state < target;
441 cpuhp_reset_state(struct cpuhp_cpu_state *st, enum cpuhp_state prev_state)
446 * If we have st->last we need to undo partial multi_instance of this
447 * state first. Otherwise start undo at the previous state.
456 st->target = prev_state;
457 st->bringup = !st->bringup;
460 /* Regular hotplug invocation of the AP hotplug thread */
461 static void __cpuhp_kick_ap(struct cpuhp_cpu_state *st)
463 if (!st->single && st->state == st->target)
468 * Make sure the above stores are visible before should_run becomes
469 * true. Paired with the mb() above in cpuhp_thread_fun()
472 st->should_run = true;
473 wake_up_process(st->thread);
474 wait_for_ap_thread(st, st->bringup);
477 static int cpuhp_kick_ap(struct cpuhp_cpu_state *st, enum cpuhp_state target)
479 enum cpuhp_state prev_state;
482 prev_state = cpuhp_set_state(st, target);
484 if ((ret = st->result)) {
485 cpuhp_reset_state(st, prev_state);
492 static int bringup_wait_for_ap(unsigned int cpu)
494 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
496 /* Wait for the CPU to reach CPUHP_AP_ONLINE_IDLE */
497 wait_for_ap_thread(st, true);
498 if (WARN_ON_ONCE((!cpu_online(cpu))))
501 /* Unpark the stopper thread and the hotplug thread of the target cpu */
502 stop_machine_unpark(cpu);
503 kthread_unpark(st->thread);
506 * SMT soft disabling on X86 requires to bring the CPU out of the
507 * BIOS 'wait for SIPI' state in order to set the CR4.MCE bit. The
508 * CPU marked itself as booted_once in cpu_notify_starting() so the
509 * cpu_smt_allowed() check will now return false if this is not the
512 if (!cpu_smt_allowed(cpu))
515 if (st->target <= CPUHP_AP_ONLINE_IDLE)
518 return cpuhp_kick_ap(st, st->target);
521 static int bringup_cpu(unsigned int cpu)
523 struct task_struct *idle = idle_thread_get(cpu);
527 * Some architectures have to walk the irq descriptors to
528 * setup the vector space for the cpu which comes online.
529 * Prevent irq alloc/free across the bringup.
533 /* Arch-specific enabling code. */
534 ret = __cpu_up(cpu, idle);
538 return bringup_wait_for_ap(cpu);
542 * Hotplug state machine related functions
545 static void undo_cpu_up(unsigned int cpu, struct cpuhp_cpu_state *st)
547 for (st->state--; st->state > st->target; st->state--) {
548 struct cpuhp_step *step = cpuhp_get_step(st->state);
550 if (!step->skip_onerr)
551 cpuhp_invoke_callback(cpu, st->state, false, NULL, NULL);
555 static int cpuhp_up_callbacks(unsigned int cpu, struct cpuhp_cpu_state *st,
556 enum cpuhp_state target)
558 enum cpuhp_state prev_state = st->state;
561 while (st->state < target) {
563 ret = cpuhp_invoke_callback(cpu, st->state, true, NULL, NULL);
565 st->target = prev_state;
566 undo_cpu_up(cpu, st);
574 * The cpu hotplug threads manage the bringup and teardown of the cpus
576 static void cpuhp_create(unsigned int cpu)
578 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
580 init_completion(&st->done_up);
581 init_completion(&st->done_down);
584 static int cpuhp_should_run(unsigned int cpu)
586 struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
588 return st->should_run;
592 * Execute teardown/startup callbacks on the plugged cpu. Also used to invoke
593 * callbacks when a state gets [un]installed at runtime.
595 * Each invocation of this function by the smpboot thread does a single AP
598 * It has 3 modes of operation:
599 * - single: runs st->cb_state
600 * - up: runs ++st->state, while st->state < st->target
601 * - down: runs st->state--, while st->state > st->target
603 * When complete or on error, should_run is cleared and the completion is fired.
605 static void cpuhp_thread_fun(unsigned int cpu)
607 struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
608 bool bringup = st->bringup;
609 enum cpuhp_state state;
612 * ACQUIRE for the cpuhp_should_run() load of ->should_run. Ensures
613 * that if we see ->should_run we also see the rest of the state.
617 if (WARN_ON_ONCE(!st->should_run))
620 cpuhp_lock_acquire(bringup);
623 state = st->cb_state;
624 st->should_run = false;
629 st->should_run = (st->state < st->target);
630 WARN_ON_ONCE(st->state > st->target);
634 st->should_run = (st->state > st->target);
635 WARN_ON_ONCE(st->state < st->target);
639 WARN_ON_ONCE(!cpuhp_is_ap_state(state));
642 struct cpuhp_step *step = cpuhp_get_step(state);
643 if (step->skip_onerr)
647 if (cpuhp_is_atomic_state(state)) {
649 st->result = cpuhp_invoke_callback(cpu, state, bringup, st->node, &st->last);
653 * STARTING/DYING must not fail!
655 WARN_ON_ONCE(st->result);
657 st->result = cpuhp_invoke_callback(cpu, state, bringup, st->node, &st->last);
662 * If we fail on a rollback, we're up a creek without no
663 * paddle, no way forward, no way back. We loose, thanks for
666 WARN_ON_ONCE(st->rollback);
667 st->should_run = false;
671 cpuhp_lock_release(bringup);
674 complete_ap_thread(st, bringup);
677 /* Invoke a single callback on a remote cpu */
679 cpuhp_invoke_ap_callback(int cpu, enum cpuhp_state state, bool bringup,
680 struct hlist_node *node)
682 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
685 if (!cpu_online(cpu))
688 cpuhp_lock_acquire(false);
689 cpuhp_lock_release(false);
691 cpuhp_lock_acquire(true);
692 cpuhp_lock_release(true);
695 * If we are up and running, use the hotplug thread. For early calls
696 * we invoke the thread function directly.
699 return cpuhp_invoke_callback(cpu, state, bringup, node, NULL);
701 st->rollback = false;
705 st->bringup = bringup;
706 st->cb_state = state;
712 * If we failed and did a partial, do a rollback.
714 if ((ret = st->result) && st->last) {
716 st->bringup = !bringup;
722 * Clean up the leftovers so the next hotplug operation wont use stale
725 st->node = st->last = NULL;
729 static int cpuhp_kick_ap_work(unsigned int cpu)
731 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
732 enum cpuhp_state prev_state = st->state;
735 cpuhp_lock_acquire(false);
736 cpuhp_lock_release(false);
738 cpuhp_lock_acquire(true);
739 cpuhp_lock_release(true);
741 trace_cpuhp_enter(cpu, st->target, prev_state, cpuhp_kick_ap_work);
742 ret = cpuhp_kick_ap(st, st->target);
743 trace_cpuhp_exit(cpu, st->state, prev_state, ret);
748 static struct smp_hotplug_thread cpuhp_threads = {
749 .store = &cpuhp_state.thread,
750 .create = &cpuhp_create,
751 .thread_should_run = cpuhp_should_run,
752 .thread_fn = cpuhp_thread_fun,
753 .thread_comm = "cpuhp/%u",
757 void __init cpuhp_threads_init(void)
759 BUG_ON(smpboot_register_percpu_thread(&cpuhp_threads));
760 kthread_unpark(this_cpu_read(cpuhp_state.thread));
763 #ifdef CONFIG_HOTPLUG_CPU
765 * clear_tasks_mm_cpumask - Safely clear tasks' mm_cpumask for a CPU
768 * This function walks all processes, finds a valid mm struct for each one and
769 * then clears a corresponding bit in mm's cpumask. While this all sounds
770 * trivial, there are various non-obvious corner cases, which this function
771 * tries to solve in a safe manner.
773 * Also note that the function uses a somewhat relaxed locking scheme, so it may
774 * be called only for an already offlined CPU.
776 void clear_tasks_mm_cpumask(int cpu)
778 struct task_struct *p;
781 * This function is called after the cpu is taken down and marked
782 * offline, so its not like new tasks will ever get this cpu set in
783 * their mm mask. -- Peter Zijlstra
784 * Thus, we may use rcu_read_lock() here, instead of grabbing
785 * full-fledged tasklist_lock.
787 WARN_ON(cpu_online(cpu));
789 for_each_process(p) {
790 struct task_struct *t;
793 * Main thread might exit, but other threads may still have
794 * a valid mm. Find one.
796 t = find_lock_task_mm(p);
799 cpumask_clear_cpu(cpu, mm_cpumask(t->mm));
805 /* Take this CPU down. */
806 static int take_cpu_down(void *_param)
808 struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
809 enum cpuhp_state target = max((int)st->target, CPUHP_AP_OFFLINE);
810 int err, cpu = smp_processor_id();
813 /* Ensure this CPU doesn't handle any more interrupts. */
814 err = __cpu_disable();
819 * We get here while we are in CPUHP_TEARDOWN_CPU state and we must not
820 * do this step again.
822 WARN_ON(st->state != CPUHP_TEARDOWN_CPU);
824 /* Invoke the former CPU_DYING callbacks */
825 for (; st->state > target; st->state--) {
826 ret = cpuhp_invoke_callback(cpu, st->state, false, NULL, NULL);
828 * DYING must not fail!
833 /* Give up timekeeping duties */
834 tick_handover_do_timer();
835 /* Park the stopper thread */
836 stop_machine_park(cpu);
840 static int takedown_cpu(unsigned int cpu)
842 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
845 /* Park the smpboot threads */
846 kthread_park(per_cpu_ptr(&cpuhp_state, cpu)->thread);
849 * Prevent irq alloc/free while the dying cpu reorganizes the
850 * interrupt affinities.
855 * So now all preempt/rcu users must observe !cpu_active().
857 err = stop_machine_cpuslocked(take_cpu_down, NULL, cpumask_of(cpu));
859 /* CPU refused to die */
861 /* Unpark the hotplug thread so we can rollback there */
862 kthread_unpark(per_cpu_ptr(&cpuhp_state, cpu)->thread);
865 BUG_ON(cpu_online(cpu));
868 * The teardown callback for CPUHP_AP_SCHED_STARTING will have removed
869 * all runnable tasks from the CPU, there's only the idle task left now
870 * that the migration thread is done doing the stop_machine thing.
872 * Wait for the stop thread to go away.
874 wait_for_ap_thread(st, false);
875 BUG_ON(st->state != CPUHP_AP_IDLE_DEAD);
877 /* Interrupts are moved away from the dying cpu, reenable alloc/free */
880 hotplug_cpu__broadcast_tick_pull(cpu);
881 /* This actually kills the CPU. */
884 tick_cleanup_dead_cpu(cpu);
885 rcutree_migrate_callbacks(cpu);
889 static void cpuhp_complete_idle_dead(void *arg)
891 struct cpuhp_cpu_state *st = arg;
893 complete_ap_thread(st, false);
896 void cpuhp_report_idle_dead(void)
898 struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
900 BUG_ON(st->state != CPUHP_AP_OFFLINE);
901 rcu_report_dead(smp_processor_id());
902 st->state = CPUHP_AP_IDLE_DEAD;
904 * We cannot call complete after rcu_report_dead() so we delegate it
907 smp_call_function_single(cpumask_first(cpu_online_mask),
908 cpuhp_complete_idle_dead, st, 0);
911 static void undo_cpu_down(unsigned int cpu, struct cpuhp_cpu_state *st)
913 for (st->state++; st->state < st->target; st->state++) {
914 struct cpuhp_step *step = cpuhp_get_step(st->state);
916 if (!step->skip_onerr)
917 cpuhp_invoke_callback(cpu, st->state, true, NULL, NULL);
921 static int cpuhp_down_callbacks(unsigned int cpu, struct cpuhp_cpu_state *st,
922 enum cpuhp_state target)
924 enum cpuhp_state prev_state = st->state;
927 for (; st->state > target; st->state--) {
928 ret = cpuhp_invoke_callback(cpu, st->state, false, NULL, NULL);
930 st->target = prev_state;
931 undo_cpu_down(cpu, st);
938 /* Requires cpu_add_remove_lock to be held */
939 static int __ref _cpu_down(unsigned int cpu, int tasks_frozen,
940 enum cpuhp_state target)
942 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
943 int prev_state, ret = 0;
945 if (num_online_cpus() == 1)
948 if (!cpu_present(cpu))
953 cpuhp_tasks_frozen = tasks_frozen;
955 prev_state = cpuhp_set_state(st, target);
957 * If the current CPU state is in the range of the AP hotplug thread,
958 * then we need to kick the thread.
960 if (st->state > CPUHP_TEARDOWN_CPU) {
961 st->target = max((int)target, CPUHP_TEARDOWN_CPU);
962 ret = cpuhp_kick_ap_work(cpu);
964 * The AP side has done the error rollback already. Just
965 * return the error code..
971 * We might have stopped still in the range of the AP hotplug
972 * thread. Nothing to do anymore.
974 if (st->state > CPUHP_TEARDOWN_CPU)
980 * The AP brought itself down to CPUHP_TEARDOWN_CPU. So we need
981 * to do the further cleanups.
983 ret = cpuhp_down_callbacks(cpu, st, target);
984 if (ret && st->state > CPUHP_TEARDOWN_CPU && st->state < prev_state) {
985 cpuhp_reset_state(st, prev_state);
992 * Do post unplug cleanup. This is still protected against
993 * concurrent CPU hotplug via cpu_add_remove_lock.
995 lockup_detector_cleanup();
999 static int cpu_down_maps_locked(unsigned int cpu, enum cpuhp_state target)
1001 if (cpu_hotplug_disabled)
1003 return _cpu_down(cpu, 0, target);
1006 static int do_cpu_down(unsigned int cpu, enum cpuhp_state target)
1010 cpu_maps_update_begin();
1011 err = cpu_down_maps_locked(cpu, target);
1012 cpu_maps_update_done();
1016 int cpu_down(unsigned int cpu)
1018 return do_cpu_down(cpu, CPUHP_OFFLINE);
1020 EXPORT_SYMBOL(cpu_down);
1023 #define takedown_cpu NULL
1024 #endif /*CONFIG_HOTPLUG_CPU*/
1027 * notify_cpu_starting(cpu) - Invoke the callbacks on the starting CPU
1028 * @cpu: cpu that just started
1030 * It must be called by the arch code on the new cpu, before the new cpu
1031 * enables interrupts and before the "boot" cpu returns from __cpu_up().
1033 void notify_cpu_starting(unsigned int cpu)
1035 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1036 enum cpuhp_state target = min((int)st->target, CPUHP_AP_ONLINE);
1039 rcu_cpu_starting(cpu); /* Enables RCU usage on this CPU. */
1040 st->booted_once = true;
1041 while (st->state < target) {
1043 ret = cpuhp_invoke_callback(cpu, st->state, true, NULL, NULL);
1045 * STARTING must not fail!
1052 * Called from the idle task. Wake up the controlling task which brings the
1053 * stopper and the hotplug thread of the upcoming CPU up and then delegates
1054 * the rest of the online bringup to the hotplug thread.
1056 void cpuhp_online_idle(enum cpuhp_state state)
1058 struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
1060 /* Happens for the boot cpu */
1061 if (state != CPUHP_AP_ONLINE_IDLE)
1064 st->state = CPUHP_AP_ONLINE_IDLE;
1065 complete_ap_thread(st, true);
1068 /* Requires cpu_add_remove_lock to be held */
1069 static int _cpu_up(unsigned int cpu, int tasks_frozen, enum cpuhp_state target)
1071 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1072 struct task_struct *idle;
1077 if (!cpu_present(cpu)) {
1083 * The caller of do_cpu_up might have raced with another
1084 * caller. Ignore it for now.
1086 if (st->state >= target)
1089 if (st->state == CPUHP_OFFLINE) {
1090 /* Let it fail before we try to bring the cpu up */
1091 idle = idle_thread_get(cpu);
1093 ret = PTR_ERR(idle);
1098 cpuhp_tasks_frozen = tasks_frozen;
1100 cpuhp_set_state(st, target);
1102 * If the current CPU state is in the range of the AP hotplug thread,
1103 * then we need to kick the thread once more.
1105 if (st->state > CPUHP_BRINGUP_CPU) {
1106 ret = cpuhp_kick_ap_work(cpu);
1108 * The AP side has done the error rollback already. Just
1109 * return the error code..
1116 * Try to reach the target state. We max out on the BP at
1117 * CPUHP_BRINGUP_CPU. After that the AP hotplug thread is
1118 * responsible for bringing it up to the target state.
1120 target = min((int)target, CPUHP_BRINGUP_CPU);
1121 ret = cpuhp_up_callbacks(cpu, st, target);
1123 cpus_write_unlock();
1127 static int do_cpu_up(unsigned int cpu, enum cpuhp_state target)
1131 if (!cpu_possible(cpu)) {
1132 pr_err("can't online cpu %d because it is not configured as may-hotadd at boot time\n",
1134 #if defined(CONFIG_IA64)
1135 pr_err("please check additional_cpus= boot parameter\n");
1140 err = try_online_node(cpu_to_node(cpu));
1144 cpu_maps_update_begin();
1146 if (cpu_hotplug_disabled) {
1150 if (!cpu_smt_allowed(cpu)) {
1155 err = _cpu_up(cpu, 0, target);
1157 cpu_maps_update_done();
1161 int cpu_up(unsigned int cpu)
1163 return do_cpu_up(cpu, CPUHP_ONLINE);
1165 EXPORT_SYMBOL_GPL(cpu_up);
1167 #ifdef CONFIG_PM_SLEEP_SMP
1168 static cpumask_var_t frozen_cpus;
1170 int freeze_secondary_cpus(int primary)
1174 cpu_maps_update_begin();
1175 if (!cpu_online(primary))
1176 primary = cpumask_first(cpu_online_mask);
1178 * We take down all of the non-boot CPUs in one shot to avoid races
1179 * with the userspace trying to use the CPU hotplug at the same time
1181 cpumask_clear(frozen_cpus);
1183 pr_info("Disabling non-boot CPUs ...\n");
1184 for_each_online_cpu(cpu) {
1187 trace_suspend_resume(TPS("CPU_OFF"), cpu, true);
1188 error = _cpu_down(cpu, 1, CPUHP_OFFLINE);
1189 trace_suspend_resume(TPS("CPU_OFF"), cpu, false);
1191 cpumask_set_cpu(cpu, frozen_cpus);
1193 pr_err("Error taking CPU%d down: %d\n", cpu, error);
1199 BUG_ON(num_online_cpus() > 1);
1201 pr_err("Non-boot CPUs are not disabled\n");
1204 * Make sure the CPUs won't be enabled by someone else. We need to do
1205 * this even in case of failure as all disable_nonboot_cpus() users are
1206 * supposed to do enable_nonboot_cpus() on the failure path.
1208 cpu_hotplug_disabled++;
1210 cpu_maps_update_done();
1214 void __weak arch_enable_nonboot_cpus_begin(void)
1218 void __weak arch_enable_nonboot_cpus_end(void)
1222 void enable_nonboot_cpus(void)
1226 /* Allow everyone to use the CPU hotplug again */
1227 cpu_maps_update_begin();
1228 __cpu_hotplug_enable();
1229 if (cpumask_empty(frozen_cpus))
1232 pr_info("Enabling non-boot CPUs ...\n");
1234 arch_enable_nonboot_cpus_begin();
1236 for_each_cpu(cpu, frozen_cpus) {
1237 trace_suspend_resume(TPS("CPU_ON"), cpu, true);
1238 error = _cpu_up(cpu, 1, CPUHP_ONLINE);
1239 trace_suspend_resume(TPS("CPU_ON"), cpu, false);
1241 pr_info("CPU%d is up\n", cpu);
1244 pr_warn("Error taking CPU%d up: %d\n", cpu, error);
1247 arch_enable_nonboot_cpus_end();
1249 cpumask_clear(frozen_cpus);
1251 cpu_maps_update_done();
1254 static int __init alloc_frozen_cpus(void)
1256 if (!alloc_cpumask_var(&frozen_cpus, GFP_KERNEL|__GFP_ZERO))
1260 core_initcall(alloc_frozen_cpus);
1263 * When callbacks for CPU hotplug notifications are being executed, we must
1264 * ensure that the state of the system with respect to the tasks being frozen
1265 * or not, as reported by the notification, remains unchanged *throughout the
1266 * duration* of the execution of the callbacks.
1267 * Hence we need to prevent the freezer from racing with regular CPU hotplug.
1269 * This synchronization is implemented by mutually excluding regular CPU
1270 * hotplug and Suspend/Hibernate call paths by hooking onto the Suspend/
1271 * Hibernate notifications.
1274 cpu_hotplug_pm_callback(struct notifier_block *nb,
1275 unsigned long action, void *ptr)
1279 case PM_SUSPEND_PREPARE:
1280 case PM_HIBERNATION_PREPARE:
1281 cpu_hotplug_disable();
1284 case PM_POST_SUSPEND:
1285 case PM_POST_HIBERNATION:
1286 cpu_hotplug_enable();
1297 static int __init cpu_hotplug_pm_sync_init(void)
1300 * cpu_hotplug_pm_callback has higher priority than x86
1301 * bsp_pm_callback which depends on cpu_hotplug_pm_callback
1302 * to disable cpu hotplug to avoid cpu hotplug race.
1304 pm_notifier(cpu_hotplug_pm_callback, 0);
1307 core_initcall(cpu_hotplug_pm_sync_init);
1309 #endif /* CONFIG_PM_SLEEP_SMP */
1313 #endif /* CONFIG_SMP */
1315 /* Boot processor state steps */
1316 static struct cpuhp_step cpuhp_hp_states[] = {
1319 .startup.single = NULL,
1320 .teardown.single = NULL,
1323 [CPUHP_CREATE_THREADS]= {
1324 .name = "threads:prepare",
1325 .startup.single = smpboot_create_threads,
1326 .teardown.single = NULL,
1329 [CPUHP_PERF_PREPARE] = {
1330 .name = "perf:prepare",
1331 .startup.single = perf_event_init_cpu,
1332 .teardown.single = perf_event_exit_cpu,
1334 [CPUHP_WORKQUEUE_PREP] = {
1335 .name = "workqueue:prepare",
1336 .startup.single = workqueue_prepare_cpu,
1337 .teardown.single = NULL,
1339 [CPUHP_HRTIMERS_PREPARE] = {
1340 .name = "hrtimers:prepare",
1341 .startup.single = hrtimers_prepare_cpu,
1342 .teardown.single = hrtimers_dead_cpu,
1344 [CPUHP_SMPCFD_PREPARE] = {
1345 .name = "smpcfd:prepare",
1346 .startup.single = smpcfd_prepare_cpu,
1347 .teardown.single = smpcfd_dead_cpu,
1349 [CPUHP_RELAY_PREPARE] = {
1350 .name = "relay:prepare",
1351 .startup.single = relay_prepare_cpu,
1352 .teardown.single = NULL,
1354 [CPUHP_SLAB_PREPARE] = {
1355 .name = "slab:prepare",
1356 .startup.single = slab_prepare_cpu,
1357 .teardown.single = slab_dead_cpu,
1359 [CPUHP_RCUTREE_PREP] = {
1360 .name = "RCU/tree:prepare",
1361 .startup.single = rcutree_prepare_cpu,
1362 .teardown.single = rcutree_dead_cpu,
1365 * On the tear-down path, timers_dead_cpu() must be invoked
1366 * before blk_mq_queue_reinit_notify() from notify_dead(),
1367 * otherwise a RCU stall occurs.
1369 [CPUHP_TIMERS_PREPARE] = {
1370 .name = "timers:prepare",
1371 .startup.single = timers_prepare_cpu,
1372 .teardown.single = timers_dead_cpu,
1374 /* Kicks the plugged cpu into life */
1375 [CPUHP_BRINGUP_CPU] = {
1376 .name = "cpu:bringup",
1377 .startup.single = bringup_cpu,
1378 .teardown.single = NULL,
1381 /* Final state before CPU kills itself */
1382 [CPUHP_AP_IDLE_DEAD] = {
1383 .name = "idle:dead",
1386 * Last state before CPU enters the idle loop to die. Transient state
1387 * for synchronization.
1389 [CPUHP_AP_OFFLINE] = {
1390 .name = "ap:offline",
1393 /* First state is scheduler control. Interrupts are disabled */
1394 [CPUHP_AP_SCHED_STARTING] = {
1395 .name = "sched:starting",
1396 .startup.single = sched_cpu_starting,
1397 .teardown.single = sched_cpu_dying,
1399 [CPUHP_AP_RCUTREE_DYING] = {
1400 .name = "RCU/tree:dying",
1401 .startup.single = NULL,
1402 .teardown.single = rcutree_dying_cpu,
1404 [CPUHP_AP_SMPCFD_DYING] = {
1405 .name = "smpcfd:dying",
1406 .startup.single = NULL,
1407 .teardown.single = smpcfd_dying_cpu,
1409 /* Entry state on starting. Interrupts enabled from here on. Transient
1410 * state for synchronsization */
1411 [CPUHP_AP_ONLINE] = {
1412 .name = "ap:online",
1415 * Handled on controll processor until the plugged processor manages
1418 [CPUHP_TEARDOWN_CPU] = {
1419 .name = "cpu:teardown",
1420 .startup.single = NULL,
1421 .teardown.single = takedown_cpu,
1424 /* Handle smpboot threads park/unpark */
1425 [CPUHP_AP_SMPBOOT_THREADS] = {
1426 .name = "smpboot/threads:online",
1427 .startup.single = smpboot_unpark_threads,
1428 .teardown.single = smpboot_park_threads,
1430 [CPUHP_AP_IRQ_AFFINITY_ONLINE] = {
1431 .name = "irq/affinity:online",
1432 .startup.single = irq_affinity_online_cpu,
1433 .teardown.single = NULL,
1435 [CPUHP_AP_PERF_ONLINE] = {
1436 .name = "perf:online",
1437 .startup.single = perf_event_init_cpu,
1438 .teardown.single = perf_event_exit_cpu,
1440 [CPUHP_AP_WATCHDOG_ONLINE] = {
1441 .name = "lockup_detector:online",
1442 .startup.single = lockup_detector_online_cpu,
1443 .teardown.single = lockup_detector_offline_cpu,
1445 [CPUHP_AP_WORKQUEUE_ONLINE] = {
1446 .name = "workqueue:online",
1447 .startup.single = workqueue_online_cpu,
1448 .teardown.single = workqueue_offline_cpu,
1450 [CPUHP_AP_RCUTREE_ONLINE] = {
1451 .name = "RCU/tree:online",
1452 .startup.single = rcutree_online_cpu,
1453 .teardown.single = rcutree_offline_cpu,
1457 * The dynamically registered state space is here
1461 /* Last state is scheduler control setting the cpu active */
1462 [CPUHP_AP_ACTIVE] = {
1463 .name = "sched:active",
1464 .startup.single = sched_cpu_activate,
1465 .teardown.single = sched_cpu_deactivate,
1469 /* CPU is fully up and running. */
1472 .startup.single = NULL,
1473 .teardown.single = NULL,
1477 /* Sanity check for callbacks */
1478 static int cpuhp_cb_check(enum cpuhp_state state)
1480 if (state <= CPUHP_OFFLINE || state >= CPUHP_ONLINE)
1486 * Returns a free for dynamic slot assignment of the Online state. The states
1487 * are protected by the cpuhp_slot_states mutex and an empty slot is identified
1488 * by having no name assigned.
1490 static int cpuhp_reserve_state(enum cpuhp_state state)
1492 enum cpuhp_state i, end;
1493 struct cpuhp_step *step;
1496 case CPUHP_AP_ONLINE_DYN:
1497 step = cpuhp_hp_states + CPUHP_AP_ONLINE_DYN;
1498 end = CPUHP_AP_ONLINE_DYN_END;
1500 case CPUHP_BP_PREPARE_DYN:
1501 step = cpuhp_hp_states + CPUHP_BP_PREPARE_DYN;
1502 end = CPUHP_BP_PREPARE_DYN_END;
1508 for (i = state; i <= end; i++, step++) {
1512 WARN(1, "No more dynamic states available for CPU hotplug\n");
1516 static int cpuhp_store_callbacks(enum cpuhp_state state, const char *name,
1517 int (*startup)(unsigned int cpu),
1518 int (*teardown)(unsigned int cpu),
1519 bool multi_instance)
1521 /* (Un)Install the callbacks for further cpu hotplug operations */
1522 struct cpuhp_step *sp;
1526 * If name is NULL, then the state gets removed.
1528 * CPUHP_AP_ONLINE_DYN and CPUHP_BP_PREPARE_DYN are handed out on
1529 * the first allocation from these dynamic ranges, so the removal
1530 * would trigger a new allocation and clear the wrong (already
1531 * empty) state, leaving the callbacks of the to be cleared state
1532 * dangling, which causes wreckage on the next hotplug operation.
1534 if (name && (state == CPUHP_AP_ONLINE_DYN ||
1535 state == CPUHP_BP_PREPARE_DYN)) {
1536 ret = cpuhp_reserve_state(state);
1541 sp = cpuhp_get_step(state);
1542 if (name && sp->name)
1545 sp->startup.single = startup;
1546 sp->teardown.single = teardown;
1548 sp->multi_instance = multi_instance;
1549 INIT_HLIST_HEAD(&sp->list);
1553 static void *cpuhp_get_teardown_cb(enum cpuhp_state state)
1555 return cpuhp_get_step(state)->teardown.single;
1559 * Call the startup/teardown function for a step either on the AP or
1560 * on the current CPU.
1562 static int cpuhp_issue_call(int cpu, enum cpuhp_state state, bool bringup,
1563 struct hlist_node *node)
1565 struct cpuhp_step *sp = cpuhp_get_step(state);
1569 * If there's nothing to do, we done.
1570 * Relies on the union for multi_instance.
1572 if ((bringup && !sp->startup.single) ||
1573 (!bringup && !sp->teardown.single))
1576 * The non AP bound callbacks can fail on bringup. On teardown
1577 * e.g. module removal we crash for now.
1580 if (cpuhp_is_ap_state(state))
1581 ret = cpuhp_invoke_ap_callback(cpu, state, bringup, node);
1583 ret = cpuhp_invoke_callback(cpu, state, bringup, node, NULL);
1585 ret = cpuhp_invoke_callback(cpu, state, bringup, node, NULL);
1587 BUG_ON(ret && !bringup);
1592 * Called from __cpuhp_setup_state on a recoverable failure.
1594 * Note: The teardown callbacks for rollback are not allowed to fail!
1596 static void cpuhp_rollback_install(int failedcpu, enum cpuhp_state state,
1597 struct hlist_node *node)
1601 /* Roll back the already executed steps on the other cpus */
1602 for_each_present_cpu(cpu) {
1603 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1604 int cpustate = st->state;
1606 if (cpu >= failedcpu)
1609 /* Did we invoke the startup call on that cpu ? */
1610 if (cpustate >= state)
1611 cpuhp_issue_call(cpu, state, false, node);
1615 int __cpuhp_state_add_instance_cpuslocked(enum cpuhp_state state,
1616 struct hlist_node *node,
1619 struct cpuhp_step *sp;
1623 lockdep_assert_cpus_held();
1625 sp = cpuhp_get_step(state);
1626 if (sp->multi_instance == false)
1629 mutex_lock(&cpuhp_state_mutex);
1631 if (!invoke || !sp->startup.multi)
1635 * Try to call the startup callback for each present cpu
1636 * depending on the hotplug state of the cpu.
1638 for_each_present_cpu(cpu) {
1639 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1640 int cpustate = st->state;
1642 if (cpustate < state)
1645 ret = cpuhp_issue_call(cpu, state, true, node);
1647 if (sp->teardown.multi)
1648 cpuhp_rollback_install(cpu, state, node);
1654 hlist_add_head(node, &sp->list);
1656 mutex_unlock(&cpuhp_state_mutex);
1660 int __cpuhp_state_add_instance(enum cpuhp_state state, struct hlist_node *node,
1666 ret = __cpuhp_state_add_instance_cpuslocked(state, node, invoke);
1670 EXPORT_SYMBOL_GPL(__cpuhp_state_add_instance);
1673 * __cpuhp_setup_state_cpuslocked - Setup the callbacks for an hotplug machine state
1674 * @state: The state to setup
1675 * @invoke: If true, the startup function is invoked for cpus where
1676 * cpu state >= @state
1677 * @startup: startup callback function
1678 * @teardown: teardown callback function
1679 * @multi_instance: State is set up for multiple instances which get
1682 * The caller needs to hold cpus read locked while calling this function.
1685 * Positive state number if @state is CPUHP_AP_ONLINE_DYN
1686 * 0 for all other states
1687 * On failure: proper (negative) error code
1689 int __cpuhp_setup_state_cpuslocked(enum cpuhp_state state,
1690 const char *name, bool invoke,
1691 int (*startup)(unsigned int cpu),
1692 int (*teardown)(unsigned int cpu),
1693 bool multi_instance)
1698 lockdep_assert_cpus_held();
1700 if (cpuhp_cb_check(state) || !name)
1703 mutex_lock(&cpuhp_state_mutex);
1705 ret = cpuhp_store_callbacks(state, name, startup, teardown,
1708 dynstate = state == CPUHP_AP_ONLINE_DYN;
1709 if (ret > 0 && dynstate) {
1714 if (ret || !invoke || !startup)
1718 * Try to call the startup callback for each present cpu
1719 * depending on the hotplug state of the cpu.
1721 for_each_present_cpu(cpu) {
1722 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1723 int cpustate = st->state;
1725 if (cpustate < state)
1728 ret = cpuhp_issue_call(cpu, state, true, NULL);
1731 cpuhp_rollback_install(cpu, state, NULL);
1732 cpuhp_store_callbacks(state, NULL, NULL, NULL, false);
1737 mutex_unlock(&cpuhp_state_mutex);
1739 * If the requested state is CPUHP_AP_ONLINE_DYN, return the
1740 * dynamically allocated state in case of success.
1742 if (!ret && dynstate)
1746 EXPORT_SYMBOL(__cpuhp_setup_state_cpuslocked);
1748 int __cpuhp_setup_state(enum cpuhp_state state,
1749 const char *name, bool invoke,
1750 int (*startup)(unsigned int cpu),
1751 int (*teardown)(unsigned int cpu),
1752 bool multi_instance)
1757 ret = __cpuhp_setup_state_cpuslocked(state, name, invoke, startup,
1758 teardown, multi_instance);
1762 EXPORT_SYMBOL(__cpuhp_setup_state);
1764 int __cpuhp_state_remove_instance(enum cpuhp_state state,
1765 struct hlist_node *node, bool invoke)
1767 struct cpuhp_step *sp = cpuhp_get_step(state);
1770 BUG_ON(cpuhp_cb_check(state));
1772 if (!sp->multi_instance)
1776 mutex_lock(&cpuhp_state_mutex);
1778 if (!invoke || !cpuhp_get_teardown_cb(state))
1781 * Call the teardown callback for each present cpu depending
1782 * on the hotplug state of the cpu. This function is not
1783 * allowed to fail currently!
1785 for_each_present_cpu(cpu) {
1786 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1787 int cpustate = st->state;
1789 if (cpustate >= state)
1790 cpuhp_issue_call(cpu, state, false, node);
1795 mutex_unlock(&cpuhp_state_mutex);
1800 EXPORT_SYMBOL_GPL(__cpuhp_state_remove_instance);
1803 * __cpuhp_remove_state_cpuslocked - Remove the callbacks for an hotplug machine state
1804 * @state: The state to remove
1805 * @invoke: If true, the teardown function is invoked for cpus where
1806 * cpu state >= @state
1808 * The caller needs to hold cpus read locked while calling this function.
1809 * The teardown callback is currently not allowed to fail. Think
1810 * about module removal!
1812 void __cpuhp_remove_state_cpuslocked(enum cpuhp_state state, bool invoke)
1814 struct cpuhp_step *sp = cpuhp_get_step(state);
1817 BUG_ON(cpuhp_cb_check(state));
1819 lockdep_assert_cpus_held();
1821 mutex_lock(&cpuhp_state_mutex);
1822 if (sp->multi_instance) {
1823 WARN(!hlist_empty(&sp->list),
1824 "Error: Removing state %d which has instances left.\n",
1829 if (!invoke || !cpuhp_get_teardown_cb(state))
1833 * Call the teardown callback for each present cpu depending
1834 * on the hotplug state of the cpu. This function is not
1835 * allowed to fail currently!
1837 for_each_present_cpu(cpu) {
1838 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1839 int cpustate = st->state;
1841 if (cpustate >= state)
1842 cpuhp_issue_call(cpu, state, false, NULL);
1845 cpuhp_store_callbacks(state, NULL, NULL, NULL, false);
1846 mutex_unlock(&cpuhp_state_mutex);
1848 EXPORT_SYMBOL(__cpuhp_remove_state_cpuslocked);
1850 void __cpuhp_remove_state(enum cpuhp_state state, bool invoke)
1853 __cpuhp_remove_state_cpuslocked(state, invoke);
1856 EXPORT_SYMBOL(__cpuhp_remove_state);
1858 #if defined(CONFIG_SYSFS) && defined(CONFIG_HOTPLUG_CPU)
1859 static ssize_t show_cpuhp_state(struct device *dev,
1860 struct device_attribute *attr, char *buf)
1862 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id);
1864 return sprintf(buf, "%d\n", st->state);
1866 static DEVICE_ATTR(state, 0444, show_cpuhp_state, NULL);
1868 static ssize_t write_cpuhp_target(struct device *dev,
1869 struct device_attribute *attr,
1870 const char *buf, size_t count)
1872 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id);
1873 struct cpuhp_step *sp;
1876 ret = kstrtoint(buf, 10, &target);
1880 #ifdef CONFIG_CPU_HOTPLUG_STATE_CONTROL
1881 if (target < CPUHP_OFFLINE || target > CPUHP_ONLINE)
1884 if (target != CPUHP_OFFLINE && target != CPUHP_ONLINE)
1888 ret = lock_device_hotplug_sysfs();
1892 mutex_lock(&cpuhp_state_mutex);
1893 sp = cpuhp_get_step(target);
1894 ret = !sp->name || sp->cant_stop ? -EINVAL : 0;
1895 mutex_unlock(&cpuhp_state_mutex);
1899 if (st->state < target)
1900 ret = do_cpu_up(dev->id, target);
1902 ret = do_cpu_down(dev->id, target);
1904 unlock_device_hotplug();
1905 return ret ? ret : count;
1908 static ssize_t show_cpuhp_target(struct device *dev,
1909 struct device_attribute *attr, char *buf)
1911 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id);
1913 return sprintf(buf, "%d\n", st->target);
1915 static DEVICE_ATTR(target, 0644, show_cpuhp_target, write_cpuhp_target);
1918 static ssize_t write_cpuhp_fail(struct device *dev,
1919 struct device_attribute *attr,
1920 const char *buf, size_t count)
1922 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id);
1923 struct cpuhp_step *sp;
1926 ret = kstrtoint(buf, 10, &fail);
1931 * Cannot fail STARTING/DYING callbacks.
1933 if (cpuhp_is_atomic_state(fail))
1937 * Cannot fail anything that doesn't have callbacks.
1939 mutex_lock(&cpuhp_state_mutex);
1940 sp = cpuhp_get_step(fail);
1941 if (!sp->startup.single && !sp->teardown.single)
1943 mutex_unlock(&cpuhp_state_mutex);
1952 static ssize_t show_cpuhp_fail(struct device *dev,
1953 struct device_attribute *attr, char *buf)
1955 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id);
1957 return sprintf(buf, "%d\n", st->fail);
1960 static DEVICE_ATTR(fail, 0644, show_cpuhp_fail, write_cpuhp_fail);
1962 static struct attribute *cpuhp_cpu_attrs[] = {
1963 &dev_attr_state.attr,
1964 &dev_attr_target.attr,
1965 &dev_attr_fail.attr,
1969 static const struct attribute_group cpuhp_cpu_attr_group = {
1970 .attrs = cpuhp_cpu_attrs,
1975 static ssize_t show_cpuhp_states(struct device *dev,
1976 struct device_attribute *attr, char *buf)
1978 ssize_t cur, res = 0;
1981 mutex_lock(&cpuhp_state_mutex);
1982 for (i = CPUHP_OFFLINE; i <= CPUHP_ONLINE; i++) {
1983 struct cpuhp_step *sp = cpuhp_get_step(i);
1986 cur = sprintf(buf, "%3d: %s\n", i, sp->name);
1991 mutex_unlock(&cpuhp_state_mutex);
1994 static DEVICE_ATTR(states, 0444, show_cpuhp_states, NULL);
1996 static struct attribute *cpuhp_cpu_root_attrs[] = {
1997 &dev_attr_states.attr,
2001 static const struct attribute_group cpuhp_cpu_root_attr_group = {
2002 .attrs = cpuhp_cpu_root_attrs,
2007 #ifdef CONFIG_HOTPLUG_SMT
2009 static const char *smt_states[] = {
2010 [CPU_SMT_ENABLED] = "on",
2011 [CPU_SMT_DISABLED] = "off",
2012 [CPU_SMT_FORCE_DISABLED] = "forceoff",
2013 [CPU_SMT_NOT_SUPPORTED] = "notsupported",
2017 show_smt_control(struct device *dev, struct device_attribute *attr, char *buf)
2019 return snprintf(buf, PAGE_SIZE - 2, "%s\n", smt_states[cpu_smt_control]);
2022 static void cpuhp_offline_cpu_device(unsigned int cpu)
2024 struct device *dev = get_cpu_device(cpu);
2026 dev->offline = true;
2027 /* Tell user space about the state change */
2028 kobject_uevent(&dev->kobj, KOBJ_OFFLINE);
2031 static void cpuhp_online_cpu_device(unsigned int cpu)
2033 struct device *dev = get_cpu_device(cpu);
2035 dev->offline = false;
2036 /* Tell user space about the state change */
2037 kobject_uevent(&dev->kobj, KOBJ_ONLINE);
2040 static int cpuhp_smt_disable(enum cpuhp_smt_control ctrlval)
2044 cpu_maps_update_begin();
2045 for_each_online_cpu(cpu) {
2046 if (topology_is_primary_thread(cpu))
2048 ret = cpu_down_maps_locked(cpu, CPUHP_OFFLINE);
2052 * As this needs to hold the cpu maps lock it's impossible
2053 * to call device_offline() because that ends up calling
2054 * cpu_down() which takes cpu maps lock. cpu maps lock
2055 * needs to be held as this might race against in kernel
2056 * abusers of the hotplug machinery (thermal management).
2058 * So nothing would update device:offline state. That would
2059 * leave the sysfs entry stale and prevent onlining after
2060 * smt control has been changed to 'off' again. This is
2061 * called under the sysfs hotplug lock, so it is properly
2062 * serialized against the regular offline usage.
2064 cpuhp_offline_cpu_device(cpu);
2067 cpu_smt_control = ctrlval;
2068 cpu_maps_update_done();
2072 static int cpuhp_smt_enable(void)
2076 cpu_maps_update_begin();
2077 cpu_smt_control = CPU_SMT_ENABLED;
2078 for_each_present_cpu(cpu) {
2079 /* Skip online CPUs and CPUs on offline nodes */
2080 if (cpu_online(cpu) || !node_online(cpu_to_node(cpu)))
2082 ret = _cpu_up(cpu, 0, CPUHP_ONLINE);
2085 /* See comment in cpuhp_smt_disable() */
2086 cpuhp_online_cpu_device(cpu);
2088 cpu_maps_update_done();
2093 store_smt_control(struct device *dev, struct device_attribute *attr,
2094 const char *buf, size_t count)
2098 if (sysfs_streq(buf, "on"))
2099 ctrlval = CPU_SMT_ENABLED;
2100 else if (sysfs_streq(buf, "off"))
2101 ctrlval = CPU_SMT_DISABLED;
2102 else if (sysfs_streq(buf, "forceoff"))
2103 ctrlval = CPU_SMT_FORCE_DISABLED;
2107 if (cpu_smt_control == CPU_SMT_FORCE_DISABLED)
2110 if (cpu_smt_control == CPU_SMT_NOT_SUPPORTED)
2113 ret = lock_device_hotplug_sysfs();
2117 if (ctrlval != cpu_smt_control) {
2119 case CPU_SMT_ENABLED:
2120 ret = cpuhp_smt_enable();
2122 case CPU_SMT_DISABLED:
2123 case CPU_SMT_FORCE_DISABLED:
2124 ret = cpuhp_smt_disable(ctrlval);
2129 unlock_device_hotplug();
2130 return ret ? ret : count;
2132 static DEVICE_ATTR(control, 0644, show_smt_control, store_smt_control);
2135 show_smt_active(struct device *dev, struct device_attribute *attr, char *buf)
2137 bool active = topology_max_smt_threads() > 1;
2139 return snprintf(buf, PAGE_SIZE - 2, "%d\n", active);
2141 static DEVICE_ATTR(active, 0444, show_smt_active, NULL);
2143 static struct attribute *cpuhp_smt_attrs[] = {
2144 &dev_attr_control.attr,
2145 &dev_attr_active.attr,
2149 static const struct attribute_group cpuhp_smt_attr_group = {
2150 .attrs = cpuhp_smt_attrs,
2155 static int __init cpu_smt_state_init(void)
2157 return sysfs_create_group(&cpu_subsys.dev_root->kobj,
2158 &cpuhp_smt_attr_group);
2162 static inline int cpu_smt_state_init(void) { return 0; }
2165 static int __init cpuhp_sysfs_init(void)
2169 ret = cpu_smt_state_init();
2173 ret = sysfs_create_group(&cpu_subsys.dev_root->kobj,
2174 &cpuhp_cpu_root_attr_group);
2178 for_each_possible_cpu(cpu) {
2179 struct device *dev = get_cpu_device(cpu);
2183 ret = sysfs_create_group(&dev->kobj, &cpuhp_cpu_attr_group);
2189 device_initcall(cpuhp_sysfs_init);
2193 * cpu_bit_bitmap[] is a special, "compressed" data structure that
2194 * represents all NR_CPUS bits binary values of 1<<nr.
2196 * It is used by cpumask_of() to get a constant address to a CPU
2197 * mask value that has a single bit set only.
2200 /* cpu_bit_bitmap[0] is empty - so we can back into it */
2201 #define MASK_DECLARE_1(x) [x+1][0] = (1UL << (x))
2202 #define MASK_DECLARE_2(x) MASK_DECLARE_1(x), MASK_DECLARE_1(x+1)
2203 #define MASK_DECLARE_4(x) MASK_DECLARE_2(x), MASK_DECLARE_2(x+2)
2204 #define MASK_DECLARE_8(x) MASK_DECLARE_4(x), MASK_DECLARE_4(x+4)
2206 const unsigned long cpu_bit_bitmap[BITS_PER_LONG+1][BITS_TO_LONGS(NR_CPUS)] = {
2208 MASK_DECLARE_8(0), MASK_DECLARE_8(8),
2209 MASK_DECLARE_8(16), MASK_DECLARE_8(24),
2210 #if BITS_PER_LONG > 32
2211 MASK_DECLARE_8(32), MASK_DECLARE_8(40),
2212 MASK_DECLARE_8(48), MASK_DECLARE_8(56),
2215 EXPORT_SYMBOL_GPL(cpu_bit_bitmap);
2217 const DECLARE_BITMAP(cpu_all_bits, NR_CPUS) = CPU_BITS_ALL;
2218 EXPORT_SYMBOL(cpu_all_bits);
2220 #ifdef CONFIG_INIT_ALL_POSSIBLE
2221 struct cpumask __cpu_possible_mask __read_mostly
2224 struct cpumask __cpu_possible_mask __read_mostly;
2226 EXPORT_SYMBOL(__cpu_possible_mask);
2228 struct cpumask __cpu_online_mask __read_mostly;
2229 EXPORT_SYMBOL(__cpu_online_mask);
2231 struct cpumask __cpu_present_mask __read_mostly;
2232 EXPORT_SYMBOL(__cpu_present_mask);
2234 struct cpumask __cpu_active_mask __read_mostly;
2235 EXPORT_SYMBOL(__cpu_active_mask);
2237 void init_cpu_present(const struct cpumask *src)
2239 cpumask_copy(&__cpu_present_mask, src);
2242 void init_cpu_possible(const struct cpumask *src)
2244 cpumask_copy(&__cpu_possible_mask, src);
2247 void init_cpu_online(const struct cpumask *src)
2249 cpumask_copy(&__cpu_online_mask, src);
2253 * Activate the first processor.
2255 void __init boot_cpu_init(void)
2257 int cpu = smp_processor_id();
2259 /* Mark the boot cpu "present", "online" etc for SMP and UP case */
2260 set_cpu_online(cpu, true);
2261 set_cpu_active(cpu, true);
2262 set_cpu_present(cpu, true);
2263 set_cpu_possible(cpu, true);
2266 __boot_cpu_id = cpu;
2271 * Must be called _AFTER_ setting up the per_cpu areas
2273 void __init boot_cpu_hotplug_init(void)
2275 this_cpu_write(cpuhp_state.booted_once, true);
2276 this_cpu_write(cpuhp_state.state, CPUHP_ONLINE);