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;
63 struct hlist_node *node;
64 struct hlist_node *last;
65 enum cpuhp_state cb_state;
67 struct completion done_up;
68 struct completion done_down;
72 static DEFINE_PER_CPU(struct cpuhp_cpu_state, cpuhp_state) = {
73 .fail = CPUHP_INVALID,
76 #if defined(CONFIG_LOCKDEP) && defined(CONFIG_SMP)
77 static struct lockdep_map cpuhp_state_up_map =
78 STATIC_LOCKDEP_MAP_INIT("cpuhp_state-up", &cpuhp_state_up_map);
79 static struct lockdep_map cpuhp_state_down_map =
80 STATIC_LOCKDEP_MAP_INIT("cpuhp_state-down", &cpuhp_state_down_map);
83 static void inline cpuhp_lock_acquire(bool bringup)
85 lock_map_acquire(bringup ? &cpuhp_state_up_map : &cpuhp_state_down_map);
88 static void inline cpuhp_lock_release(bool bringup)
90 lock_map_release(bringup ? &cpuhp_state_up_map : &cpuhp_state_down_map);
94 static void inline cpuhp_lock_acquire(bool bringup) { }
95 static void inline cpuhp_lock_release(bool bringup) { }
100 * cpuhp_step - Hotplug state machine step
101 * @name: Name of the step
102 * @startup: Startup function of the step
103 * @teardown: Teardown function of the step
104 * @skip_onerr: Do not invoke the functions on error rollback
105 * Will go away once the notifiers are gone
106 * @cant_stop: Bringup/teardown can't be stopped at this step
111 int (*single)(unsigned int cpu);
112 int (*multi)(unsigned int cpu,
113 struct hlist_node *node);
116 int (*single)(unsigned int cpu);
117 int (*multi)(unsigned int cpu,
118 struct hlist_node *node);
120 struct hlist_head list;
126 static DEFINE_MUTEX(cpuhp_state_mutex);
127 static struct cpuhp_step cpuhp_bp_states[];
128 static struct cpuhp_step cpuhp_ap_states[];
130 static bool cpuhp_is_ap_state(enum cpuhp_state state)
133 * The extra check for CPUHP_TEARDOWN_CPU is only for documentation
134 * purposes as that state is handled explicitly in cpu_down.
136 return state > CPUHP_BRINGUP_CPU && state != CPUHP_TEARDOWN_CPU;
139 static struct cpuhp_step *cpuhp_get_step(enum cpuhp_state state)
141 struct cpuhp_step *sp;
143 sp = cpuhp_is_ap_state(state) ? cpuhp_ap_states : cpuhp_bp_states;
148 * cpuhp_invoke_callback _ Invoke the callbacks for a given state
149 * @cpu: The cpu for which the callback should be invoked
150 * @state: The state to do callbacks for
151 * @bringup: True if the bringup callback should be invoked
152 * @node: For multi-instance, do a single entry callback for install/remove
153 * @lastp: For multi-instance rollback, remember how far we got
155 * Called from cpu hotplug and from the state register machinery.
157 static int cpuhp_invoke_callback(unsigned int cpu, enum cpuhp_state state,
158 bool bringup, struct hlist_node *node,
159 struct hlist_node **lastp)
161 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
162 struct cpuhp_step *step = cpuhp_get_step(state);
163 int (*cbm)(unsigned int cpu, struct hlist_node *node);
164 int (*cb)(unsigned int cpu);
167 if (st->fail == state) {
168 st->fail = CPUHP_INVALID;
170 if (!(bringup ? step->startup.single : step->teardown.single))
176 if (!step->multi_instance) {
177 WARN_ON_ONCE(lastp && *lastp);
178 cb = bringup ? step->startup.single : step->teardown.single;
181 trace_cpuhp_enter(cpu, st->target, state, cb);
183 trace_cpuhp_exit(cpu, st->state, state, ret);
186 cbm = bringup ? step->startup.multi : step->teardown.multi;
190 /* Single invocation for instance add/remove */
192 WARN_ON_ONCE(lastp && *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);
199 /* State transition. Invoke on all instances */
201 hlist_for_each(node, &step->list) {
202 if (lastp && node == *lastp)
205 trace_cpuhp_multi_enter(cpu, st->target, state, cbm, node);
206 ret = cbm(cpu, node);
207 trace_cpuhp_exit(cpu, st->state, state, ret);
221 /* Rollback the instances if one failed */
222 cbm = !bringup ? step->startup.multi : step->teardown.multi;
226 hlist_for_each(node, &step->list) {
230 trace_cpuhp_multi_enter(cpu, st->target, state, cbm, node);
231 ret = cbm(cpu, node);
232 trace_cpuhp_exit(cpu, st->state, state, ret);
234 * Rollback must not fail,
242 static inline void wait_for_ap_thread(struct cpuhp_cpu_state *st, bool bringup)
244 struct completion *done = bringup ? &st->done_up : &st->done_down;
245 wait_for_completion(done);
248 static inline void complete_ap_thread(struct cpuhp_cpu_state *st, bool bringup)
250 struct completion *done = bringup ? &st->done_up : &st->done_down;
255 * The former STARTING/DYING states, ran with IRQs disabled and must not fail.
257 static bool cpuhp_is_atomic_state(enum cpuhp_state state)
259 return CPUHP_AP_IDLE_DEAD <= state && state < CPUHP_AP_ONLINE;
262 /* Serializes the updates to cpu_online_mask, cpu_present_mask */
263 static DEFINE_MUTEX(cpu_add_remove_lock);
264 bool cpuhp_tasks_frozen;
265 EXPORT_SYMBOL_GPL(cpuhp_tasks_frozen);
268 * The following two APIs (cpu_maps_update_begin/done) must be used when
269 * attempting to serialize the updates to cpu_online_mask & cpu_present_mask.
271 void cpu_maps_update_begin(void)
273 mutex_lock(&cpu_add_remove_lock);
276 void cpu_maps_update_done(void)
278 mutex_unlock(&cpu_add_remove_lock);
282 * If set, cpu_up and cpu_down will return -EBUSY and do nothing.
283 * Should always be manipulated under cpu_add_remove_lock
285 static int cpu_hotplug_disabled;
287 #ifdef CONFIG_HOTPLUG_CPU
289 DEFINE_STATIC_PERCPU_RWSEM(cpu_hotplug_lock);
291 void cpus_read_lock(void)
293 percpu_down_read(&cpu_hotplug_lock);
295 EXPORT_SYMBOL_GPL(cpus_read_lock);
297 void cpus_read_unlock(void)
299 percpu_up_read(&cpu_hotplug_lock);
301 EXPORT_SYMBOL_GPL(cpus_read_unlock);
303 void cpus_write_lock(void)
305 percpu_down_write(&cpu_hotplug_lock);
308 void cpus_write_unlock(void)
310 percpu_up_write(&cpu_hotplug_lock);
313 void lockdep_assert_cpus_held(void)
315 percpu_rwsem_assert_held(&cpu_hotplug_lock);
319 * Wait for currently running CPU hotplug operations to complete (if any) and
320 * disable future CPU hotplug (from sysfs). The 'cpu_add_remove_lock' protects
321 * the 'cpu_hotplug_disabled' flag. The same lock is also acquired by the
322 * hotplug path before performing hotplug operations. So acquiring that lock
323 * guarantees mutual exclusion from any currently running hotplug operations.
325 void cpu_hotplug_disable(void)
327 cpu_maps_update_begin();
328 cpu_hotplug_disabled++;
329 cpu_maps_update_done();
331 EXPORT_SYMBOL_GPL(cpu_hotplug_disable);
333 static void __cpu_hotplug_enable(void)
335 if (WARN_ONCE(!cpu_hotplug_disabled, "Unbalanced cpu hotplug enable\n"))
337 cpu_hotplug_disabled--;
340 void cpu_hotplug_enable(void)
342 cpu_maps_update_begin();
343 __cpu_hotplug_enable();
344 cpu_maps_update_done();
346 EXPORT_SYMBOL_GPL(cpu_hotplug_enable);
347 #endif /* CONFIG_HOTPLUG_CPU */
349 static inline enum cpuhp_state
350 cpuhp_set_state(struct cpuhp_cpu_state *st, enum cpuhp_state target)
352 enum cpuhp_state prev_state = st->state;
354 st->rollback = false;
359 st->bringup = st->state < target;
365 cpuhp_reset_state(struct cpuhp_cpu_state *st, enum cpuhp_state prev_state)
370 * If we have st->last we need to undo partial multi_instance of this
371 * state first. Otherwise start undo at the previous state.
380 st->target = prev_state;
381 st->bringup = !st->bringup;
384 /* Regular hotplug invocation of the AP hotplug thread */
385 static void __cpuhp_kick_ap(struct cpuhp_cpu_state *st)
387 if (!st->single && st->state == st->target)
392 * Make sure the above stores are visible before should_run becomes
393 * true. Paired with the mb() above in cpuhp_thread_fun()
396 st->should_run = true;
397 wake_up_process(st->thread);
398 wait_for_ap_thread(st, st->bringup);
401 static int cpuhp_kick_ap(struct cpuhp_cpu_state *st, enum cpuhp_state target)
403 enum cpuhp_state prev_state;
406 prev_state = cpuhp_set_state(st, target);
408 if ((ret = st->result)) {
409 cpuhp_reset_state(st, prev_state);
416 static int bringup_wait_for_ap(unsigned int cpu)
418 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
420 /* Wait for the CPU to reach CPUHP_AP_ONLINE_IDLE */
421 wait_for_ap_thread(st, true);
422 if (WARN_ON_ONCE((!cpu_online(cpu))))
425 /* Unpark the stopper thread and the hotplug thread of the target cpu */
426 stop_machine_unpark(cpu);
427 kthread_unpark(st->thread);
429 if (st->target <= CPUHP_AP_ONLINE_IDLE)
432 return cpuhp_kick_ap(st, st->target);
435 static int bringup_cpu(unsigned int cpu)
437 struct task_struct *idle = idle_thread_get(cpu);
441 * Some architectures have to walk the irq descriptors to
442 * setup the vector space for the cpu which comes online.
443 * Prevent irq alloc/free across the bringup.
447 /* Arch-specific enabling code. */
448 ret = __cpu_up(cpu, idle);
452 return bringup_wait_for_ap(cpu);
456 * Hotplug state machine related functions
459 static void undo_cpu_up(unsigned int cpu, struct cpuhp_cpu_state *st)
461 for (st->state--; st->state > st->target; st->state--) {
462 struct cpuhp_step *step = cpuhp_get_step(st->state);
464 if (!step->skip_onerr)
465 cpuhp_invoke_callback(cpu, st->state, false, NULL, NULL);
469 static int cpuhp_up_callbacks(unsigned int cpu, struct cpuhp_cpu_state *st,
470 enum cpuhp_state target)
472 enum cpuhp_state prev_state = st->state;
475 while (st->state < target) {
477 ret = cpuhp_invoke_callback(cpu, st->state, true, NULL, NULL);
479 st->target = prev_state;
480 undo_cpu_up(cpu, st);
488 * The cpu hotplug threads manage the bringup and teardown of the cpus
490 static void cpuhp_create(unsigned int cpu)
492 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
494 init_completion(&st->done_up);
495 init_completion(&st->done_down);
498 static int cpuhp_should_run(unsigned int cpu)
500 struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
502 return st->should_run;
506 * Execute teardown/startup callbacks on the plugged cpu. Also used to invoke
507 * callbacks when a state gets [un]installed at runtime.
509 * Each invocation of this function by the smpboot thread does a single AP
512 * It has 3 modes of operation:
513 * - single: runs st->cb_state
514 * - up: runs ++st->state, while st->state < st->target
515 * - down: runs st->state--, while st->state > st->target
517 * When complete or on error, should_run is cleared and the completion is fired.
519 static void cpuhp_thread_fun(unsigned int cpu)
521 struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
522 bool bringup = st->bringup;
523 enum cpuhp_state state;
526 * ACQUIRE for the cpuhp_should_run() load of ->should_run. Ensures
527 * that if we see ->should_run we also see the rest of the state.
531 if (WARN_ON_ONCE(!st->should_run))
534 cpuhp_lock_acquire(bringup);
537 state = st->cb_state;
538 st->should_run = false;
543 st->should_run = (st->state < st->target);
544 WARN_ON_ONCE(st->state > st->target);
548 st->should_run = (st->state > st->target);
549 WARN_ON_ONCE(st->state < st->target);
553 WARN_ON_ONCE(!cpuhp_is_ap_state(state));
556 struct cpuhp_step *step = cpuhp_get_step(state);
557 if (step->skip_onerr)
561 if (cpuhp_is_atomic_state(state)) {
563 st->result = cpuhp_invoke_callback(cpu, state, bringup, st->node, &st->last);
567 * STARTING/DYING must not fail!
569 WARN_ON_ONCE(st->result);
571 st->result = cpuhp_invoke_callback(cpu, state, bringup, st->node, &st->last);
576 * If we fail on a rollback, we're up a creek without no
577 * paddle, no way forward, no way back. We loose, thanks for
580 WARN_ON_ONCE(st->rollback);
581 st->should_run = false;
585 cpuhp_lock_release(bringup);
588 complete_ap_thread(st, bringup);
591 /* Invoke a single callback on a remote cpu */
593 cpuhp_invoke_ap_callback(int cpu, enum cpuhp_state state, bool bringup,
594 struct hlist_node *node)
596 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
599 if (!cpu_online(cpu))
602 cpuhp_lock_acquire(false);
603 cpuhp_lock_release(false);
605 cpuhp_lock_acquire(true);
606 cpuhp_lock_release(true);
609 * If we are up and running, use the hotplug thread. For early calls
610 * we invoke the thread function directly.
613 return cpuhp_invoke_callback(cpu, state, bringup, node, NULL);
615 st->rollback = false;
619 st->bringup = bringup;
620 st->cb_state = state;
626 * If we failed and did a partial, do a rollback.
628 if ((ret = st->result) && st->last) {
630 st->bringup = !bringup;
636 * Clean up the leftovers so the next hotplug operation wont use stale
639 st->node = st->last = NULL;
643 static int cpuhp_kick_ap_work(unsigned int cpu)
645 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
646 enum cpuhp_state prev_state = st->state;
649 cpuhp_lock_acquire(false);
650 cpuhp_lock_release(false);
652 cpuhp_lock_acquire(true);
653 cpuhp_lock_release(true);
655 trace_cpuhp_enter(cpu, st->target, prev_state, cpuhp_kick_ap_work);
656 ret = cpuhp_kick_ap(st, st->target);
657 trace_cpuhp_exit(cpu, st->state, prev_state, ret);
662 static struct smp_hotplug_thread cpuhp_threads = {
663 .store = &cpuhp_state.thread,
664 .create = &cpuhp_create,
665 .thread_should_run = cpuhp_should_run,
666 .thread_fn = cpuhp_thread_fun,
667 .thread_comm = "cpuhp/%u",
671 void __init cpuhp_threads_init(void)
673 BUG_ON(smpboot_register_percpu_thread(&cpuhp_threads));
674 kthread_unpark(this_cpu_read(cpuhp_state.thread));
677 #ifdef CONFIG_HOTPLUG_CPU
679 * clear_tasks_mm_cpumask - Safely clear tasks' mm_cpumask for a CPU
682 * This function walks all processes, finds a valid mm struct for each one and
683 * then clears a corresponding bit in mm's cpumask. While this all sounds
684 * trivial, there are various non-obvious corner cases, which this function
685 * tries to solve in a safe manner.
687 * Also note that the function uses a somewhat relaxed locking scheme, so it may
688 * be called only for an already offlined CPU.
690 void clear_tasks_mm_cpumask(int cpu)
692 struct task_struct *p;
695 * This function is called after the cpu is taken down and marked
696 * offline, so its not like new tasks will ever get this cpu set in
697 * their mm mask. -- Peter Zijlstra
698 * Thus, we may use rcu_read_lock() here, instead of grabbing
699 * full-fledged tasklist_lock.
701 WARN_ON(cpu_online(cpu));
703 for_each_process(p) {
704 struct task_struct *t;
707 * Main thread might exit, but other threads may still have
708 * a valid mm. Find one.
710 t = find_lock_task_mm(p);
713 cpumask_clear_cpu(cpu, mm_cpumask(t->mm));
719 /* Take this CPU down. */
720 static int take_cpu_down(void *_param)
722 struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
723 enum cpuhp_state target = max((int)st->target, CPUHP_AP_OFFLINE);
724 int err, cpu = smp_processor_id();
727 /* Ensure this CPU doesn't handle any more interrupts. */
728 err = __cpu_disable();
733 * We get here while we are in CPUHP_TEARDOWN_CPU state and we must not
734 * do this step again.
736 WARN_ON(st->state != CPUHP_TEARDOWN_CPU);
738 /* Invoke the former CPU_DYING callbacks */
739 for (; st->state > target; st->state--) {
740 ret = cpuhp_invoke_callback(cpu, st->state, false, NULL, NULL);
742 * DYING must not fail!
747 /* Give up timekeeping duties */
748 tick_handover_do_timer();
749 /* Park the stopper thread */
750 stop_machine_park(cpu);
754 static int takedown_cpu(unsigned int cpu)
756 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
759 /* Park the smpboot threads */
760 kthread_park(per_cpu_ptr(&cpuhp_state, cpu)->thread);
763 * Prevent irq alloc/free while the dying cpu reorganizes the
764 * interrupt affinities.
769 * So now all preempt/rcu users must observe !cpu_active().
771 err = stop_machine_cpuslocked(take_cpu_down, NULL, cpumask_of(cpu));
773 /* CPU refused to die */
775 /* Unpark the hotplug thread so we can rollback there */
776 kthread_unpark(per_cpu_ptr(&cpuhp_state, cpu)->thread);
779 BUG_ON(cpu_online(cpu));
782 * The CPUHP_AP_SCHED_MIGRATE_DYING callback will have removed all
783 * runnable tasks from the cpu, there's only the idle task left now
784 * that the migration thread is done doing the stop_machine thing.
786 * Wait for the stop thread to go away.
788 wait_for_ap_thread(st, false);
789 BUG_ON(st->state != CPUHP_AP_IDLE_DEAD);
791 /* Interrupts are moved away from the dying cpu, reenable alloc/free */
794 hotplug_cpu__broadcast_tick_pull(cpu);
795 /* This actually kills the CPU. */
798 tick_cleanup_dead_cpu(cpu);
799 rcutree_migrate_callbacks(cpu);
803 static void cpuhp_complete_idle_dead(void *arg)
805 struct cpuhp_cpu_state *st = arg;
807 complete_ap_thread(st, false);
810 void cpuhp_report_idle_dead(void)
812 struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
814 BUG_ON(st->state != CPUHP_AP_OFFLINE);
815 rcu_report_dead(smp_processor_id());
816 st->state = CPUHP_AP_IDLE_DEAD;
818 * We cannot call complete after rcu_report_dead() so we delegate it
821 smp_call_function_single(cpumask_first(cpu_online_mask),
822 cpuhp_complete_idle_dead, st, 0);
825 static void undo_cpu_down(unsigned int cpu, struct cpuhp_cpu_state *st)
827 for (st->state++; st->state < st->target; st->state++) {
828 struct cpuhp_step *step = cpuhp_get_step(st->state);
830 if (!step->skip_onerr)
831 cpuhp_invoke_callback(cpu, st->state, true, NULL, NULL);
835 static int cpuhp_down_callbacks(unsigned int cpu, struct cpuhp_cpu_state *st,
836 enum cpuhp_state target)
838 enum cpuhp_state prev_state = st->state;
841 for (; st->state > target; st->state--) {
842 ret = cpuhp_invoke_callback(cpu, st->state, false, NULL, NULL);
844 st->target = prev_state;
845 undo_cpu_down(cpu, st);
852 /* Requires cpu_add_remove_lock to be held */
853 static int __ref _cpu_down(unsigned int cpu, int tasks_frozen,
854 enum cpuhp_state target)
856 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
857 int prev_state, ret = 0;
859 if (num_online_cpus() == 1)
862 if (!cpu_present(cpu))
867 cpuhp_tasks_frozen = tasks_frozen;
869 prev_state = cpuhp_set_state(st, target);
871 * If the current CPU state is in the range of the AP hotplug thread,
872 * then we need to kick the thread.
874 if (st->state > CPUHP_TEARDOWN_CPU) {
875 st->target = max((int)target, CPUHP_TEARDOWN_CPU);
876 ret = cpuhp_kick_ap_work(cpu);
878 * The AP side has done the error rollback already. Just
879 * return the error code..
885 * We might have stopped still in the range of the AP hotplug
886 * thread. Nothing to do anymore.
888 if (st->state > CPUHP_TEARDOWN_CPU)
894 * The AP brought itself down to CPUHP_TEARDOWN_CPU. So we need
895 * to do the further cleanups.
897 ret = cpuhp_down_callbacks(cpu, st, target);
898 if (ret && st->state > CPUHP_TEARDOWN_CPU && st->state < prev_state) {
899 cpuhp_reset_state(st, prev_state);
906 * Do post unplug cleanup. This is still protected against
907 * concurrent CPU hotplug via cpu_add_remove_lock.
909 lockup_detector_cleanup();
913 static int cpu_down_maps_locked(unsigned int cpu, enum cpuhp_state target)
915 if (cpu_hotplug_disabled)
917 return _cpu_down(cpu, 0, target);
920 static int do_cpu_down(unsigned int cpu, enum cpuhp_state target)
924 cpu_maps_update_begin();
925 err = cpu_down_maps_locked(cpu, target);
926 cpu_maps_update_done();
930 int cpu_down(unsigned int cpu)
932 return do_cpu_down(cpu, CPUHP_OFFLINE);
934 EXPORT_SYMBOL(cpu_down);
937 #define takedown_cpu NULL
938 #endif /*CONFIG_HOTPLUG_CPU*/
940 #ifdef CONFIG_HOTPLUG_SMT
941 enum cpuhp_smt_control cpu_smt_control __read_mostly = CPU_SMT_ENABLED;
943 static int __init smt_cmdline_disable(char *str)
945 cpu_smt_control = CPU_SMT_DISABLED;
946 if (str && !strcmp(str, "force")) {
947 pr_info("SMT: Force disabled\n");
948 cpu_smt_control = CPU_SMT_FORCE_DISABLED;
952 early_param("nosmt", smt_cmdline_disable);
954 static inline bool cpu_smt_allowed(unsigned int cpu)
956 return cpu_smt_control == CPU_SMT_ENABLED ||
957 topology_is_primary_thread(cpu);
960 static inline bool cpu_smt_allowed(unsigned int cpu) { return true; }
964 * notify_cpu_starting(cpu) - Invoke the callbacks on the starting CPU
965 * @cpu: cpu that just started
967 * It must be called by the arch code on the new cpu, before the new cpu
968 * enables interrupts and before the "boot" cpu returns from __cpu_up().
970 void notify_cpu_starting(unsigned int cpu)
972 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
973 enum cpuhp_state target = min((int)st->target, CPUHP_AP_ONLINE);
976 rcu_cpu_starting(cpu); /* Enables RCU usage on this CPU. */
977 while (st->state < target) {
979 ret = cpuhp_invoke_callback(cpu, st->state, true, NULL, NULL);
981 * STARTING must not fail!
988 * Called from the idle task. Wake up the controlling task which brings the
989 * stopper and the hotplug thread of the upcoming CPU up and then delegates
990 * the rest of the online bringup to the hotplug thread.
992 void cpuhp_online_idle(enum cpuhp_state state)
994 struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
996 /* Happens for the boot cpu */
997 if (state != CPUHP_AP_ONLINE_IDLE)
1000 st->state = CPUHP_AP_ONLINE_IDLE;
1001 complete_ap_thread(st, true);
1004 /* Requires cpu_add_remove_lock to be held */
1005 static int _cpu_up(unsigned int cpu, int tasks_frozen, enum cpuhp_state target)
1007 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1008 struct task_struct *idle;
1013 if (!cpu_present(cpu)) {
1019 * The caller of do_cpu_up might have raced with another
1020 * caller. Ignore it for now.
1022 if (st->state >= target)
1025 if (st->state == CPUHP_OFFLINE) {
1026 /* Let it fail before we try to bring the cpu up */
1027 idle = idle_thread_get(cpu);
1029 ret = PTR_ERR(idle);
1034 cpuhp_tasks_frozen = tasks_frozen;
1036 cpuhp_set_state(st, target);
1038 * If the current CPU state is in the range of the AP hotplug thread,
1039 * then we need to kick the thread once more.
1041 if (st->state > CPUHP_BRINGUP_CPU) {
1042 ret = cpuhp_kick_ap_work(cpu);
1044 * The AP side has done the error rollback already. Just
1045 * return the error code..
1052 * Try to reach the target state. We max out on the BP at
1053 * CPUHP_BRINGUP_CPU. After that the AP hotplug thread is
1054 * responsible for bringing it up to the target state.
1056 target = min((int)target, CPUHP_BRINGUP_CPU);
1057 ret = cpuhp_up_callbacks(cpu, st, target);
1059 cpus_write_unlock();
1063 static int do_cpu_up(unsigned int cpu, enum cpuhp_state target)
1067 if (!cpu_possible(cpu)) {
1068 pr_err("can't online cpu %d because it is not configured as may-hotadd at boot time\n",
1070 #if defined(CONFIG_IA64)
1071 pr_err("please check additional_cpus= boot parameter\n");
1076 err = try_online_node(cpu_to_node(cpu));
1080 cpu_maps_update_begin();
1082 if (cpu_hotplug_disabled) {
1086 if (!cpu_smt_allowed(cpu)) {
1091 err = _cpu_up(cpu, 0, target);
1093 cpu_maps_update_done();
1097 int cpu_up(unsigned int cpu)
1099 return do_cpu_up(cpu, CPUHP_ONLINE);
1101 EXPORT_SYMBOL_GPL(cpu_up);
1103 #ifdef CONFIG_PM_SLEEP_SMP
1104 static cpumask_var_t frozen_cpus;
1106 int freeze_secondary_cpus(int primary)
1110 cpu_maps_update_begin();
1111 if (!cpu_online(primary))
1112 primary = cpumask_first(cpu_online_mask);
1114 * We take down all of the non-boot CPUs in one shot to avoid races
1115 * with the userspace trying to use the CPU hotplug at the same time
1117 cpumask_clear(frozen_cpus);
1119 pr_info("Disabling non-boot CPUs ...\n");
1120 for_each_online_cpu(cpu) {
1123 trace_suspend_resume(TPS("CPU_OFF"), cpu, true);
1124 error = _cpu_down(cpu, 1, CPUHP_OFFLINE);
1125 trace_suspend_resume(TPS("CPU_OFF"), cpu, false);
1127 cpumask_set_cpu(cpu, frozen_cpus);
1129 pr_err("Error taking CPU%d down: %d\n", cpu, error);
1135 BUG_ON(num_online_cpus() > 1);
1137 pr_err("Non-boot CPUs are not disabled\n");
1140 * Make sure the CPUs won't be enabled by someone else. We need to do
1141 * this even in case of failure as all disable_nonboot_cpus() users are
1142 * supposed to do enable_nonboot_cpus() on the failure path.
1144 cpu_hotplug_disabled++;
1146 cpu_maps_update_done();
1150 void __weak arch_enable_nonboot_cpus_begin(void)
1154 void __weak arch_enable_nonboot_cpus_end(void)
1158 void enable_nonboot_cpus(void)
1162 /* Allow everyone to use the CPU hotplug again */
1163 cpu_maps_update_begin();
1164 __cpu_hotplug_enable();
1165 if (cpumask_empty(frozen_cpus))
1168 pr_info("Enabling non-boot CPUs ...\n");
1170 arch_enable_nonboot_cpus_begin();
1172 for_each_cpu(cpu, frozen_cpus) {
1173 trace_suspend_resume(TPS("CPU_ON"), cpu, true);
1174 error = _cpu_up(cpu, 1, CPUHP_ONLINE);
1175 trace_suspend_resume(TPS("CPU_ON"), cpu, false);
1177 pr_info("CPU%d is up\n", cpu);
1180 pr_warn("Error taking CPU%d up: %d\n", cpu, error);
1183 arch_enable_nonboot_cpus_end();
1185 cpumask_clear(frozen_cpus);
1187 cpu_maps_update_done();
1190 static int __init alloc_frozen_cpus(void)
1192 if (!alloc_cpumask_var(&frozen_cpus, GFP_KERNEL|__GFP_ZERO))
1196 core_initcall(alloc_frozen_cpus);
1199 * When callbacks for CPU hotplug notifications are being executed, we must
1200 * ensure that the state of the system with respect to the tasks being frozen
1201 * or not, as reported by the notification, remains unchanged *throughout the
1202 * duration* of the execution of the callbacks.
1203 * Hence we need to prevent the freezer from racing with regular CPU hotplug.
1205 * This synchronization is implemented by mutually excluding regular CPU
1206 * hotplug and Suspend/Hibernate call paths by hooking onto the Suspend/
1207 * Hibernate notifications.
1210 cpu_hotplug_pm_callback(struct notifier_block *nb,
1211 unsigned long action, void *ptr)
1215 case PM_SUSPEND_PREPARE:
1216 case PM_HIBERNATION_PREPARE:
1217 cpu_hotplug_disable();
1220 case PM_POST_SUSPEND:
1221 case PM_POST_HIBERNATION:
1222 cpu_hotplug_enable();
1233 static int __init cpu_hotplug_pm_sync_init(void)
1236 * cpu_hotplug_pm_callback has higher priority than x86
1237 * bsp_pm_callback which depends on cpu_hotplug_pm_callback
1238 * to disable cpu hotplug to avoid cpu hotplug race.
1240 pm_notifier(cpu_hotplug_pm_callback, 0);
1243 core_initcall(cpu_hotplug_pm_sync_init);
1245 #endif /* CONFIG_PM_SLEEP_SMP */
1249 #endif /* CONFIG_SMP */
1251 /* Boot processor state steps */
1252 static struct cpuhp_step cpuhp_bp_states[] = {
1255 .startup.single = NULL,
1256 .teardown.single = NULL,
1259 [CPUHP_CREATE_THREADS]= {
1260 .name = "threads:prepare",
1261 .startup.single = smpboot_create_threads,
1262 .teardown.single = NULL,
1265 [CPUHP_PERF_PREPARE] = {
1266 .name = "perf:prepare",
1267 .startup.single = perf_event_init_cpu,
1268 .teardown.single = perf_event_exit_cpu,
1270 [CPUHP_WORKQUEUE_PREP] = {
1271 .name = "workqueue:prepare",
1272 .startup.single = workqueue_prepare_cpu,
1273 .teardown.single = NULL,
1275 [CPUHP_HRTIMERS_PREPARE] = {
1276 .name = "hrtimers:prepare",
1277 .startup.single = hrtimers_prepare_cpu,
1278 .teardown.single = hrtimers_dead_cpu,
1280 [CPUHP_SMPCFD_PREPARE] = {
1281 .name = "smpcfd:prepare",
1282 .startup.single = smpcfd_prepare_cpu,
1283 .teardown.single = smpcfd_dead_cpu,
1285 [CPUHP_RELAY_PREPARE] = {
1286 .name = "relay:prepare",
1287 .startup.single = relay_prepare_cpu,
1288 .teardown.single = NULL,
1290 [CPUHP_SLAB_PREPARE] = {
1291 .name = "slab:prepare",
1292 .startup.single = slab_prepare_cpu,
1293 .teardown.single = slab_dead_cpu,
1295 [CPUHP_RCUTREE_PREP] = {
1296 .name = "RCU/tree:prepare",
1297 .startup.single = rcutree_prepare_cpu,
1298 .teardown.single = rcutree_dead_cpu,
1301 * On the tear-down path, timers_dead_cpu() must be invoked
1302 * before blk_mq_queue_reinit_notify() from notify_dead(),
1303 * otherwise a RCU stall occurs.
1305 [CPUHP_TIMERS_PREPARE] = {
1306 .name = "timers:dead",
1307 .startup.single = timers_prepare_cpu,
1308 .teardown.single = timers_dead_cpu,
1310 /* Kicks the plugged cpu into life */
1311 [CPUHP_BRINGUP_CPU] = {
1312 .name = "cpu:bringup",
1313 .startup.single = bringup_cpu,
1314 .teardown.single = NULL,
1318 * Handled on controll processor until the plugged processor manages
1321 [CPUHP_TEARDOWN_CPU] = {
1322 .name = "cpu:teardown",
1323 .startup.single = NULL,
1324 .teardown.single = takedown_cpu,
1328 [CPUHP_BRINGUP_CPU] = { },
1332 /* Application processor state steps */
1333 static struct cpuhp_step cpuhp_ap_states[] = {
1335 /* Final state before CPU kills itself */
1336 [CPUHP_AP_IDLE_DEAD] = {
1337 .name = "idle:dead",
1340 * Last state before CPU enters the idle loop to die. Transient state
1341 * for synchronization.
1343 [CPUHP_AP_OFFLINE] = {
1344 .name = "ap:offline",
1347 /* First state is scheduler control. Interrupts are disabled */
1348 [CPUHP_AP_SCHED_STARTING] = {
1349 .name = "sched:starting",
1350 .startup.single = sched_cpu_starting,
1351 .teardown.single = sched_cpu_dying,
1353 [CPUHP_AP_RCUTREE_DYING] = {
1354 .name = "RCU/tree:dying",
1355 .startup.single = NULL,
1356 .teardown.single = rcutree_dying_cpu,
1358 [CPUHP_AP_SMPCFD_DYING] = {
1359 .name = "smpcfd:dying",
1360 .startup.single = NULL,
1361 .teardown.single = smpcfd_dying_cpu,
1363 /* Entry state on starting. Interrupts enabled from here on. Transient
1364 * state for synchronsization */
1365 [CPUHP_AP_ONLINE] = {
1366 .name = "ap:online",
1368 /* Handle smpboot threads park/unpark */
1369 [CPUHP_AP_SMPBOOT_THREADS] = {
1370 .name = "smpboot/threads:online",
1371 .startup.single = smpboot_unpark_threads,
1372 .teardown.single = smpboot_park_threads,
1374 [CPUHP_AP_IRQ_AFFINITY_ONLINE] = {
1375 .name = "irq/affinity:online",
1376 .startup.single = irq_affinity_online_cpu,
1377 .teardown.single = NULL,
1379 [CPUHP_AP_PERF_ONLINE] = {
1380 .name = "perf:online",
1381 .startup.single = perf_event_init_cpu,
1382 .teardown.single = perf_event_exit_cpu,
1384 [CPUHP_AP_WORKQUEUE_ONLINE] = {
1385 .name = "workqueue:online",
1386 .startup.single = workqueue_online_cpu,
1387 .teardown.single = workqueue_offline_cpu,
1389 [CPUHP_AP_RCUTREE_ONLINE] = {
1390 .name = "RCU/tree:online",
1391 .startup.single = rcutree_online_cpu,
1392 .teardown.single = rcutree_offline_cpu,
1396 * The dynamically registered state space is here
1400 /* Last state is scheduler control setting the cpu active */
1401 [CPUHP_AP_ACTIVE] = {
1402 .name = "sched:active",
1403 .startup.single = sched_cpu_activate,
1404 .teardown.single = sched_cpu_deactivate,
1408 /* CPU is fully up and running. */
1411 .startup.single = NULL,
1412 .teardown.single = NULL,
1416 /* Sanity check for callbacks */
1417 static int cpuhp_cb_check(enum cpuhp_state state)
1419 if (state <= CPUHP_OFFLINE || state >= CPUHP_ONLINE)
1425 * Returns a free for dynamic slot assignment of the Online state. The states
1426 * are protected by the cpuhp_slot_states mutex and an empty slot is identified
1427 * by having no name assigned.
1429 static int cpuhp_reserve_state(enum cpuhp_state state)
1431 enum cpuhp_state i, end;
1432 struct cpuhp_step *step;
1435 case CPUHP_AP_ONLINE_DYN:
1436 step = cpuhp_ap_states + CPUHP_AP_ONLINE_DYN;
1437 end = CPUHP_AP_ONLINE_DYN_END;
1439 case CPUHP_BP_PREPARE_DYN:
1440 step = cpuhp_bp_states + CPUHP_BP_PREPARE_DYN;
1441 end = CPUHP_BP_PREPARE_DYN_END;
1447 for (i = state; i <= end; i++, step++) {
1451 WARN(1, "No more dynamic states available for CPU hotplug\n");
1455 static int cpuhp_store_callbacks(enum cpuhp_state state, const char *name,
1456 int (*startup)(unsigned int cpu),
1457 int (*teardown)(unsigned int cpu),
1458 bool multi_instance)
1460 /* (Un)Install the callbacks for further cpu hotplug operations */
1461 struct cpuhp_step *sp;
1465 * If name is NULL, then the state gets removed.
1467 * CPUHP_AP_ONLINE_DYN and CPUHP_BP_PREPARE_DYN are handed out on
1468 * the first allocation from these dynamic ranges, so the removal
1469 * would trigger a new allocation and clear the wrong (already
1470 * empty) state, leaving the callbacks of the to be cleared state
1471 * dangling, which causes wreckage on the next hotplug operation.
1473 if (name && (state == CPUHP_AP_ONLINE_DYN ||
1474 state == CPUHP_BP_PREPARE_DYN)) {
1475 ret = cpuhp_reserve_state(state);
1480 sp = cpuhp_get_step(state);
1481 if (name && sp->name)
1484 sp->startup.single = startup;
1485 sp->teardown.single = teardown;
1487 sp->multi_instance = multi_instance;
1488 INIT_HLIST_HEAD(&sp->list);
1492 static void *cpuhp_get_teardown_cb(enum cpuhp_state state)
1494 return cpuhp_get_step(state)->teardown.single;
1498 * Call the startup/teardown function for a step either on the AP or
1499 * on the current CPU.
1501 static int cpuhp_issue_call(int cpu, enum cpuhp_state state, bool bringup,
1502 struct hlist_node *node)
1504 struct cpuhp_step *sp = cpuhp_get_step(state);
1508 * If there's nothing to do, we done.
1509 * Relies on the union for multi_instance.
1511 if ((bringup && !sp->startup.single) ||
1512 (!bringup && !sp->teardown.single))
1515 * The non AP bound callbacks can fail on bringup. On teardown
1516 * e.g. module removal we crash for now.
1519 if (cpuhp_is_ap_state(state))
1520 ret = cpuhp_invoke_ap_callback(cpu, state, bringup, node);
1522 ret = cpuhp_invoke_callback(cpu, state, bringup, node, NULL);
1524 ret = cpuhp_invoke_callback(cpu, state, bringup, node, NULL);
1526 BUG_ON(ret && !bringup);
1531 * Called from __cpuhp_setup_state on a recoverable failure.
1533 * Note: The teardown callbacks for rollback are not allowed to fail!
1535 static void cpuhp_rollback_install(int failedcpu, enum cpuhp_state state,
1536 struct hlist_node *node)
1540 /* Roll back the already executed steps on the other cpus */
1541 for_each_present_cpu(cpu) {
1542 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1543 int cpustate = st->state;
1545 if (cpu >= failedcpu)
1548 /* Did we invoke the startup call on that cpu ? */
1549 if (cpustate >= state)
1550 cpuhp_issue_call(cpu, state, false, node);
1554 int __cpuhp_state_add_instance_cpuslocked(enum cpuhp_state state,
1555 struct hlist_node *node,
1558 struct cpuhp_step *sp;
1562 lockdep_assert_cpus_held();
1564 sp = cpuhp_get_step(state);
1565 if (sp->multi_instance == false)
1568 mutex_lock(&cpuhp_state_mutex);
1570 if (!invoke || !sp->startup.multi)
1574 * Try to call the startup callback for each present cpu
1575 * depending on the hotplug state of the cpu.
1577 for_each_present_cpu(cpu) {
1578 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1579 int cpustate = st->state;
1581 if (cpustate < state)
1584 ret = cpuhp_issue_call(cpu, state, true, node);
1586 if (sp->teardown.multi)
1587 cpuhp_rollback_install(cpu, state, node);
1593 hlist_add_head(node, &sp->list);
1595 mutex_unlock(&cpuhp_state_mutex);
1599 int __cpuhp_state_add_instance(enum cpuhp_state state, struct hlist_node *node,
1605 ret = __cpuhp_state_add_instance_cpuslocked(state, node, invoke);
1609 EXPORT_SYMBOL_GPL(__cpuhp_state_add_instance);
1612 * __cpuhp_setup_state_cpuslocked - Setup the callbacks for an hotplug machine state
1613 * @state: The state to setup
1614 * @invoke: If true, the startup function is invoked for cpus where
1615 * cpu state >= @state
1616 * @startup: startup callback function
1617 * @teardown: teardown callback function
1618 * @multi_instance: State is set up for multiple instances which get
1621 * The caller needs to hold cpus read locked while calling this function.
1624 * Positive state number if @state is CPUHP_AP_ONLINE_DYN
1625 * 0 for all other states
1626 * On failure: proper (negative) error code
1628 int __cpuhp_setup_state_cpuslocked(enum cpuhp_state state,
1629 const char *name, bool invoke,
1630 int (*startup)(unsigned int cpu),
1631 int (*teardown)(unsigned int cpu),
1632 bool multi_instance)
1637 lockdep_assert_cpus_held();
1639 if (cpuhp_cb_check(state) || !name)
1642 mutex_lock(&cpuhp_state_mutex);
1644 ret = cpuhp_store_callbacks(state, name, startup, teardown,
1647 dynstate = state == CPUHP_AP_ONLINE_DYN;
1648 if (ret > 0 && dynstate) {
1653 if (ret || !invoke || !startup)
1657 * Try to call the startup callback for each present cpu
1658 * depending on the hotplug state of the cpu.
1660 for_each_present_cpu(cpu) {
1661 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1662 int cpustate = st->state;
1664 if (cpustate < state)
1667 ret = cpuhp_issue_call(cpu, state, true, NULL);
1670 cpuhp_rollback_install(cpu, state, NULL);
1671 cpuhp_store_callbacks(state, NULL, NULL, NULL, false);
1676 mutex_unlock(&cpuhp_state_mutex);
1678 * If the requested state is CPUHP_AP_ONLINE_DYN, return the
1679 * dynamically allocated state in case of success.
1681 if (!ret && dynstate)
1685 EXPORT_SYMBOL(__cpuhp_setup_state_cpuslocked);
1687 int __cpuhp_setup_state(enum cpuhp_state state,
1688 const char *name, bool invoke,
1689 int (*startup)(unsigned int cpu),
1690 int (*teardown)(unsigned int cpu),
1691 bool multi_instance)
1696 ret = __cpuhp_setup_state_cpuslocked(state, name, invoke, startup,
1697 teardown, multi_instance);
1701 EXPORT_SYMBOL(__cpuhp_setup_state);
1703 int __cpuhp_state_remove_instance(enum cpuhp_state state,
1704 struct hlist_node *node, bool invoke)
1706 struct cpuhp_step *sp = cpuhp_get_step(state);
1709 BUG_ON(cpuhp_cb_check(state));
1711 if (!sp->multi_instance)
1715 mutex_lock(&cpuhp_state_mutex);
1717 if (!invoke || !cpuhp_get_teardown_cb(state))
1720 * Call the teardown callback for each present cpu depending
1721 * on the hotplug state of the cpu. This function is not
1722 * allowed to fail currently!
1724 for_each_present_cpu(cpu) {
1725 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1726 int cpustate = st->state;
1728 if (cpustate >= state)
1729 cpuhp_issue_call(cpu, state, false, node);
1734 mutex_unlock(&cpuhp_state_mutex);
1739 EXPORT_SYMBOL_GPL(__cpuhp_state_remove_instance);
1742 * __cpuhp_remove_state_cpuslocked - Remove the callbacks for an hotplug machine state
1743 * @state: The state to remove
1744 * @invoke: If true, the teardown function is invoked for cpus where
1745 * cpu state >= @state
1747 * The caller needs to hold cpus read locked while calling this function.
1748 * The teardown callback is currently not allowed to fail. Think
1749 * about module removal!
1751 void __cpuhp_remove_state_cpuslocked(enum cpuhp_state state, bool invoke)
1753 struct cpuhp_step *sp = cpuhp_get_step(state);
1756 BUG_ON(cpuhp_cb_check(state));
1758 lockdep_assert_cpus_held();
1760 mutex_lock(&cpuhp_state_mutex);
1761 if (sp->multi_instance) {
1762 WARN(!hlist_empty(&sp->list),
1763 "Error: Removing state %d which has instances left.\n",
1768 if (!invoke || !cpuhp_get_teardown_cb(state))
1772 * Call the teardown callback for each present cpu depending
1773 * on the hotplug state of the cpu. This function is not
1774 * allowed to fail currently!
1776 for_each_present_cpu(cpu) {
1777 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1778 int cpustate = st->state;
1780 if (cpustate >= state)
1781 cpuhp_issue_call(cpu, state, false, NULL);
1784 cpuhp_store_callbacks(state, NULL, NULL, NULL, false);
1785 mutex_unlock(&cpuhp_state_mutex);
1787 EXPORT_SYMBOL(__cpuhp_remove_state_cpuslocked);
1789 void __cpuhp_remove_state(enum cpuhp_state state, bool invoke)
1792 __cpuhp_remove_state_cpuslocked(state, invoke);
1795 EXPORT_SYMBOL(__cpuhp_remove_state);
1797 #if defined(CONFIG_SYSFS) && defined(CONFIG_HOTPLUG_CPU)
1798 static ssize_t show_cpuhp_state(struct device *dev,
1799 struct device_attribute *attr, char *buf)
1801 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id);
1803 return sprintf(buf, "%d\n", st->state);
1805 static DEVICE_ATTR(state, 0444, show_cpuhp_state, NULL);
1807 static ssize_t write_cpuhp_target(struct device *dev,
1808 struct device_attribute *attr,
1809 const char *buf, size_t count)
1811 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id);
1812 struct cpuhp_step *sp;
1815 ret = kstrtoint(buf, 10, &target);
1819 #ifdef CONFIG_CPU_HOTPLUG_STATE_CONTROL
1820 if (target < CPUHP_OFFLINE || target > CPUHP_ONLINE)
1823 if (target != CPUHP_OFFLINE && target != CPUHP_ONLINE)
1827 ret = lock_device_hotplug_sysfs();
1831 mutex_lock(&cpuhp_state_mutex);
1832 sp = cpuhp_get_step(target);
1833 ret = !sp->name || sp->cant_stop ? -EINVAL : 0;
1834 mutex_unlock(&cpuhp_state_mutex);
1838 if (st->state < target)
1839 ret = do_cpu_up(dev->id, target);
1841 ret = do_cpu_down(dev->id, target);
1843 unlock_device_hotplug();
1844 return ret ? ret : count;
1847 static ssize_t show_cpuhp_target(struct device *dev,
1848 struct device_attribute *attr, char *buf)
1850 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id);
1852 return sprintf(buf, "%d\n", st->target);
1854 static DEVICE_ATTR(target, 0644, show_cpuhp_target, write_cpuhp_target);
1857 static ssize_t write_cpuhp_fail(struct device *dev,
1858 struct device_attribute *attr,
1859 const char *buf, size_t count)
1861 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id);
1862 struct cpuhp_step *sp;
1865 ret = kstrtoint(buf, 10, &fail);
1870 * Cannot fail STARTING/DYING callbacks.
1872 if (cpuhp_is_atomic_state(fail))
1876 * Cannot fail anything that doesn't have callbacks.
1878 mutex_lock(&cpuhp_state_mutex);
1879 sp = cpuhp_get_step(fail);
1880 if (!sp->startup.single && !sp->teardown.single)
1882 mutex_unlock(&cpuhp_state_mutex);
1891 static ssize_t show_cpuhp_fail(struct device *dev,
1892 struct device_attribute *attr, char *buf)
1894 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id);
1896 return sprintf(buf, "%d\n", st->fail);
1899 static DEVICE_ATTR(fail, 0644, show_cpuhp_fail, write_cpuhp_fail);
1901 static struct attribute *cpuhp_cpu_attrs[] = {
1902 &dev_attr_state.attr,
1903 &dev_attr_target.attr,
1904 &dev_attr_fail.attr,
1908 static const struct attribute_group cpuhp_cpu_attr_group = {
1909 .attrs = cpuhp_cpu_attrs,
1914 static ssize_t show_cpuhp_states(struct device *dev,
1915 struct device_attribute *attr, char *buf)
1917 ssize_t cur, res = 0;
1920 mutex_lock(&cpuhp_state_mutex);
1921 for (i = CPUHP_OFFLINE; i <= CPUHP_ONLINE; i++) {
1922 struct cpuhp_step *sp = cpuhp_get_step(i);
1925 cur = sprintf(buf, "%3d: %s\n", i, sp->name);
1930 mutex_unlock(&cpuhp_state_mutex);
1933 static DEVICE_ATTR(states, 0444, show_cpuhp_states, NULL);
1935 static struct attribute *cpuhp_cpu_root_attrs[] = {
1936 &dev_attr_states.attr,
1940 static const struct attribute_group cpuhp_cpu_root_attr_group = {
1941 .attrs = cpuhp_cpu_root_attrs,
1946 #ifdef CONFIG_HOTPLUG_SMT
1948 static const char *smt_states[] = {
1949 [CPU_SMT_ENABLED] = "on",
1950 [CPU_SMT_DISABLED] = "off",
1951 [CPU_SMT_FORCE_DISABLED] = "forceoff",
1952 [CPU_SMT_NOT_SUPPORTED] = "notsupported",
1956 show_smt_control(struct device *dev, struct device_attribute *attr, char *buf)
1958 return snprintf(buf, PAGE_SIZE - 2, "%s\n", smt_states[cpu_smt_control]);
1961 static void cpuhp_offline_cpu_device(unsigned int cpu)
1963 struct device *dev = get_cpu_device(cpu);
1965 dev->offline = true;
1966 /* Tell user space about the state change */
1967 kobject_uevent(&dev->kobj, KOBJ_OFFLINE);
1970 static int cpuhp_smt_disable(enum cpuhp_smt_control ctrlval)
1974 cpu_maps_update_begin();
1975 for_each_online_cpu(cpu) {
1976 if (topology_is_primary_thread(cpu))
1978 ret = cpu_down_maps_locked(cpu, CPUHP_OFFLINE);
1982 * As this needs to hold the cpu maps lock it's impossible
1983 * to call device_offline() because that ends up calling
1984 * cpu_down() which takes cpu maps lock. cpu maps lock
1985 * needs to be held as this might race against in kernel
1986 * abusers of the hotplug machinery (thermal management).
1988 * So nothing would update device:offline state. That would
1989 * leave the sysfs entry stale and prevent onlining after
1990 * smt control has been changed to 'off' again. This is
1991 * called under the sysfs hotplug lock, so it is properly
1992 * serialized against the regular offline usage.
1994 cpuhp_offline_cpu_device(cpu);
1997 cpu_smt_control = ctrlval;
1998 cpu_maps_update_done();
2002 static void cpuhp_smt_enable(void)
2004 cpu_maps_update_begin();
2005 cpu_smt_control = CPU_SMT_ENABLED;
2006 cpu_maps_update_done();
2010 store_smt_control(struct device *dev, struct device_attribute *attr,
2011 const char *buf, size_t count)
2015 if (sysfs_streq(buf, "on"))
2016 ctrlval = CPU_SMT_ENABLED;
2017 else if (sysfs_streq(buf, "off"))
2018 ctrlval = CPU_SMT_DISABLED;
2019 else if (sysfs_streq(buf, "forceoff"))
2020 ctrlval = CPU_SMT_FORCE_DISABLED;
2024 if (cpu_smt_control == CPU_SMT_FORCE_DISABLED)
2027 if (cpu_smt_control == CPU_SMT_NOT_SUPPORTED)
2030 ret = lock_device_hotplug_sysfs();
2034 if (ctrlval != cpu_smt_control) {
2036 case CPU_SMT_ENABLED:
2039 case CPU_SMT_DISABLED:
2040 case CPU_SMT_FORCE_DISABLED:
2041 ret = cpuhp_smt_disable(ctrlval);
2046 unlock_device_hotplug();
2047 return ret ? ret : count;
2049 static DEVICE_ATTR(control, 0644, show_smt_control, store_smt_control);
2052 show_smt_active(struct device *dev, struct device_attribute *attr, char *buf)
2054 bool active = topology_max_smt_threads() > 1;
2056 return snprintf(buf, PAGE_SIZE - 2, "%d\n", active);
2058 static DEVICE_ATTR(active, 0444, show_smt_active, NULL);
2060 static struct attribute *cpuhp_smt_attrs[] = {
2061 &dev_attr_control.attr,
2062 &dev_attr_active.attr,
2066 static const struct attribute_group cpuhp_smt_attr_group = {
2067 .attrs = cpuhp_smt_attrs,
2072 static int __init cpu_smt_state_init(void)
2074 if (!topology_smt_supported())
2075 cpu_smt_control = CPU_SMT_NOT_SUPPORTED;
2077 return sysfs_create_group(&cpu_subsys.dev_root->kobj,
2078 &cpuhp_smt_attr_group);
2082 static inline int cpu_smt_state_init(void) { return 0; }
2085 static int __init cpuhp_sysfs_init(void)
2089 ret = cpu_smt_state_init();
2093 ret = sysfs_create_group(&cpu_subsys.dev_root->kobj,
2094 &cpuhp_cpu_root_attr_group);
2098 for_each_possible_cpu(cpu) {
2099 struct device *dev = get_cpu_device(cpu);
2103 ret = sysfs_create_group(&dev->kobj, &cpuhp_cpu_attr_group);
2109 device_initcall(cpuhp_sysfs_init);
2113 * cpu_bit_bitmap[] is a special, "compressed" data structure that
2114 * represents all NR_CPUS bits binary values of 1<<nr.
2116 * It is used by cpumask_of() to get a constant address to a CPU
2117 * mask value that has a single bit set only.
2120 /* cpu_bit_bitmap[0] is empty - so we can back into it */
2121 #define MASK_DECLARE_1(x) [x+1][0] = (1UL << (x))
2122 #define MASK_DECLARE_2(x) MASK_DECLARE_1(x), MASK_DECLARE_1(x+1)
2123 #define MASK_DECLARE_4(x) MASK_DECLARE_2(x), MASK_DECLARE_2(x+2)
2124 #define MASK_DECLARE_8(x) MASK_DECLARE_4(x), MASK_DECLARE_4(x+4)
2126 const unsigned long cpu_bit_bitmap[BITS_PER_LONG+1][BITS_TO_LONGS(NR_CPUS)] = {
2128 MASK_DECLARE_8(0), MASK_DECLARE_8(8),
2129 MASK_DECLARE_8(16), MASK_DECLARE_8(24),
2130 #if BITS_PER_LONG > 32
2131 MASK_DECLARE_8(32), MASK_DECLARE_8(40),
2132 MASK_DECLARE_8(48), MASK_DECLARE_8(56),
2135 EXPORT_SYMBOL_GPL(cpu_bit_bitmap);
2137 const DECLARE_BITMAP(cpu_all_bits, NR_CPUS) = CPU_BITS_ALL;
2138 EXPORT_SYMBOL(cpu_all_bits);
2140 #ifdef CONFIG_INIT_ALL_POSSIBLE
2141 struct cpumask __cpu_possible_mask __read_mostly
2144 struct cpumask __cpu_possible_mask __read_mostly;
2146 EXPORT_SYMBOL(__cpu_possible_mask);
2148 struct cpumask __cpu_online_mask __read_mostly;
2149 EXPORT_SYMBOL(__cpu_online_mask);
2151 struct cpumask __cpu_present_mask __read_mostly;
2152 EXPORT_SYMBOL(__cpu_present_mask);
2154 struct cpumask __cpu_active_mask __read_mostly;
2155 EXPORT_SYMBOL(__cpu_active_mask);
2157 void init_cpu_present(const struct cpumask *src)
2159 cpumask_copy(&__cpu_present_mask, src);
2162 void init_cpu_possible(const struct cpumask *src)
2164 cpumask_copy(&__cpu_possible_mask, src);
2167 void init_cpu_online(const struct cpumask *src)
2169 cpumask_copy(&__cpu_online_mask, src);
2173 * Activate the first processor.
2175 void __init boot_cpu_init(void)
2177 int cpu = smp_processor_id();
2179 /* Mark the boot cpu "present", "online" etc for SMP and UP case */
2180 set_cpu_online(cpu, true);
2181 set_cpu_active(cpu, true);
2182 set_cpu_present(cpu, true);
2183 set_cpu_possible(cpu, true);
2186 __boot_cpu_id = cpu;
2191 * Must be called _AFTER_ setting up the per_cpu areas
2193 void __init boot_cpu_hotplug_init(void)
2195 per_cpu_ptr(&cpuhp_state, smp_processor_id())->state = CPUHP_ONLINE;