sh_eth: fix EESIPR values for SH77{34|63}
[linux/fpc-iii.git] / kernel / cpu.c
blobf75c4d031eeb2152c75182466dbf5ec97e93c008
1 /* CPU control.
2 * (C) 2001, 2002, 2003, 2004 Rusty Russell
4 * This code is licenced under the GPL.
5 */
6 #include <linux/proc_fs.h>
7 #include <linux/smp.h>
8 #include <linux/init.h>
9 #include <linux/notifier.h>
10 #include <linux/sched.h>
11 #include <linux/unistd.h>
12 #include <linux/cpu.h>
13 #include <linux/oom.h>
14 #include <linux/rcupdate.h>
15 #include <linux/export.h>
16 #include <linux/bug.h>
17 #include <linux/kthread.h>
18 #include <linux/stop_machine.h>
19 #include <linux/mutex.h>
20 #include <linux/gfp.h>
21 #include <linux/suspend.h>
22 #include <linux/lockdep.h>
23 #include <linux/tick.h>
24 #include <linux/irq.h>
25 #include <linux/smpboot.h>
26 #include <linux/relay.h>
27 #include <linux/slab.h>
29 #include <trace/events/power.h>
30 #define CREATE_TRACE_POINTS
31 #include <trace/events/cpuhp.h>
33 #include "smpboot.h"
35 /**
36 * cpuhp_cpu_state - Per cpu hotplug state storage
37 * @state: The current cpu state
38 * @target: The target state
39 * @thread: Pointer to the hotplug thread
40 * @should_run: Thread should execute
41 * @rollback: Perform a rollback
42 * @single: Single callback invocation
43 * @bringup: Single callback bringup or teardown selector
44 * @cb_state: The state for a single callback (install/uninstall)
45 * @result: Result of the operation
46 * @done: Signal completion to the issuer of the task
48 struct cpuhp_cpu_state {
49 enum cpuhp_state state;
50 enum cpuhp_state target;
51 #ifdef CONFIG_SMP
52 struct task_struct *thread;
53 bool should_run;
54 bool rollback;
55 bool single;
56 bool bringup;
57 struct hlist_node *node;
58 enum cpuhp_state cb_state;
59 int result;
60 struct completion done;
61 #endif
64 static DEFINE_PER_CPU(struct cpuhp_cpu_state, cpuhp_state);
66 /**
67 * cpuhp_step - Hotplug state machine step
68 * @name: Name of the step
69 * @startup: Startup function of the step
70 * @teardown: Teardown function of the step
71 * @skip_onerr: Do not invoke the functions on error rollback
72 * Will go away once the notifiers are gone
73 * @cant_stop: Bringup/teardown can't be stopped at this step
75 struct cpuhp_step {
76 const char *name;
77 union {
78 int (*single)(unsigned int cpu);
79 int (*multi)(unsigned int cpu,
80 struct hlist_node *node);
81 } startup;
82 union {
83 int (*single)(unsigned int cpu);
84 int (*multi)(unsigned int cpu,
85 struct hlist_node *node);
86 } teardown;
87 struct hlist_head list;
88 bool skip_onerr;
89 bool cant_stop;
90 bool multi_instance;
93 static DEFINE_MUTEX(cpuhp_state_mutex);
94 static struct cpuhp_step cpuhp_bp_states[];
95 static struct cpuhp_step cpuhp_ap_states[];
97 static bool cpuhp_is_ap_state(enum cpuhp_state state)
100 * The extra check for CPUHP_TEARDOWN_CPU is only for documentation
101 * purposes as that state is handled explicitly in cpu_down.
103 return state > CPUHP_BRINGUP_CPU && state != CPUHP_TEARDOWN_CPU;
106 static struct cpuhp_step *cpuhp_get_step(enum cpuhp_state state)
108 struct cpuhp_step *sp;
110 sp = cpuhp_is_ap_state(state) ? cpuhp_ap_states : cpuhp_bp_states;
111 return sp + state;
115 * cpuhp_invoke_callback _ Invoke the callbacks for a given state
116 * @cpu: The cpu for which the callback should be invoked
117 * @step: The step in the state machine
118 * @bringup: True if the bringup callback should be invoked
120 * Called from cpu hotplug and from the state register machinery.
122 static int cpuhp_invoke_callback(unsigned int cpu, enum cpuhp_state state,
123 bool bringup, struct hlist_node *node)
125 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
126 struct cpuhp_step *step = cpuhp_get_step(state);
127 int (*cbm)(unsigned int cpu, struct hlist_node *node);
128 int (*cb)(unsigned int cpu);
129 int ret, cnt;
131 if (!step->multi_instance) {
132 cb = bringup ? step->startup.single : step->teardown.single;
133 if (!cb)
134 return 0;
135 trace_cpuhp_enter(cpu, st->target, state, cb);
136 ret = cb(cpu);
137 trace_cpuhp_exit(cpu, st->state, state, ret);
138 return ret;
140 cbm = bringup ? step->startup.multi : step->teardown.multi;
141 if (!cbm)
142 return 0;
144 /* Single invocation for instance add/remove */
145 if (node) {
146 trace_cpuhp_multi_enter(cpu, st->target, state, cbm, node);
147 ret = cbm(cpu, node);
148 trace_cpuhp_exit(cpu, st->state, state, ret);
149 return ret;
152 /* State transition. Invoke on all instances */
153 cnt = 0;
154 hlist_for_each(node, &step->list) {
155 trace_cpuhp_multi_enter(cpu, st->target, state, cbm, node);
156 ret = cbm(cpu, node);
157 trace_cpuhp_exit(cpu, st->state, state, ret);
158 if (ret)
159 goto err;
160 cnt++;
162 return 0;
163 err:
164 /* Rollback the instances if one failed */
165 cbm = !bringup ? step->startup.multi : step->teardown.multi;
166 if (!cbm)
167 return ret;
169 hlist_for_each(node, &step->list) {
170 if (!cnt--)
171 break;
172 cbm(cpu, node);
174 return ret;
177 #ifdef CONFIG_SMP
178 /* Serializes the updates to cpu_online_mask, cpu_present_mask */
179 static DEFINE_MUTEX(cpu_add_remove_lock);
180 bool cpuhp_tasks_frozen;
181 EXPORT_SYMBOL_GPL(cpuhp_tasks_frozen);
184 * The following two APIs (cpu_maps_update_begin/done) must be used when
185 * attempting to serialize the updates to cpu_online_mask & cpu_present_mask.
187 void cpu_maps_update_begin(void)
189 mutex_lock(&cpu_add_remove_lock);
192 void cpu_maps_update_done(void)
194 mutex_unlock(&cpu_add_remove_lock);
197 /* If set, cpu_up and cpu_down will return -EBUSY and do nothing.
198 * Should always be manipulated under cpu_add_remove_lock
200 static int cpu_hotplug_disabled;
202 #ifdef CONFIG_HOTPLUG_CPU
204 static struct {
205 struct task_struct *active_writer;
206 /* wait queue to wake up the active_writer */
207 wait_queue_head_t wq;
208 /* verifies that no writer will get active while readers are active */
209 struct mutex lock;
211 * Also blocks the new readers during
212 * an ongoing cpu hotplug operation.
214 atomic_t refcount;
216 #ifdef CONFIG_DEBUG_LOCK_ALLOC
217 struct lockdep_map dep_map;
218 #endif
219 } cpu_hotplug = {
220 .active_writer = NULL,
221 .wq = __WAIT_QUEUE_HEAD_INITIALIZER(cpu_hotplug.wq),
222 .lock = __MUTEX_INITIALIZER(cpu_hotplug.lock),
223 #ifdef CONFIG_DEBUG_LOCK_ALLOC
224 .dep_map = STATIC_LOCKDEP_MAP_INIT("cpu_hotplug.dep_map", &cpu_hotplug.dep_map),
225 #endif
228 /* Lockdep annotations for get/put_online_cpus() and cpu_hotplug_begin/end() */
229 #define cpuhp_lock_acquire_read() lock_map_acquire_read(&cpu_hotplug.dep_map)
230 #define cpuhp_lock_acquire_tryread() \
231 lock_map_acquire_tryread(&cpu_hotplug.dep_map)
232 #define cpuhp_lock_acquire() lock_map_acquire(&cpu_hotplug.dep_map)
233 #define cpuhp_lock_release() lock_map_release(&cpu_hotplug.dep_map)
236 void get_online_cpus(void)
238 might_sleep();
239 if (cpu_hotplug.active_writer == current)
240 return;
241 cpuhp_lock_acquire_read();
242 mutex_lock(&cpu_hotplug.lock);
243 atomic_inc(&cpu_hotplug.refcount);
244 mutex_unlock(&cpu_hotplug.lock);
246 EXPORT_SYMBOL_GPL(get_online_cpus);
248 void put_online_cpus(void)
250 int refcount;
252 if (cpu_hotplug.active_writer == current)
253 return;
255 refcount = atomic_dec_return(&cpu_hotplug.refcount);
256 if (WARN_ON(refcount < 0)) /* try to fix things up */
257 atomic_inc(&cpu_hotplug.refcount);
259 if (refcount <= 0 && waitqueue_active(&cpu_hotplug.wq))
260 wake_up(&cpu_hotplug.wq);
262 cpuhp_lock_release();
265 EXPORT_SYMBOL_GPL(put_online_cpus);
268 * This ensures that the hotplug operation can begin only when the
269 * refcount goes to zero.
271 * Note that during a cpu-hotplug operation, the new readers, if any,
272 * will be blocked by the cpu_hotplug.lock
274 * Since cpu_hotplug_begin() is always called after invoking
275 * cpu_maps_update_begin(), we can be sure that only one writer is active.
277 * Note that theoretically, there is a possibility of a livelock:
278 * - Refcount goes to zero, last reader wakes up the sleeping
279 * writer.
280 * - Last reader unlocks the cpu_hotplug.lock.
281 * - A new reader arrives at this moment, bumps up the refcount.
282 * - The writer acquires the cpu_hotplug.lock finds the refcount
283 * non zero and goes to sleep again.
285 * However, this is very difficult to achieve in practice since
286 * get_online_cpus() not an api which is called all that often.
289 void cpu_hotplug_begin(void)
291 DEFINE_WAIT(wait);
293 cpu_hotplug.active_writer = current;
294 cpuhp_lock_acquire();
296 for (;;) {
297 mutex_lock(&cpu_hotplug.lock);
298 prepare_to_wait(&cpu_hotplug.wq, &wait, TASK_UNINTERRUPTIBLE);
299 if (likely(!atomic_read(&cpu_hotplug.refcount)))
300 break;
301 mutex_unlock(&cpu_hotplug.lock);
302 schedule();
304 finish_wait(&cpu_hotplug.wq, &wait);
307 void cpu_hotplug_done(void)
309 cpu_hotplug.active_writer = NULL;
310 mutex_unlock(&cpu_hotplug.lock);
311 cpuhp_lock_release();
315 * Wait for currently running CPU hotplug operations to complete (if any) and
316 * disable future CPU hotplug (from sysfs). The 'cpu_add_remove_lock' protects
317 * the 'cpu_hotplug_disabled' flag. The same lock is also acquired by the
318 * hotplug path before performing hotplug operations. So acquiring that lock
319 * guarantees mutual exclusion from any currently running hotplug operations.
321 void cpu_hotplug_disable(void)
323 cpu_maps_update_begin();
324 cpu_hotplug_disabled++;
325 cpu_maps_update_done();
327 EXPORT_SYMBOL_GPL(cpu_hotplug_disable);
329 static void __cpu_hotplug_enable(void)
331 if (WARN_ONCE(!cpu_hotplug_disabled, "Unbalanced cpu hotplug enable\n"))
332 return;
333 cpu_hotplug_disabled--;
336 void cpu_hotplug_enable(void)
338 cpu_maps_update_begin();
339 __cpu_hotplug_enable();
340 cpu_maps_update_done();
342 EXPORT_SYMBOL_GPL(cpu_hotplug_enable);
343 #endif /* CONFIG_HOTPLUG_CPU */
345 /* Notifier wrappers for transitioning to state machine */
347 static int bringup_wait_for_ap(unsigned int cpu)
349 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
351 wait_for_completion(&st->done);
352 return st->result;
355 static int bringup_cpu(unsigned int cpu)
357 struct task_struct *idle = idle_thread_get(cpu);
358 int ret;
361 * Some architectures have to walk the irq descriptors to
362 * setup the vector space for the cpu which comes online.
363 * Prevent irq alloc/free across the bringup.
365 irq_lock_sparse();
367 /* Arch-specific enabling code. */
368 ret = __cpu_up(cpu, idle);
369 irq_unlock_sparse();
370 if (ret)
371 return ret;
372 ret = bringup_wait_for_ap(cpu);
373 BUG_ON(!cpu_online(cpu));
374 return ret;
378 * Hotplug state machine related functions
380 static void undo_cpu_down(unsigned int cpu, struct cpuhp_cpu_state *st)
382 for (st->state++; st->state < st->target; st->state++) {
383 struct cpuhp_step *step = cpuhp_get_step(st->state);
385 if (!step->skip_onerr)
386 cpuhp_invoke_callback(cpu, st->state, true, NULL);
390 static int cpuhp_down_callbacks(unsigned int cpu, struct cpuhp_cpu_state *st,
391 enum cpuhp_state target)
393 enum cpuhp_state prev_state = st->state;
394 int ret = 0;
396 for (; st->state > target; st->state--) {
397 ret = cpuhp_invoke_callback(cpu, st->state, false, NULL);
398 if (ret) {
399 st->target = prev_state;
400 undo_cpu_down(cpu, st);
401 break;
404 return ret;
407 static void undo_cpu_up(unsigned int cpu, struct cpuhp_cpu_state *st)
409 for (st->state--; st->state > st->target; st->state--) {
410 struct cpuhp_step *step = cpuhp_get_step(st->state);
412 if (!step->skip_onerr)
413 cpuhp_invoke_callback(cpu, st->state, false, NULL);
417 static int cpuhp_up_callbacks(unsigned int cpu, struct cpuhp_cpu_state *st,
418 enum cpuhp_state target)
420 enum cpuhp_state prev_state = st->state;
421 int ret = 0;
423 while (st->state < target) {
424 st->state++;
425 ret = cpuhp_invoke_callback(cpu, st->state, true, NULL);
426 if (ret) {
427 st->target = prev_state;
428 undo_cpu_up(cpu, st);
429 break;
432 return ret;
436 * The cpu hotplug threads manage the bringup and teardown of the cpus
438 static void cpuhp_create(unsigned int cpu)
440 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
442 init_completion(&st->done);
445 static int cpuhp_should_run(unsigned int cpu)
447 struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
449 return st->should_run;
452 /* Execute the teardown callbacks. Used to be CPU_DOWN_PREPARE */
453 static int cpuhp_ap_offline(unsigned int cpu, struct cpuhp_cpu_state *st)
455 enum cpuhp_state target = max((int)st->target, CPUHP_TEARDOWN_CPU);
457 return cpuhp_down_callbacks(cpu, st, target);
460 /* Execute the online startup callbacks. Used to be CPU_ONLINE */
461 static int cpuhp_ap_online(unsigned int cpu, struct cpuhp_cpu_state *st)
463 return cpuhp_up_callbacks(cpu, st, st->target);
467 * Execute teardown/startup callbacks on the plugged cpu. Also used to invoke
468 * callbacks when a state gets [un]installed at runtime.
470 static void cpuhp_thread_fun(unsigned int cpu)
472 struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
473 int ret = 0;
476 * Paired with the mb() in cpuhp_kick_ap_work and
477 * cpuhp_invoke_ap_callback, so the work set is consistent visible.
479 smp_mb();
480 if (!st->should_run)
481 return;
483 st->should_run = false;
485 /* Single callback invocation for [un]install ? */
486 if (st->single) {
487 if (st->cb_state < CPUHP_AP_ONLINE) {
488 local_irq_disable();
489 ret = cpuhp_invoke_callback(cpu, st->cb_state,
490 st->bringup, st->node);
491 local_irq_enable();
492 } else {
493 ret = cpuhp_invoke_callback(cpu, st->cb_state,
494 st->bringup, st->node);
496 } else if (st->rollback) {
497 BUG_ON(st->state < CPUHP_AP_ONLINE_IDLE);
499 undo_cpu_down(cpu, st);
500 st->rollback = false;
501 } else {
502 /* Cannot happen .... */
503 BUG_ON(st->state < CPUHP_AP_ONLINE_IDLE);
505 /* Regular hotplug work */
506 if (st->state < st->target)
507 ret = cpuhp_ap_online(cpu, st);
508 else if (st->state > st->target)
509 ret = cpuhp_ap_offline(cpu, st);
511 st->result = ret;
512 complete(&st->done);
515 /* Invoke a single callback on a remote cpu */
516 static int
517 cpuhp_invoke_ap_callback(int cpu, enum cpuhp_state state, bool bringup,
518 struct hlist_node *node)
520 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
522 if (!cpu_online(cpu))
523 return 0;
526 * If we are up and running, use the hotplug thread. For early calls
527 * we invoke the thread function directly.
529 if (!st->thread)
530 return cpuhp_invoke_callback(cpu, state, bringup, node);
532 st->cb_state = state;
533 st->single = true;
534 st->bringup = bringup;
535 st->node = node;
538 * Make sure the above stores are visible before should_run becomes
539 * true. Paired with the mb() above in cpuhp_thread_fun()
541 smp_mb();
542 st->should_run = true;
543 wake_up_process(st->thread);
544 wait_for_completion(&st->done);
545 return st->result;
548 /* Regular hotplug invocation of the AP hotplug thread */
549 static void __cpuhp_kick_ap_work(struct cpuhp_cpu_state *st)
551 st->result = 0;
552 st->single = false;
554 * Make sure the above stores are visible before should_run becomes
555 * true. Paired with the mb() above in cpuhp_thread_fun()
557 smp_mb();
558 st->should_run = true;
559 wake_up_process(st->thread);
562 static int cpuhp_kick_ap_work(unsigned int cpu)
564 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
565 enum cpuhp_state state = st->state;
567 trace_cpuhp_enter(cpu, st->target, state, cpuhp_kick_ap_work);
568 __cpuhp_kick_ap_work(st);
569 wait_for_completion(&st->done);
570 trace_cpuhp_exit(cpu, st->state, state, st->result);
571 return st->result;
574 static struct smp_hotplug_thread cpuhp_threads = {
575 .store = &cpuhp_state.thread,
576 .create = &cpuhp_create,
577 .thread_should_run = cpuhp_should_run,
578 .thread_fn = cpuhp_thread_fun,
579 .thread_comm = "cpuhp/%u",
580 .selfparking = true,
583 void __init cpuhp_threads_init(void)
585 BUG_ON(smpboot_register_percpu_thread(&cpuhp_threads));
586 kthread_unpark(this_cpu_read(cpuhp_state.thread));
589 #ifdef CONFIG_HOTPLUG_CPU
591 * clear_tasks_mm_cpumask - Safely clear tasks' mm_cpumask for a CPU
592 * @cpu: a CPU id
594 * This function walks all processes, finds a valid mm struct for each one and
595 * then clears a corresponding bit in mm's cpumask. While this all sounds
596 * trivial, there are various non-obvious corner cases, which this function
597 * tries to solve in a safe manner.
599 * Also note that the function uses a somewhat relaxed locking scheme, so it may
600 * be called only for an already offlined CPU.
602 void clear_tasks_mm_cpumask(int cpu)
604 struct task_struct *p;
607 * This function is called after the cpu is taken down and marked
608 * offline, so its not like new tasks will ever get this cpu set in
609 * their mm mask. -- Peter Zijlstra
610 * Thus, we may use rcu_read_lock() here, instead of grabbing
611 * full-fledged tasklist_lock.
613 WARN_ON(cpu_online(cpu));
614 rcu_read_lock();
615 for_each_process(p) {
616 struct task_struct *t;
619 * Main thread might exit, but other threads may still have
620 * a valid mm. Find one.
622 t = find_lock_task_mm(p);
623 if (!t)
624 continue;
625 cpumask_clear_cpu(cpu, mm_cpumask(t->mm));
626 task_unlock(t);
628 rcu_read_unlock();
631 static inline void check_for_tasks(int dead_cpu)
633 struct task_struct *g, *p;
635 read_lock(&tasklist_lock);
636 for_each_process_thread(g, p) {
637 if (!p->on_rq)
638 continue;
640 * We do the check with unlocked task_rq(p)->lock.
641 * Order the reading to do not warn about a task,
642 * which was running on this cpu in the past, and
643 * it's just been woken on another cpu.
645 rmb();
646 if (task_cpu(p) != dead_cpu)
647 continue;
649 pr_warn("Task %s (pid=%d) is on cpu %d (state=%ld, flags=%x)\n",
650 p->comm, task_pid_nr(p), dead_cpu, p->state, p->flags);
652 read_unlock(&tasklist_lock);
655 /* Take this CPU down. */
656 static int take_cpu_down(void *_param)
658 struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
659 enum cpuhp_state target = max((int)st->target, CPUHP_AP_OFFLINE);
660 int err, cpu = smp_processor_id();
662 /* Ensure this CPU doesn't handle any more interrupts. */
663 err = __cpu_disable();
664 if (err < 0)
665 return err;
668 * We get here while we are in CPUHP_TEARDOWN_CPU state and we must not
669 * do this step again.
671 WARN_ON(st->state != CPUHP_TEARDOWN_CPU);
672 st->state--;
673 /* Invoke the former CPU_DYING callbacks */
674 for (; st->state > target; st->state--)
675 cpuhp_invoke_callback(cpu, st->state, false, NULL);
677 /* Give up timekeeping duties */
678 tick_handover_do_timer();
679 /* Park the stopper thread */
680 stop_machine_park(cpu);
681 return 0;
684 static int takedown_cpu(unsigned int cpu)
686 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
687 int err;
689 /* Park the smpboot threads */
690 kthread_park(per_cpu_ptr(&cpuhp_state, cpu)->thread);
691 smpboot_park_threads(cpu);
694 * Prevent irq alloc/free while the dying cpu reorganizes the
695 * interrupt affinities.
697 irq_lock_sparse();
700 * So now all preempt/rcu users must observe !cpu_active().
702 err = stop_machine(take_cpu_down, NULL, cpumask_of(cpu));
703 if (err) {
704 /* CPU refused to die */
705 irq_unlock_sparse();
706 /* Unpark the hotplug thread so we can rollback there */
707 kthread_unpark(per_cpu_ptr(&cpuhp_state, cpu)->thread);
708 return err;
710 BUG_ON(cpu_online(cpu));
713 * The CPUHP_AP_SCHED_MIGRATE_DYING callback will have removed all
714 * runnable tasks from the cpu, there's only the idle task left now
715 * that the migration thread is done doing the stop_machine thing.
717 * Wait for the stop thread to go away.
719 wait_for_completion(&st->done);
720 BUG_ON(st->state != CPUHP_AP_IDLE_DEAD);
722 /* Interrupts are moved away from the dying cpu, reenable alloc/free */
723 irq_unlock_sparse();
725 hotplug_cpu__broadcast_tick_pull(cpu);
726 /* This actually kills the CPU. */
727 __cpu_die(cpu);
729 tick_cleanup_dead_cpu(cpu);
730 return 0;
733 static void cpuhp_complete_idle_dead(void *arg)
735 struct cpuhp_cpu_state *st = arg;
737 complete(&st->done);
740 void cpuhp_report_idle_dead(void)
742 struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
744 BUG_ON(st->state != CPUHP_AP_OFFLINE);
745 rcu_report_dead(smp_processor_id());
746 st->state = CPUHP_AP_IDLE_DEAD;
748 * We cannot call complete after rcu_report_dead() so we delegate it
749 * to an online cpu.
751 smp_call_function_single(cpumask_first(cpu_online_mask),
752 cpuhp_complete_idle_dead, st, 0);
755 #else
756 #define takedown_cpu NULL
757 #endif
759 #ifdef CONFIG_HOTPLUG_CPU
761 /* Requires cpu_add_remove_lock to be held */
762 static int __ref _cpu_down(unsigned int cpu, int tasks_frozen,
763 enum cpuhp_state target)
765 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
766 int prev_state, ret = 0;
767 bool hasdied = false;
769 if (num_online_cpus() == 1)
770 return -EBUSY;
772 if (!cpu_present(cpu))
773 return -EINVAL;
775 cpu_hotplug_begin();
777 cpuhp_tasks_frozen = tasks_frozen;
779 prev_state = st->state;
780 st->target = target;
782 * If the current CPU state is in the range of the AP hotplug thread,
783 * then we need to kick the thread.
785 if (st->state > CPUHP_TEARDOWN_CPU) {
786 ret = cpuhp_kick_ap_work(cpu);
788 * The AP side has done the error rollback already. Just
789 * return the error code..
791 if (ret)
792 goto out;
795 * We might have stopped still in the range of the AP hotplug
796 * thread. Nothing to do anymore.
798 if (st->state > CPUHP_TEARDOWN_CPU)
799 goto out;
802 * The AP brought itself down to CPUHP_TEARDOWN_CPU. So we need
803 * to do the further cleanups.
805 ret = cpuhp_down_callbacks(cpu, st, target);
806 if (ret && st->state > CPUHP_TEARDOWN_CPU && st->state < prev_state) {
807 st->target = prev_state;
808 st->rollback = true;
809 cpuhp_kick_ap_work(cpu);
812 hasdied = prev_state != st->state && st->state == CPUHP_OFFLINE;
813 out:
814 cpu_hotplug_done();
815 return ret;
818 static int do_cpu_down(unsigned int cpu, enum cpuhp_state target)
820 int err;
822 cpu_maps_update_begin();
824 if (cpu_hotplug_disabled) {
825 err = -EBUSY;
826 goto out;
829 err = _cpu_down(cpu, 0, target);
831 out:
832 cpu_maps_update_done();
833 return err;
835 int cpu_down(unsigned int cpu)
837 return do_cpu_down(cpu, CPUHP_OFFLINE);
839 EXPORT_SYMBOL(cpu_down);
840 #endif /*CONFIG_HOTPLUG_CPU*/
843 * notify_cpu_starting(cpu) - Invoke the callbacks on the starting CPU
844 * @cpu: cpu that just started
846 * It must be called by the arch code on the new cpu, before the new cpu
847 * enables interrupts and before the "boot" cpu returns from __cpu_up().
849 void notify_cpu_starting(unsigned int cpu)
851 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
852 enum cpuhp_state target = min((int)st->target, CPUHP_AP_ONLINE);
854 rcu_cpu_starting(cpu); /* Enables RCU usage on this CPU. */
855 while (st->state < target) {
856 st->state++;
857 cpuhp_invoke_callback(cpu, st->state, true, NULL);
862 * Called from the idle task. We need to set active here, so we can kick off
863 * the stopper thread and unpark the smpboot threads. If the target state is
864 * beyond CPUHP_AP_ONLINE_IDLE we kick cpuhp thread and let it bring up the
865 * cpu further.
867 void cpuhp_online_idle(enum cpuhp_state state)
869 struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
870 unsigned int cpu = smp_processor_id();
872 /* Happens for the boot cpu */
873 if (state != CPUHP_AP_ONLINE_IDLE)
874 return;
876 st->state = CPUHP_AP_ONLINE_IDLE;
878 /* Unpark the stopper thread and the hotplug thread of this cpu */
879 stop_machine_unpark(cpu);
880 kthread_unpark(st->thread);
882 /* Should we go further up ? */
883 if (st->target > CPUHP_AP_ONLINE_IDLE)
884 __cpuhp_kick_ap_work(st);
885 else
886 complete(&st->done);
889 /* Requires cpu_add_remove_lock to be held */
890 static int _cpu_up(unsigned int cpu, int tasks_frozen, enum cpuhp_state target)
892 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
893 struct task_struct *idle;
894 int ret = 0;
896 cpu_hotplug_begin();
898 if (!cpu_present(cpu)) {
899 ret = -EINVAL;
900 goto out;
904 * The caller of do_cpu_up might have raced with another
905 * caller. Ignore it for now.
907 if (st->state >= target)
908 goto out;
910 if (st->state == CPUHP_OFFLINE) {
911 /* Let it fail before we try to bring the cpu up */
912 idle = idle_thread_get(cpu);
913 if (IS_ERR(idle)) {
914 ret = PTR_ERR(idle);
915 goto out;
919 cpuhp_tasks_frozen = tasks_frozen;
921 st->target = target;
923 * If the current CPU state is in the range of the AP hotplug thread,
924 * then we need to kick the thread once more.
926 if (st->state > CPUHP_BRINGUP_CPU) {
927 ret = cpuhp_kick_ap_work(cpu);
929 * The AP side has done the error rollback already. Just
930 * return the error code..
932 if (ret)
933 goto out;
937 * Try to reach the target state. We max out on the BP at
938 * CPUHP_BRINGUP_CPU. After that the AP hotplug thread is
939 * responsible for bringing it up to the target state.
941 target = min((int)target, CPUHP_BRINGUP_CPU);
942 ret = cpuhp_up_callbacks(cpu, st, target);
943 out:
944 cpu_hotplug_done();
945 return ret;
948 static int do_cpu_up(unsigned int cpu, enum cpuhp_state target)
950 int err = 0;
952 if (!cpu_possible(cpu)) {
953 pr_err("can't online cpu %d because it is not configured as may-hotadd at boot time\n",
954 cpu);
955 #if defined(CONFIG_IA64)
956 pr_err("please check additional_cpus= boot parameter\n");
957 #endif
958 return -EINVAL;
961 err = try_online_node(cpu_to_node(cpu));
962 if (err)
963 return err;
965 cpu_maps_update_begin();
967 if (cpu_hotplug_disabled) {
968 err = -EBUSY;
969 goto out;
972 err = _cpu_up(cpu, 0, target);
973 out:
974 cpu_maps_update_done();
975 return err;
978 int cpu_up(unsigned int cpu)
980 return do_cpu_up(cpu, CPUHP_ONLINE);
982 EXPORT_SYMBOL_GPL(cpu_up);
984 #ifdef CONFIG_PM_SLEEP_SMP
985 static cpumask_var_t frozen_cpus;
987 int freeze_secondary_cpus(int primary)
989 int cpu, error = 0;
991 cpu_maps_update_begin();
992 if (!cpu_online(primary))
993 primary = cpumask_first(cpu_online_mask);
995 * We take down all of the non-boot CPUs in one shot to avoid races
996 * with the userspace trying to use the CPU hotplug at the same time
998 cpumask_clear(frozen_cpus);
1000 pr_info("Disabling non-boot CPUs ...\n");
1001 for_each_online_cpu(cpu) {
1002 if (cpu == primary)
1003 continue;
1004 trace_suspend_resume(TPS("CPU_OFF"), cpu, true);
1005 error = _cpu_down(cpu, 1, CPUHP_OFFLINE);
1006 trace_suspend_resume(TPS("CPU_OFF"), cpu, false);
1007 if (!error)
1008 cpumask_set_cpu(cpu, frozen_cpus);
1009 else {
1010 pr_err("Error taking CPU%d down: %d\n", cpu, error);
1011 break;
1015 if (!error)
1016 BUG_ON(num_online_cpus() > 1);
1017 else
1018 pr_err("Non-boot CPUs are not disabled\n");
1021 * Make sure the CPUs won't be enabled by someone else. We need to do
1022 * this even in case of failure as all disable_nonboot_cpus() users are
1023 * supposed to do enable_nonboot_cpus() on the failure path.
1025 cpu_hotplug_disabled++;
1027 cpu_maps_update_done();
1028 return error;
1031 void __weak arch_enable_nonboot_cpus_begin(void)
1035 void __weak arch_enable_nonboot_cpus_end(void)
1039 void enable_nonboot_cpus(void)
1041 int cpu, error;
1043 /* Allow everyone to use the CPU hotplug again */
1044 cpu_maps_update_begin();
1045 __cpu_hotplug_enable();
1046 if (cpumask_empty(frozen_cpus))
1047 goto out;
1049 pr_info("Enabling non-boot CPUs ...\n");
1051 arch_enable_nonboot_cpus_begin();
1053 for_each_cpu(cpu, frozen_cpus) {
1054 trace_suspend_resume(TPS("CPU_ON"), cpu, true);
1055 error = _cpu_up(cpu, 1, CPUHP_ONLINE);
1056 trace_suspend_resume(TPS("CPU_ON"), cpu, false);
1057 if (!error) {
1058 pr_info("CPU%d is up\n", cpu);
1059 continue;
1061 pr_warn("Error taking CPU%d up: %d\n", cpu, error);
1064 arch_enable_nonboot_cpus_end();
1066 cpumask_clear(frozen_cpus);
1067 out:
1068 cpu_maps_update_done();
1071 static int __init alloc_frozen_cpus(void)
1073 if (!alloc_cpumask_var(&frozen_cpus, GFP_KERNEL|__GFP_ZERO))
1074 return -ENOMEM;
1075 return 0;
1077 core_initcall(alloc_frozen_cpus);
1080 * When callbacks for CPU hotplug notifications are being executed, we must
1081 * ensure that the state of the system with respect to the tasks being frozen
1082 * or not, as reported by the notification, remains unchanged *throughout the
1083 * duration* of the execution of the callbacks.
1084 * Hence we need to prevent the freezer from racing with regular CPU hotplug.
1086 * This synchronization is implemented by mutually excluding regular CPU
1087 * hotplug and Suspend/Hibernate call paths by hooking onto the Suspend/
1088 * Hibernate notifications.
1090 static int
1091 cpu_hotplug_pm_callback(struct notifier_block *nb,
1092 unsigned long action, void *ptr)
1094 switch (action) {
1096 case PM_SUSPEND_PREPARE:
1097 case PM_HIBERNATION_PREPARE:
1098 cpu_hotplug_disable();
1099 break;
1101 case PM_POST_SUSPEND:
1102 case PM_POST_HIBERNATION:
1103 cpu_hotplug_enable();
1104 break;
1106 default:
1107 return NOTIFY_DONE;
1110 return NOTIFY_OK;
1114 static int __init cpu_hotplug_pm_sync_init(void)
1117 * cpu_hotplug_pm_callback has higher priority than x86
1118 * bsp_pm_callback which depends on cpu_hotplug_pm_callback
1119 * to disable cpu hotplug to avoid cpu hotplug race.
1121 pm_notifier(cpu_hotplug_pm_callback, 0);
1122 return 0;
1124 core_initcall(cpu_hotplug_pm_sync_init);
1126 #endif /* CONFIG_PM_SLEEP_SMP */
1128 #endif /* CONFIG_SMP */
1130 /* Boot processor state steps */
1131 static struct cpuhp_step cpuhp_bp_states[] = {
1132 [CPUHP_OFFLINE] = {
1133 .name = "offline",
1134 .startup.single = NULL,
1135 .teardown.single = NULL,
1137 #ifdef CONFIG_SMP
1138 [CPUHP_CREATE_THREADS]= {
1139 .name = "threads:prepare",
1140 .startup.single = smpboot_create_threads,
1141 .teardown.single = NULL,
1142 .cant_stop = true,
1144 [CPUHP_PERF_PREPARE] = {
1145 .name = "perf:prepare",
1146 .startup.single = perf_event_init_cpu,
1147 .teardown.single = perf_event_exit_cpu,
1149 [CPUHP_WORKQUEUE_PREP] = {
1150 .name = "workqueue:prepare",
1151 .startup.single = workqueue_prepare_cpu,
1152 .teardown.single = NULL,
1154 [CPUHP_HRTIMERS_PREPARE] = {
1155 .name = "hrtimers:prepare",
1156 .startup.single = hrtimers_prepare_cpu,
1157 .teardown.single = hrtimers_dead_cpu,
1159 [CPUHP_SMPCFD_PREPARE] = {
1160 .name = "smpcfd:prepare",
1161 .startup.single = smpcfd_prepare_cpu,
1162 .teardown.single = smpcfd_dead_cpu,
1164 [CPUHP_RELAY_PREPARE] = {
1165 .name = "relay:prepare",
1166 .startup.single = relay_prepare_cpu,
1167 .teardown.single = NULL,
1169 [CPUHP_SLAB_PREPARE] = {
1170 .name = "slab:prepare",
1171 .startup.single = slab_prepare_cpu,
1172 .teardown.single = slab_dead_cpu,
1174 [CPUHP_RCUTREE_PREP] = {
1175 .name = "RCU/tree:prepare",
1176 .startup.single = rcutree_prepare_cpu,
1177 .teardown.single = rcutree_dead_cpu,
1180 * On the tear-down path, timers_dead_cpu() must be invoked
1181 * before blk_mq_queue_reinit_notify() from notify_dead(),
1182 * otherwise a RCU stall occurs.
1184 [CPUHP_TIMERS_DEAD] = {
1185 .name = "timers:dead",
1186 .startup.single = NULL,
1187 .teardown.single = timers_dead_cpu,
1189 /* Kicks the plugged cpu into life */
1190 [CPUHP_BRINGUP_CPU] = {
1191 .name = "cpu:bringup",
1192 .startup.single = bringup_cpu,
1193 .teardown.single = NULL,
1194 .cant_stop = true,
1196 [CPUHP_AP_SMPCFD_DYING] = {
1197 .name = "smpcfd:dying",
1198 .startup.single = NULL,
1199 .teardown.single = smpcfd_dying_cpu,
1202 * Handled on controll processor until the plugged processor manages
1203 * this itself.
1205 [CPUHP_TEARDOWN_CPU] = {
1206 .name = "cpu:teardown",
1207 .startup.single = NULL,
1208 .teardown.single = takedown_cpu,
1209 .cant_stop = true,
1211 #else
1212 [CPUHP_BRINGUP_CPU] = { },
1213 #endif
1216 /* Application processor state steps */
1217 static struct cpuhp_step cpuhp_ap_states[] = {
1218 #ifdef CONFIG_SMP
1219 /* Final state before CPU kills itself */
1220 [CPUHP_AP_IDLE_DEAD] = {
1221 .name = "idle:dead",
1224 * Last state before CPU enters the idle loop to die. Transient state
1225 * for synchronization.
1227 [CPUHP_AP_OFFLINE] = {
1228 .name = "ap:offline",
1229 .cant_stop = true,
1231 /* First state is scheduler control. Interrupts are disabled */
1232 [CPUHP_AP_SCHED_STARTING] = {
1233 .name = "sched:starting",
1234 .startup.single = sched_cpu_starting,
1235 .teardown.single = sched_cpu_dying,
1237 [CPUHP_AP_RCUTREE_DYING] = {
1238 .name = "RCU/tree:dying",
1239 .startup.single = NULL,
1240 .teardown.single = rcutree_dying_cpu,
1242 /* Entry state on starting. Interrupts enabled from here on. Transient
1243 * state for synchronsization */
1244 [CPUHP_AP_ONLINE] = {
1245 .name = "ap:online",
1247 /* Handle smpboot threads park/unpark */
1248 [CPUHP_AP_SMPBOOT_THREADS] = {
1249 .name = "smpboot/threads:online",
1250 .startup.single = smpboot_unpark_threads,
1251 .teardown.single = NULL,
1253 [CPUHP_AP_PERF_ONLINE] = {
1254 .name = "perf:online",
1255 .startup.single = perf_event_init_cpu,
1256 .teardown.single = perf_event_exit_cpu,
1258 [CPUHP_AP_WORKQUEUE_ONLINE] = {
1259 .name = "workqueue:online",
1260 .startup.single = workqueue_online_cpu,
1261 .teardown.single = workqueue_offline_cpu,
1263 [CPUHP_AP_RCUTREE_ONLINE] = {
1264 .name = "RCU/tree:online",
1265 .startup.single = rcutree_online_cpu,
1266 .teardown.single = rcutree_offline_cpu,
1268 #endif
1270 * The dynamically registered state space is here
1273 #ifdef CONFIG_SMP
1274 /* Last state is scheduler control setting the cpu active */
1275 [CPUHP_AP_ACTIVE] = {
1276 .name = "sched:active",
1277 .startup.single = sched_cpu_activate,
1278 .teardown.single = sched_cpu_deactivate,
1280 #endif
1282 /* CPU is fully up and running. */
1283 [CPUHP_ONLINE] = {
1284 .name = "online",
1285 .startup.single = NULL,
1286 .teardown.single = NULL,
1290 /* Sanity check for callbacks */
1291 static int cpuhp_cb_check(enum cpuhp_state state)
1293 if (state <= CPUHP_OFFLINE || state >= CPUHP_ONLINE)
1294 return -EINVAL;
1295 return 0;
1299 * Returns a free for dynamic slot assignment of the Online state. The states
1300 * are protected by the cpuhp_slot_states mutex and an empty slot is identified
1301 * by having no name assigned.
1303 static int cpuhp_reserve_state(enum cpuhp_state state)
1305 enum cpuhp_state i;
1307 for (i = CPUHP_AP_ONLINE_DYN; i <= CPUHP_AP_ONLINE_DYN_END; i++) {
1308 if (!cpuhp_ap_states[i].name)
1309 return i;
1311 WARN(1, "No more dynamic states available for CPU hotplug\n");
1312 return -ENOSPC;
1315 static int cpuhp_store_callbacks(enum cpuhp_state state, const char *name,
1316 int (*startup)(unsigned int cpu),
1317 int (*teardown)(unsigned int cpu),
1318 bool multi_instance)
1320 /* (Un)Install the callbacks for further cpu hotplug operations */
1321 struct cpuhp_step *sp;
1322 int ret = 0;
1324 mutex_lock(&cpuhp_state_mutex);
1326 if (state == CPUHP_AP_ONLINE_DYN) {
1327 ret = cpuhp_reserve_state(state);
1328 if (ret < 0)
1329 goto out;
1330 state = ret;
1332 sp = cpuhp_get_step(state);
1333 if (name && sp->name) {
1334 ret = -EBUSY;
1335 goto out;
1337 sp->startup.single = startup;
1338 sp->teardown.single = teardown;
1339 sp->name = name;
1340 sp->multi_instance = multi_instance;
1341 INIT_HLIST_HEAD(&sp->list);
1342 out:
1343 mutex_unlock(&cpuhp_state_mutex);
1344 return ret;
1347 static void *cpuhp_get_teardown_cb(enum cpuhp_state state)
1349 return cpuhp_get_step(state)->teardown.single;
1353 * Call the startup/teardown function for a step either on the AP or
1354 * on the current CPU.
1356 static int cpuhp_issue_call(int cpu, enum cpuhp_state state, bool bringup,
1357 struct hlist_node *node)
1359 struct cpuhp_step *sp = cpuhp_get_step(state);
1360 int ret;
1362 if ((bringup && !sp->startup.single) ||
1363 (!bringup && !sp->teardown.single))
1364 return 0;
1366 * The non AP bound callbacks can fail on bringup. On teardown
1367 * e.g. module removal we crash for now.
1369 #ifdef CONFIG_SMP
1370 if (cpuhp_is_ap_state(state))
1371 ret = cpuhp_invoke_ap_callback(cpu, state, bringup, node);
1372 else
1373 ret = cpuhp_invoke_callback(cpu, state, bringup, node);
1374 #else
1375 ret = cpuhp_invoke_callback(cpu, state, bringup, node);
1376 #endif
1377 BUG_ON(ret && !bringup);
1378 return ret;
1382 * Called from __cpuhp_setup_state on a recoverable failure.
1384 * Note: The teardown callbacks for rollback are not allowed to fail!
1386 static void cpuhp_rollback_install(int failedcpu, enum cpuhp_state state,
1387 struct hlist_node *node)
1389 int cpu;
1391 /* Roll back the already executed steps on the other cpus */
1392 for_each_present_cpu(cpu) {
1393 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1394 int cpustate = st->state;
1396 if (cpu >= failedcpu)
1397 break;
1399 /* Did we invoke the startup call on that cpu ? */
1400 if (cpustate >= state)
1401 cpuhp_issue_call(cpu, state, false, node);
1405 int __cpuhp_state_add_instance(enum cpuhp_state state, struct hlist_node *node,
1406 bool invoke)
1408 struct cpuhp_step *sp;
1409 int cpu;
1410 int ret;
1412 sp = cpuhp_get_step(state);
1413 if (sp->multi_instance == false)
1414 return -EINVAL;
1416 get_online_cpus();
1418 if (!invoke || !sp->startup.multi)
1419 goto add_node;
1422 * Try to call the startup callback for each present cpu
1423 * depending on the hotplug state of the cpu.
1425 for_each_present_cpu(cpu) {
1426 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1427 int cpustate = st->state;
1429 if (cpustate < state)
1430 continue;
1432 ret = cpuhp_issue_call(cpu, state, true, node);
1433 if (ret) {
1434 if (sp->teardown.multi)
1435 cpuhp_rollback_install(cpu, state, node);
1436 goto err;
1439 add_node:
1440 ret = 0;
1441 mutex_lock(&cpuhp_state_mutex);
1442 hlist_add_head(node, &sp->list);
1443 mutex_unlock(&cpuhp_state_mutex);
1445 err:
1446 put_online_cpus();
1447 return ret;
1449 EXPORT_SYMBOL_GPL(__cpuhp_state_add_instance);
1452 * __cpuhp_setup_state - Setup the callbacks for an hotplug machine state
1453 * @state: The state to setup
1454 * @invoke: If true, the startup function is invoked for cpus where
1455 * cpu state >= @state
1456 * @startup: startup callback function
1457 * @teardown: teardown callback function
1458 * @multi_instance: State is set up for multiple instances which get
1459 * added afterwards.
1461 * Returns:
1462 * On success:
1463 * Positive state number if @state is CPUHP_AP_ONLINE_DYN
1464 * 0 for all other states
1465 * On failure: proper (negative) error code
1467 int __cpuhp_setup_state(enum cpuhp_state state,
1468 const char *name, bool invoke,
1469 int (*startup)(unsigned int cpu),
1470 int (*teardown)(unsigned int cpu),
1471 bool multi_instance)
1473 int cpu, ret = 0;
1474 bool dynstate;
1476 if (cpuhp_cb_check(state) || !name)
1477 return -EINVAL;
1479 get_online_cpus();
1481 ret = cpuhp_store_callbacks(state, name, startup, teardown,
1482 multi_instance);
1484 dynstate = state == CPUHP_AP_ONLINE_DYN;
1485 if (ret > 0 && dynstate) {
1486 state = ret;
1487 ret = 0;
1490 if (ret || !invoke || !startup)
1491 goto out;
1494 * Try to call the startup callback for each present cpu
1495 * depending on the hotplug state of the cpu.
1497 for_each_present_cpu(cpu) {
1498 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1499 int cpustate = st->state;
1501 if (cpustate < state)
1502 continue;
1504 ret = cpuhp_issue_call(cpu, state, true, NULL);
1505 if (ret) {
1506 if (teardown)
1507 cpuhp_rollback_install(cpu, state, NULL);
1508 cpuhp_store_callbacks(state, NULL, NULL, NULL, false);
1509 goto out;
1512 out:
1513 put_online_cpus();
1515 * If the requested state is CPUHP_AP_ONLINE_DYN, return the
1516 * dynamically allocated state in case of success.
1518 if (!ret && dynstate)
1519 return state;
1520 return ret;
1522 EXPORT_SYMBOL(__cpuhp_setup_state);
1524 int __cpuhp_state_remove_instance(enum cpuhp_state state,
1525 struct hlist_node *node, bool invoke)
1527 struct cpuhp_step *sp = cpuhp_get_step(state);
1528 int cpu;
1530 BUG_ON(cpuhp_cb_check(state));
1532 if (!sp->multi_instance)
1533 return -EINVAL;
1535 get_online_cpus();
1536 if (!invoke || !cpuhp_get_teardown_cb(state))
1537 goto remove;
1539 * Call the teardown callback for each present cpu depending
1540 * on the hotplug state of the cpu. This function is not
1541 * allowed to fail currently!
1543 for_each_present_cpu(cpu) {
1544 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1545 int cpustate = st->state;
1547 if (cpustate >= state)
1548 cpuhp_issue_call(cpu, state, false, node);
1551 remove:
1552 mutex_lock(&cpuhp_state_mutex);
1553 hlist_del(node);
1554 mutex_unlock(&cpuhp_state_mutex);
1555 put_online_cpus();
1557 return 0;
1559 EXPORT_SYMBOL_GPL(__cpuhp_state_remove_instance);
1561 * __cpuhp_remove_state - Remove the callbacks for an hotplug machine state
1562 * @state: The state to remove
1563 * @invoke: If true, the teardown function is invoked for cpus where
1564 * cpu state >= @state
1566 * The teardown callback is currently not allowed to fail. Think
1567 * about module removal!
1569 void __cpuhp_remove_state(enum cpuhp_state state, bool invoke)
1571 struct cpuhp_step *sp = cpuhp_get_step(state);
1572 int cpu;
1574 BUG_ON(cpuhp_cb_check(state));
1576 get_online_cpus();
1578 if (sp->multi_instance) {
1579 WARN(!hlist_empty(&sp->list),
1580 "Error: Removing state %d which has instances left.\n",
1581 state);
1582 goto remove;
1585 if (!invoke || !cpuhp_get_teardown_cb(state))
1586 goto remove;
1589 * Call the teardown callback for each present cpu depending
1590 * on the hotplug state of the cpu. This function is not
1591 * allowed to fail currently!
1593 for_each_present_cpu(cpu) {
1594 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1595 int cpustate = st->state;
1597 if (cpustate >= state)
1598 cpuhp_issue_call(cpu, state, false, NULL);
1600 remove:
1601 cpuhp_store_callbacks(state, NULL, NULL, NULL, false);
1602 put_online_cpus();
1604 EXPORT_SYMBOL(__cpuhp_remove_state);
1606 #if defined(CONFIG_SYSFS) && defined(CONFIG_HOTPLUG_CPU)
1607 static ssize_t show_cpuhp_state(struct device *dev,
1608 struct device_attribute *attr, char *buf)
1610 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id);
1612 return sprintf(buf, "%d\n", st->state);
1614 static DEVICE_ATTR(state, 0444, show_cpuhp_state, NULL);
1616 static ssize_t write_cpuhp_target(struct device *dev,
1617 struct device_attribute *attr,
1618 const char *buf, size_t count)
1620 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id);
1621 struct cpuhp_step *sp;
1622 int target, ret;
1624 ret = kstrtoint(buf, 10, &target);
1625 if (ret)
1626 return ret;
1628 #ifdef CONFIG_CPU_HOTPLUG_STATE_CONTROL
1629 if (target < CPUHP_OFFLINE || target > CPUHP_ONLINE)
1630 return -EINVAL;
1631 #else
1632 if (target != CPUHP_OFFLINE && target != CPUHP_ONLINE)
1633 return -EINVAL;
1634 #endif
1636 ret = lock_device_hotplug_sysfs();
1637 if (ret)
1638 return ret;
1640 mutex_lock(&cpuhp_state_mutex);
1641 sp = cpuhp_get_step(target);
1642 ret = !sp->name || sp->cant_stop ? -EINVAL : 0;
1643 mutex_unlock(&cpuhp_state_mutex);
1644 if (ret)
1645 return ret;
1647 if (st->state < target)
1648 ret = do_cpu_up(dev->id, target);
1649 else
1650 ret = do_cpu_down(dev->id, target);
1652 unlock_device_hotplug();
1653 return ret ? ret : count;
1656 static ssize_t show_cpuhp_target(struct device *dev,
1657 struct device_attribute *attr, char *buf)
1659 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id);
1661 return sprintf(buf, "%d\n", st->target);
1663 static DEVICE_ATTR(target, 0644, show_cpuhp_target, write_cpuhp_target);
1665 static struct attribute *cpuhp_cpu_attrs[] = {
1666 &dev_attr_state.attr,
1667 &dev_attr_target.attr,
1668 NULL
1671 static struct attribute_group cpuhp_cpu_attr_group = {
1672 .attrs = cpuhp_cpu_attrs,
1673 .name = "hotplug",
1674 NULL
1677 static ssize_t show_cpuhp_states(struct device *dev,
1678 struct device_attribute *attr, char *buf)
1680 ssize_t cur, res = 0;
1681 int i;
1683 mutex_lock(&cpuhp_state_mutex);
1684 for (i = CPUHP_OFFLINE; i <= CPUHP_ONLINE; i++) {
1685 struct cpuhp_step *sp = cpuhp_get_step(i);
1687 if (sp->name) {
1688 cur = sprintf(buf, "%3d: %s\n", i, sp->name);
1689 buf += cur;
1690 res += cur;
1693 mutex_unlock(&cpuhp_state_mutex);
1694 return res;
1696 static DEVICE_ATTR(states, 0444, show_cpuhp_states, NULL);
1698 static struct attribute *cpuhp_cpu_root_attrs[] = {
1699 &dev_attr_states.attr,
1700 NULL
1703 static struct attribute_group cpuhp_cpu_root_attr_group = {
1704 .attrs = cpuhp_cpu_root_attrs,
1705 .name = "hotplug",
1706 NULL
1709 static int __init cpuhp_sysfs_init(void)
1711 int cpu, ret;
1713 ret = sysfs_create_group(&cpu_subsys.dev_root->kobj,
1714 &cpuhp_cpu_root_attr_group);
1715 if (ret)
1716 return ret;
1718 for_each_possible_cpu(cpu) {
1719 struct device *dev = get_cpu_device(cpu);
1721 if (!dev)
1722 continue;
1723 ret = sysfs_create_group(&dev->kobj, &cpuhp_cpu_attr_group);
1724 if (ret)
1725 return ret;
1727 return 0;
1729 device_initcall(cpuhp_sysfs_init);
1730 #endif
1733 * cpu_bit_bitmap[] is a special, "compressed" data structure that
1734 * represents all NR_CPUS bits binary values of 1<<nr.
1736 * It is used by cpumask_of() to get a constant address to a CPU
1737 * mask value that has a single bit set only.
1740 /* cpu_bit_bitmap[0] is empty - so we can back into it */
1741 #define MASK_DECLARE_1(x) [x+1][0] = (1UL << (x))
1742 #define MASK_DECLARE_2(x) MASK_DECLARE_1(x), MASK_DECLARE_1(x+1)
1743 #define MASK_DECLARE_4(x) MASK_DECLARE_2(x), MASK_DECLARE_2(x+2)
1744 #define MASK_DECLARE_8(x) MASK_DECLARE_4(x), MASK_DECLARE_4(x+4)
1746 const unsigned long cpu_bit_bitmap[BITS_PER_LONG+1][BITS_TO_LONGS(NR_CPUS)] = {
1748 MASK_DECLARE_8(0), MASK_DECLARE_8(8),
1749 MASK_DECLARE_8(16), MASK_DECLARE_8(24),
1750 #if BITS_PER_LONG > 32
1751 MASK_DECLARE_8(32), MASK_DECLARE_8(40),
1752 MASK_DECLARE_8(48), MASK_DECLARE_8(56),
1753 #endif
1755 EXPORT_SYMBOL_GPL(cpu_bit_bitmap);
1757 const DECLARE_BITMAP(cpu_all_bits, NR_CPUS) = CPU_BITS_ALL;
1758 EXPORT_SYMBOL(cpu_all_bits);
1760 #ifdef CONFIG_INIT_ALL_POSSIBLE
1761 struct cpumask __cpu_possible_mask __read_mostly
1762 = {CPU_BITS_ALL};
1763 #else
1764 struct cpumask __cpu_possible_mask __read_mostly;
1765 #endif
1766 EXPORT_SYMBOL(__cpu_possible_mask);
1768 struct cpumask __cpu_online_mask __read_mostly;
1769 EXPORT_SYMBOL(__cpu_online_mask);
1771 struct cpumask __cpu_present_mask __read_mostly;
1772 EXPORT_SYMBOL(__cpu_present_mask);
1774 struct cpumask __cpu_active_mask __read_mostly;
1775 EXPORT_SYMBOL(__cpu_active_mask);
1777 void init_cpu_present(const struct cpumask *src)
1779 cpumask_copy(&__cpu_present_mask, src);
1782 void init_cpu_possible(const struct cpumask *src)
1784 cpumask_copy(&__cpu_possible_mask, src);
1787 void init_cpu_online(const struct cpumask *src)
1789 cpumask_copy(&__cpu_online_mask, src);
1793 * Activate the first processor.
1795 void __init boot_cpu_init(void)
1797 int cpu = smp_processor_id();
1799 /* Mark the boot cpu "present", "online" etc for SMP and UP case */
1800 set_cpu_online(cpu, true);
1801 set_cpu_active(cpu, true);
1802 set_cpu_present(cpu, true);
1803 set_cpu_possible(cpu, true);
1807 * Must be called _AFTER_ setting up the per_cpu areas
1809 void __init boot_cpu_state_init(void)
1811 per_cpu_ptr(&cpuhp_state, smp_processor_id())->state = CPUHP_ONLINE;