1
2
3
4
5
6#include <linux/sched/mm.h>
7#include <linux/proc_fs.h>
8#include <linux/smp.h>
9#include <linux/init.h>
10#include <linux/notifier.h>
11#include <linux/sched/signal.h>
12#include <linux/sched/hotplug.h>
13#include <linux/sched/isolation.h>
14#include <linux/sched/task.h>
15#include <linux/sched/smt.h>
16#include <linux/unistd.h>
17#include <linux/cpu.h>
18#include <linux/oom.h>
19#include <linux/rcupdate.h>
20#include <linux/export.h>
21#include <linux/bug.h>
22#include <linux/kthread.h>
23#include <linux/stop_machine.h>
24#include <linux/mutex.h>
25#include <linux/gfp.h>
26#include <linux/suspend.h>
27#include <linux/lockdep.h>
28#include <linux/tick.h>
29#include <linux/irq.h>
30#include <linux/nmi.h>
31#include <linux/smpboot.h>
32#include <linux/relay.h>
33#include <linux/slab.h>
34#include <linux/percpu-rwsem.h>
35
36#include <trace/events/power.h>
37#define CREATE_TRACE_POINTS
38#include <trace/events/cpuhp.h>
39
40#include "smpboot.h"
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56struct cpuhp_cpu_state {
57 enum cpuhp_state state;
58 enum cpuhp_state target;
59 enum cpuhp_state fail;
60#ifdef CONFIG_SMP
61 struct task_struct *thread;
62 bool should_run;
63 bool rollback;
64 bool single;
65 bool bringup;
66 struct hlist_node *node;
67 struct hlist_node *last;
68 enum cpuhp_state cb_state;
69 int result;
70 struct completion done_up;
71 struct completion done_down;
72#endif
73};
74
75static DEFINE_PER_CPU(struct cpuhp_cpu_state, cpuhp_state) = {
76 .fail = CPUHP_INVALID,
77};
78
79#ifdef CONFIG_SMP
80cpumask_t cpus_booted_once_mask;
81#endif
82
83#if defined(CONFIG_LOCKDEP) && defined(CONFIG_SMP)
84static struct lockdep_map cpuhp_state_up_map =
85 STATIC_LOCKDEP_MAP_INIT("cpuhp_state-up", &cpuhp_state_up_map);
86static struct lockdep_map cpuhp_state_down_map =
87 STATIC_LOCKDEP_MAP_INIT("cpuhp_state-down", &cpuhp_state_down_map);
88
89
90static inline void cpuhp_lock_acquire(bool bringup)
91{
92 lock_map_acquire(bringup ? &cpuhp_state_up_map : &cpuhp_state_down_map);
93}
94
95static inline void cpuhp_lock_release(bool bringup)
96{
97 lock_map_release(bringup ? &cpuhp_state_up_map : &cpuhp_state_down_map);
98}
99#else
100
101static inline void cpuhp_lock_acquire(bool bringup) { }
102static inline void cpuhp_lock_release(bool bringup) { }
103
104#endif
105
106
107
108
109
110
111
112
113struct cpuhp_step {
114 const char *name;
115 union {
116 int (*single)(unsigned int cpu);
117 int (*multi)(unsigned int cpu,
118 struct hlist_node *node);
119 } startup;
120 union {
121 int (*single)(unsigned int cpu);
122 int (*multi)(unsigned int cpu,
123 struct hlist_node *node);
124 } teardown;
125 struct hlist_head list;
126 bool cant_stop;
127 bool multi_instance;
128};
129
130static DEFINE_MUTEX(cpuhp_state_mutex);
131static struct cpuhp_step cpuhp_hp_states[];
132
133static struct cpuhp_step *cpuhp_get_step(enum cpuhp_state state)
134{
135 return cpuhp_hp_states + state;
136}
137
138
139
140
141
142
143
144
145
146
147
148static int cpuhp_invoke_callback(unsigned int cpu, enum cpuhp_state state,
149 bool bringup, struct hlist_node *node,
150 struct hlist_node **lastp)
151{
152 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
153 struct cpuhp_step *step = cpuhp_get_step(state);
154 int (*cbm)(unsigned int cpu, struct hlist_node *node);
155 int (*cb)(unsigned int cpu);
156 int ret, cnt;
157
158 if (st->fail == state) {
159 st->fail = CPUHP_INVALID;
160
161 if (!(bringup ? step->startup.single : step->teardown.single))
162 return 0;
163
164 return -EAGAIN;
165 }
166
167 if (!step->multi_instance) {
168 WARN_ON_ONCE(lastp && *lastp);
169 cb = bringup ? step->startup.single : step->teardown.single;
170 if (!cb)
171 return 0;
172 trace_cpuhp_enter(cpu, st->target, state, cb);
173 ret = cb(cpu);
174 trace_cpuhp_exit(cpu, st->state, state, ret);
175 return ret;
176 }
177 cbm = bringup ? step->startup.multi : step->teardown.multi;
178 if (!cbm)
179 return 0;
180
181
182 if (node) {
183 WARN_ON_ONCE(lastp && *lastp);
184 trace_cpuhp_multi_enter(cpu, st->target, state, cbm, node);
185 ret = cbm(cpu, node);
186 trace_cpuhp_exit(cpu, st->state, state, ret);
187 return ret;
188 }
189
190
191 cnt = 0;
192 hlist_for_each(node, &step->list) {
193 if (lastp && node == *lastp)
194 break;
195
196 trace_cpuhp_multi_enter(cpu, st->target, state, cbm, node);
197 ret = cbm(cpu, node);
198 trace_cpuhp_exit(cpu, st->state, state, ret);
199 if (ret) {
200 if (!lastp)
201 goto err;
202
203 *lastp = node;
204 return ret;
205 }
206 cnt++;
207 }
208 if (lastp)
209 *lastp = NULL;
210 return 0;
211err:
212
213 cbm = !bringup ? step->startup.multi : step->teardown.multi;
214 if (!cbm)
215 return ret;
216
217 hlist_for_each(node, &step->list) {
218 if (!cnt--)
219 break;
220
221 trace_cpuhp_multi_enter(cpu, st->target, state, cbm, node);
222 ret = cbm(cpu, node);
223 trace_cpuhp_exit(cpu, st->state, state, ret);
224
225
226
227 WARN_ON_ONCE(ret);
228 }
229 return ret;
230}
231
232#ifdef CONFIG_SMP
233static bool cpuhp_is_ap_state(enum cpuhp_state state)
234{
235
236
237
238
239 return state > CPUHP_BRINGUP_CPU && state != CPUHP_TEARDOWN_CPU;
240}
241
242static inline void wait_for_ap_thread(struct cpuhp_cpu_state *st, bool bringup)
243{
244 struct completion *done = bringup ? &st->done_up : &st->done_down;
245 wait_for_completion(done);
246}
247
248static inline void complete_ap_thread(struct cpuhp_cpu_state *st, bool bringup)
249{
250 struct completion *done = bringup ? &st->done_up : &st->done_down;
251 complete(done);
252}
253
254
255
256
257static bool cpuhp_is_atomic_state(enum cpuhp_state state)
258{
259 return CPUHP_AP_IDLE_DEAD <= state && state < CPUHP_AP_ONLINE;
260}
261
262
263static DEFINE_MUTEX(cpu_add_remove_lock);
264bool cpuhp_tasks_frozen;
265EXPORT_SYMBOL_GPL(cpuhp_tasks_frozen);
266
267
268
269
270
271void cpu_maps_update_begin(void)
272{
273 mutex_lock(&cpu_add_remove_lock);
274}
275
276void cpu_maps_update_done(void)
277{
278 mutex_unlock(&cpu_add_remove_lock);
279}
280
281
282
283
284
285static int cpu_hotplug_disabled;
286
287#ifdef CONFIG_HOTPLUG_CPU
288
289DEFINE_STATIC_PERCPU_RWSEM(cpu_hotplug_lock);
290
291void cpus_read_lock(void)
292{
293 percpu_down_read(&cpu_hotplug_lock);
294}
295EXPORT_SYMBOL_GPL(cpus_read_lock);
296
297int cpus_read_trylock(void)
298{
299 return percpu_down_read_trylock(&cpu_hotplug_lock);
300}
301EXPORT_SYMBOL_GPL(cpus_read_trylock);
302
303void cpus_read_unlock(void)
304{
305 percpu_up_read(&cpu_hotplug_lock);
306}
307EXPORT_SYMBOL_GPL(cpus_read_unlock);
308
309void cpus_write_lock(void)
310{
311 percpu_down_write(&cpu_hotplug_lock);
312}
313
314void cpus_write_unlock(void)
315{
316 percpu_up_write(&cpu_hotplug_lock);
317}
318
319void lockdep_assert_cpus_held(void)
320{
321
322
323
324
325
326
327 if (system_state < SYSTEM_RUNNING)
328 return;
329
330 percpu_rwsem_assert_held(&cpu_hotplug_lock);
331}
332
333static void lockdep_acquire_cpus_lock(void)
334{
335 rwsem_acquire(&cpu_hotplug_lock.dep_map, 0, 0, _THIS_IP_);
336}
337
338static void lockdep_release_cpus_lock(void)
339{
340 rwsem_release(&cpu_hotplug_lock.dep_map, _THIS_IP_);
341}
342
343
344
345
346
347
348
349
350void cpu_hotplug_disable(void)
351{
352 cpu_maps_update_begin();
353 cpu_hotplug_disabled++;
354 cpu_maps_update_done();
355}
356EXPORT_SYMBOL_GPL(cpu_hotplug_disable);
357
358static void __cpu_hotplug_enable(void)
359{
360 if (WARN_ONCE(!cpu_hotplug_disabled, "Unbalanced cpu hotplug enable\n"))
361 return;
362 cpu_hotplug_disabled--;
363}
364
365void cpu_hotplug_enable(void)
366{
367 cpu_maps_update_begin();
368 __cpu_hotplug_enable();
369 cpu_maps_update_done();
370}
371EXPORT_SYMBOL_GPL(cpu_hotplug_enable);
372
373#else
374
375static void lockdep_acquire_cpus_lock(void)
376{
377}
378
379static void lockdep_release_cpus_lock(void)
380{
381}
382
383#endif
384
385
386
387
388
389void __weak arch_smt_update(void) { }
390
391#ifdef CONFIG_HOTPLUG_SMT
392enum cpuhp_smt_control cpu_smt_control __read_mostly = CPU_SMT_ENABLED;
393
394void __init cpu_smt_disable(bool force)
395{
396 if (!cpu_smt_possible())
397 return;
398
399 if (force) {
400 pr_info("SMT: Force disabled\n");
401 cpu_smt_control = CPU_SMT_FORCE_DISABLED;
402 } else {
403 pr_info("SMT: disabled\n");
404 cpu_smt_control = CPU_SMT_DISABLED;
405 }
406}
407
408
409
410
411
412void __init cpu_smt_check_topology(void)
413{
414 if (!topology_smt_supported())
415 cpu_smt_control = CPU_SMT_NOT_SUPPORTED;
416}
417
418static int __init smt_cmdline_disable(char *str)
419{
420 cpu_smt_disable(str && !strcmp(str, "force"));
421 return 0;
422}
423early_param("nosmt", smt_cmdline_disable);
424
425static inline bool cpu_smt_allowed(unsigned int cpu)
426{
427 if (cpu_smt_control == CPU_SMT_ENABLED)
428 return true;
429
430 if (topology_is_primary_thread(cpu))
431 return true;
432
433
434
435
436
437
438
439 return !cpumask_test_cpu(cpu, &cpus_booted_once_mask);
440}
441
442
443bool cpu_smt_possible(void)
444{
445 return cpu_smt_control != CPU_SMT_FORCE_DISABLED &&
446 cpu_smt_control != CPU_SMT_NOT_SUPPORTED;
447}
448EXPORT_SYMBOL_GPL(cpu_smt_possible);
449#else
450static inline bool cpu_smt_allowed(unsigned int cpu) { return true; }
451#endif
452
453static inline enum cpuhp_state
454cpuhp_set_state(struct cpuhp_cpu_state *st, enum cpuhp_state target)
455{
456 enum cpuhp_state prev_state = st->state;
457
458 st->rollback = false;
459 st->last = NULL;
460
461 st->target = target;
462 st->single = false;
463 st->bringup = st->state < target;
464
465 return prev_state;
466}
467
468static inline void
469cpuhp_reset_state(struct cpuhp_cpu_state *st, enum cpuhp_state prev_state)
470{
471 st->rollback = true;
472
473
474
475
476
477 if (!st->last) {
478 if (st->bringup)
479 st->state--;
480 else
481 st->state++;
482 }
483
484 st->target = prev_state;
485 st->bringup = !st->bringup;
486}
487
488
489static void __cpuhp_kick_ap(struct cpuhp_cpu_state *st)
490{
491 if (!st->single && st->state == st->target)
492 return;
493
494 st->result = 0;
495
496
497
498
499 smp_mb();
500 st->should_run = true;
501 wake_up_process(st->thread);
502 wait_for_ap_thread(st, st->bringup);
503}
504
505static int cpuhp_kick_ap(struct cpuhp_cpu_state *st, enum cpuhp_state target)
506{
507 enum cpuhp_state prev_state;
508 int ret;
509
510 prev_state = cpuhp_set_state(st, target);
511 __cpuhp_kick_ap(st);
512 if ((ret = st->result)) {
513 cpuhp_reset_state(st, prev_state);
514 __cpuhp_kick_ap(st);
515 }
516
517 return ret;
518}
519
520static int bringup_wait_for_ap(unsigned int cpu)
521{
522 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
523
524
525 wait_for_ap_thread(st, true);
526 if (WARN_ON_ONCE((!cpu_online(cpu))))
527 return -ECANCELED;
528
529
530 kthread_unpark(st->thread);
531
532
533
534
535
536
537
538
539 if (!cpu_smt_allowed(cpu))
540 return -ECANCELED;
541
542 if (st->target <= CPUHP_AP_ONLINE_IDLE)
543 return 0;
544
545 return cpuhp_kick_ap(st, st->target);
546}
547
548static int bringup_cpu(unsigned int cpu)
549{
550 struct task_struct *idle = idle_thread_get(cpu);
551 int ret;
552
553
554
555
556
557
558 irq_lock_sparse();
559
560
561 ret = __cpu_up(cpu, idle);
562 irq_unlock_sparse();
563 if (ret)
564 return ret;
565 return bringup_wait_for_ap(cpu);
566}
567
568static int finish_cpu(unsigned int cpu)
569{
570 struct task_struct *idle = idle_thread_get(cpu);
571 struct mm_struct *mm = idle->active_mm;
572
573
574
575
576
577 if (mm != &init_mm)
578 idle->active_mm = &init_mm;
579 mmdrop(mm);
580 return 0;
581}
582
583
584
585
586
587static void undo_cpu_up(unsigned int cpu, struct cpuhp_cpu_state *st)
588{
589 for (st->state--; st->state > st->target; st->state--)
590 cpuhp_invoke_callback(cpu, st->state, false, NULL, NULL);
591}
592
593static inline bool can_rollback_cpu(struct cpuhp_cpu_state *st)
594{
595 if (IS_ENABLED(CONFIG_HOTPLUG_CPU))
596 return true;
597
598
599
600
601
602
603
604 return st->state <= CPUHP_BRINGUP_CPU;
605}
606
607static int cpuhp_up_callbacks(unsigned int cpu, struct cpuhp_cpu_state *st,
608 enum cpuhp_state target)
609{
610 enum cpuhp_state prev_state = st->state;
611 int ret = 0;
612
613 while (st->state < target) {
614 st->state++;
615 ret = cpuhp_invoke_callback(cpu, st->state, true, NULL, NULL);
616 if (ret) {
617 if (can_rollback_cpu(st)) {
618 st->target = prev_state;
619 undo_cpu_up(cpu, st);
620 }
621 break;
622 }
623 }
624 return ret;
625}
626
627
628
629
630static void cpuhp_create(unsigned int cpu)
631{
632 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
633
634 init_completion(&st->done_up);
635 init_completion(&st->done_down);
636}
637
638static int cpuhp_should_run(unsigned int cpu)
639{
640 struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
641
642 return st->should_run;
643}
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659static void cpuhp_thread_fun(unsigned int cpu)
660{
661 struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
662 bool bringup = st->bringup;
663 enum cpuhp_state state;
664
665 if (WARN_ON_ONCE(!st->should_run))
666 return;
667
668
669
670
671
672 smp_mb();
673
674
675
676
677
678
679 lockdep_acquire_cpus_lock();
680 cpuhp_lock_acquire(bringup);
681
682 if (st->single) {
683 state = st->cb_state;
684 st->should_run = false;
685 } else {
686 if (bringup) {
687 st->state++;
688 state = st->state;
689 st->should_run = (st->state < st->target);
690 WARN_ON_ONCE(st->state > st->target);
691 } else {
692 state = st->state;
693 st->state--;
694 st->should_run = (st->state > st->target);
695 WARN_ON_ONCE(st->state < st->target);
696 }
697 }
698
699 WARN_ON_ONCE(!cpuhp_is_ap_state(state));
700
701 if (cpuhp_is_atomic_state(state)) {
702 local_irq_disable();
703 st->result = cpuhp_invoke_callback(cpu, state, bringup, st->node, &st->last);
704 local_irq_enable();
705
706
707
708
709 WARN_ON_ONCE(st->result);
710 } else {
711 st->result = cpuhp_invoke_callback(cpu, state, bringup, st->node, &st->last);
712 }
713
714 if (st->result) {
715
716
717
718
719
720 WARN_ON_ONCE(st->rollback);
721 st->should_run = false;
722 }
723
724 cpuhp_lock_release(bringup);
725 lockdep_release_cpus_lock();
726
727 if (!st->should_run)
728 complete_ap_thread(st, bringup);
729}
730
731
732static int
733cpuhp_invoke_ap_callback(int cpu, enum cpuhp_state state, bool bringup,
734 struct hlist_node *node)
735{
736 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
737 int ret;
738
739 if (!cpu_online(cpu))
740 return 0;
741
742 cpuhp_lock_acquire(false);
743 cpuhp_lock_release(false);
744
745 cpuhp_lock_acquire(true);
746 cpuhp_lock_release(true);
747
748
749
750
751
752 if (!st->thread)
753 return cpuhp_invoke_callback(cpu, state, bringup, node, NULL);
754
755 st->rollback = false;
756 st->last = NULL;
757
758 st->node = node;
759 st->bringup = bringup;
760 st->cb_state = state;
761 st->single = true;
762
763 __cpuhp_kick_ap(st);
764
765
766
767
768 if ((ret = st->result) && st->last) {
769 st->rollback = true;
770 st->bringup = !bringup;
771
772 __cpuhp_kick_ap(st);
773 }
774
775
776
777
778
779 st->node = st->last = NULL;
780 return ret;
781}
782
783static int cpuhp_kick_ap_work(unsigned int cpu)
784{
785 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
786 enum cpuhp_state prev_state = st->state;
787 int ret;
788
789 cpuhp_lock_acquire(false);
790 cpuhp_lock_release(false);
791
792 cpuhp_lock_acquire(true);
793 cpuhp_lock_release(true);
794
795 trace_cpuhp_enter(cpu, st->target, prev_state, cpuhp_kick_ap_work);
796 ret = cpuhp_kick_ap(st, st->target);
797 trace_cpuhp_exit(cpu, st->state, prev_state, ret);
798
799 return ret;
800}
801
802static struct smp_hotplug_thread cpuhp_threads = {
803 .store = &cpuhp_state.thread,
804 .create = &cpuhp_create,
805 .thread_should_run = cpuhp_should_run,
806 .thread_fn = cpuhp_thread_fun,
807 .thread_comm = "cpuhp/%u",
808 .selfparking = true,
809};
810
811void __init cpuhp_threads_init(void)
812{
813 BUG_ON(smpboot_register_percpu_thread(&cpuhp_threads));
814 kthread_unpark(this_cpu_read(cpuhp_state.thread));
815}
816
817#ifdef CONFIG_HOTPLUG_CPU
818#ifndef arch_clear_mm_cpumask_cpu
819#define arch_clear_mm_cpumask_cpu(cpu, mm) cpumask_clear_cpu(cpu, mm_cpumask(mm))
820#endif
821
822
823
824
825
826
827
828
829
830
831
832
833
834void clear_tasks_mm_cpumask(int cpu)
835{
836 struct task_struct *p;
837
838
839
840
841
842
843
844
845 WARN_ON(cpu_online(cpu));
846 rcu_read_lock();
847 for_each_process(p) {
848 struct task_struct *t;
849
850
851
852
853
854 t = find_lock_task_mm(p);
855 if (!t)
856 continue;
857 arch_clear_mm_cpumask_cpu(cpu, t->mm);
858 task_unlock(t);
859 }
860 rcu_read_unlock();
861}
862
863
864static int take_cpu_down(void *_param)
865{
866 struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
867 enum cpuhp_state target = max((int)st->target, CPUHP_AP_OFFLINE);
868 int err, cpu = smp_processor_id();
869 int ret;
870
871
872 err = __cpu_disable();
873 if (err < 0)
874 return err;
875
876
877
878
879
880 WARN_ON(st->state != CPUHP_TEARDOWN_CPU);
881 st->state--;
882
883 for (; st->state > target; st->state--) {
884 ret = cpuhp_invoke_callback(cpu, st->state, false, NULL, NULL);
885
886
887
888 WARN_ON_ONCE(ret);
889 }
890
891
892 tick_handover_do_timer();
893
894 tick_offline_cpu(cpu);
895
896 stop_machine_park(cpu);
897 return 0;
898}
899
900static int takedown_cpu(unsigned int cpu)
901{
902 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
903 int err;
904
905
906 kthread_park(per_cpu_ptr(&cpuhp_state, cpu)->thread);
907
908
909
910
911
912 irq_lock_sparse();
913
914
915
916
917 err = stop_machine_cpuslocked(take_cpu_down, NULL, cpumask_of(cpu));
918 if (err) {
919
920 irq_unlock_sparse();
921
922 kthread_unpark(per_cpu_ptr(&cpuhp_state, cpu)->thread);
923 return err;
924 }
925 BUG_ON(cpu_online(cpu));
926
927
928
929
930
931
932
933
934 wait_for_ap_thread(st, false);
935 BUG_ON(st->state != CPUHP_AP_IDLE_DEAD);
936
937
938 irq_unlock_sparse();
939
940 hotplug_cpu__broadcast_tick_pull(cpu);
941
942 __cpu_die(cpu);
943
944 tick_cleanup_dead_cpu(cpu);
945 rcutree_migrate_callbacks(cpu);
946 return 0;
947}
948
949static void cpuhp_complete_idle_dead(void *arg)
950{
951 struct cpuhp_cpu_state *st = arg;
952
953 complete_ap_thread(st, false);
954}
955
956void cpuhp_report_idle_dead(void)
957{
958 struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
959
960 BUG_ON(st->state != CPUHP_AP_OFFLINE);
961 rcu_report_dead(smp_processor_id());
962 st->state = CPUHP_AP_IDLE_DEAD;
963
964
965
966
967 smp_call_function_single(cpumask_first(cpu_online_mask),
968 cpuhp_complete_idle_dead, st, 0);
969}
970
971static void undo_cpu_down(unsigned int cpu, struct cpuhp_cpu_state *st)
972{
973 for (st->state++; st->state < st->target; st->state++)
974 cpuhp_invoke_callback(cpu, st->state, true, NULL, NULL);
975}
976
977static int cpuhp_down_callbacks(unsigned int cpu, struct cpuhp_cpu_state *st,
978 enum cpuhp_state target)
979{
980 enum cpuhp_state prev_state = st->state;
981 int ret = 0;
982
983 for (; st->state > target; st->state--) {
984 ret = cpuhp_invoke_callback(cpu, st->state, false, NULL, NULL);
985 if (ret) {
986 st->target = prev_state;
987 if (st->state < prev_state)
988 undo_cpu_down(cpu, st);
989 break;
990 }
991 }
992 return ret;
993}
994
995
996static int __ref _cpu_down(unsigned int cpu, int tasks_frozen,
997 enum cpuhp_state target)
998{
999 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1000 int prev_state, ret = 0;
1001
1002 if (num_online_cpus() == 1)
1003 return -EBUSY;
1004
1005 if (!cpu_present(cpu))
1006 return -EINVAL;
1007
1008 cpus_write_lock();
1009
1010 cpuhp_tasks_frozen = tasks_frozen;
1011
1012 prev_state = cpuhp_set_state(st, target);
1013
1014
1015
1016
1017 if (st->state > CPUHP_TEARDOWN_CPU) {
1018 st->target = max((int)target, CPUHP_TEARDOWN_CPU);
1019 ret = cpuhp_kick_ap_work(cpu);
1020
1021
1022
1023
1024 if (ret)
1025 goto out;
1026
1027
1028
1029
1030
1031 if (st->state > CPUHP_TEARDOWN_CPU)
1032 goto out;
1033
1034 st->target = target;
1035 }
1036
1037
1038
1039
1040 ret = cpuhp_down_callbacks(cpu, st, target);
1041 if (ret && st->state == CPUHP_TEARDOWN_CPU && st->state < prev_state) {
1042 cpuhp_reset_state(st, prev_state);
1043 __cpuhp_kick_ap(st);
1044 }
1045
1046out:
1047 cpus_write_unlock();
1048
1049
1050
1051
1052 lockup_detector_cleanup();
1053 arch_smt_update();
1054 return ret;
1055}
1056
1057static int cpu_down_maps_locked(unsigned int cpu, enum cpuhp_state target)
1058{
1059 if (cpu_hotplug_disabled)
1060 return -EBUSY;
1061 return _cpu_down(cpu, 0, target);
1062}
1063
1064static int cpu_down(unsigned int cpu, enum cpuhp_state target)
1065{
1066 int err;
1067
1068 cpu_maps_update_begin();
1069 err = cpu_down_maps_locked(cpu, target);
1070 cpu_maps_update_done();
1071 return err;
1072}
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082int cpu_device_down(struct device *dev)
1083{
1084 return cpu_down(dev->id, CPUHP_OFFLINE);
1085}
1086
1087int remove_cpu(unsigned int cpu)
1088{
1089 int ret;
1090
1091 lock_device_hotplug();
1092 ret = device_offline(get_cpu_device(cpu));
1093 unlock_device_hotplug();
1094
1095 return ret;
1096}
1097EXPORT_SYMBOL_GPL(remove_cpu);
1098
1099void smp_shutdown_nonboot_cpus(unsigned int primary_cpu)
1100{
1101 unsigned int cpu;
1102 int error;
1103
1104 cpu_maps_update_begin();
1105
1106
1107
1108
1109
1110
1111 if (!cpu_online(primary_cpu))
1112 primary_cpu = cpumask_first(cpu_online_mask);
1113
1114 for_each_online_cpu(cpu) {
1115 if (cpu == primary_cpu)
1116 continue;
1117
1118 error = cpu_down_maps_locked(cpu, CPUHP_OFFLINE);
1119 if (error) {
1120 pr_err("Failed to offline CPU%d - error=%d",
1121 cpu, error);
1122 break;
1123 }
1124 }
1125
1126
1127
1128
1129 BUG_ON(num_online_cpus() > 1);
1130
1131
1132
1133
1134
1135
1136 cpu_hotplug_disabled++;
1137
1138 cpu_maps_update_done();
1139}
1140
1141#else
1142#define takedown_cpu NULL
1143#endif
1144
1145
1146
1147
1148
1149
1150
1151
1152void notify_cpu_starting(unsigned int cpu)
1153{
1154 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1155 enum cpuhp_state target = min((int)st->target, CPUHP_AP_ONLINE);
1156 int ret;
1157
1158 rcu_cpu_starting(cpu);
1159 cpumask_set_cpu(cpu, &cpus_booted_once_mask);
1160 while (st->state < target) {
1161 st->state++;
1162 ret = cpuhp_invoke_callback(cpu, st->state, true, NULL, NULL);
1163
1164
1165
1166 WARN_ON_ONCE(ret);
1167 }
1168}
1169
1170
1171
1172
1173
1174
1175void cpuhp_online_idle(enum cpuhp_state state)
1176{
1177 struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
1178
1179
1180 if (state != CPUHP_AP_ONLINE_IDLE)
1181 return;
1182
1183
1184
1185
1186
1187 stop_machine_unpark(smp_processor_id());
1188
1189 st->state = CPUHP_AP_ONLINE_IDLE;
1190 complete_ap_thread(st, true);
1191}
1192
1193
1194static int _cpu_up(unsigned int cpu, int tasks_frozen, enum cpuhp_state target)
1195{
1196 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1197 struct task_struct *idle;
1198 int ret = 0;
1199
1200 cpus_write_lock();
1201
1202 if (!cpu_present(cpu)) {
1203 ret = -EINVAL;
1204 goto out;
1205 }
1206
1207
1208
1209
1210
1211 if (st->state >= target)
1212 goto out;
1213
1214 if (st->state == CPUHP_OFFLINE) {
1215
1216 idle = idle_thread_get(cpu);
1217 if (IS_ERR(idle)) {
1218 ret = PTR_ERR(idle);
1219 goto out;
1220 }
1221 }
1222
1223 cpuhp_tasks_frozen = tasks_frozen;
1224
1225 cpuhp_set_state(st, target);
1226
1227
1228
1229
1230 if (st->state > CPUHP_BRINGUP_CPU) {
1231 ret = cpuhp_kick_ap_work(cpu);
1232
1233
1234
1235
1236 if (ret)
1237 goto out;
1238 }
1239
1240
1241
1242
1243
1244
1245 target = min((int)target, CPUHP_BRINGUP_CPU);
1246 ret = cpuhp_up_callbacks(cpu, st, target);
1247out:
1248 cpus_write_unlock();
1249 arch_smt_update();
1250 return ret;
1251}
1252
1253static int cpu_up(unsigned int cpu, enum cpuhp_state target)
1254{
1255 int err = 0;
1256
1257 if (!cpu_possible(cpu)) {
1258 pr_err("can't online cpu %d because it is not configured as may-hotadd at boot time\n",
1259 cpu);
1260#if defined(CONFIG_IA64)
1261 pr_err("please check additional_cpus= boot parameter\n");
1262#endif
1263 return -EINVAL;
1264 }
1265
1266 err = try_online_node(cpu_to_node(cpu));
1267 if (err)
1268 return err;
1269
1270 cpu_maps_update_begin();
1271
1272 if (cpu_hotplug_disabled) {
1273 err = -EBUSY;
1274 goto out;
1275 }
1276 if (!cpu_smt_allowed(cpu)) {
1277 err = -EPERM;
1278 goto out;
1279 }
1280
1281 err = _cpu_up(cpu, 0, target);
1282out:
1283 cpu_maps_update_done();
1284 return err;
1285}
1286
1287
1288
1289
1290
1291
1292
1293
1294
1295int cpu_device_up(struct device *dev)
1296{
1297 return cpu_up(dev->id, CPUHP_ONLINE);
1298}
1299
1300int add_cpu(unsigned int cpu)
1301{
1302 int ret;
1303
1304 lock_device_hotplug();
1305 ret = device_online(get_cpu_device(cpu));
1306 unlock_device_hotplug();
1307
1308 return ret;
1309}
1310EXPORT_SYMBOL_GPL(add_cpu);
1311
1312
1313
1314
1315
1316
1317
1318
1319
1320int bringup_hibernate_cpu(unsigned int sleep_cpu)
1321{
1322 int ret;
1323
1324 if (!cpu_online(sleep_cpu)) {
1325 pr_info("Hibernated on a CPU that is offline! Bringing CPU up.\n");
1326 ret = cpu_up(sleep_cpu, CPUHP_ONLINE);
1327 if (ret) {
1328 pr_err("Failed to bring hibernate-CPU up!\n");
1329 return ret;
1330 }
1331 }
1332 return 0;
1333}
1334
1335void bringup_nonboot_cpus(unsigned int setup_max_cpus)
1336{
1337 unsigned int cpu;
1338
1339 for_each_present_cpu(cpu) {
1340 if (num_online_cpus() >= setup_max_cpus)
1341 break;
1342 if (!cpu_online(cpu))
1343 cpu_up(cpu, CPUHP_ONLINE);
1344 }
1345}
1346
1347#ifdef CONFIG_PM_SLEEP_SMP
1348static cpumask_var_t frozen_cpus;
1349
1350int freeze_secondary_cpus(int primary)
1351{
1352 int cpu, error = 0;
1353
1354 cpu_maps_update_begin();
1355 if (primary == -1) {
1356 primary = cpumask_first(cpu_online_mask);
1357 if (!housekeeping_cpu(primary, HK_FLAG_TIMER))
1358 primary = housekeeping_any_cpu(HK_FLAG_TIMER);
1359 } else {
1360 if (!cpu_online(primary))
1361 primary = cpumask_first(cpu_online_mask);
1362 }
1363
1364
1365
1366
1367
1368 cpumask_clear(frozen_cpus);
1369
1370 pr_info("Disabling non-boot CPUs ...\n");
1371 for_each_online_cpu(cpu) {
1372 if (cpu == primary)
1373 continue;
1374
1375 if (pm_wakeup_pending()) {
1376 pr_info("Wakeup pending. Abort CPU freeze\n");
1377 error = -EBUSY;
1378 break;
1379 }
1380
1381 trace_suspend_resume(TPS("CPU_OFF"), cpu, true);
1382 error = _cpu_down(cpu, 1, CPUHP_OFFLINE);
1383 trace_suspend_resume(TPS("CPU_OFF"), cpu, false);
1384 if (!error)
1385 cpumask_set_cpu(cpu, frozen_cpus);
1386 else {
1387 pr_err("Error taking CPU%d down: %d\n", cpu, error);
1388 break;
1389 }
1390 }
1391
1392 if (!error)
1393 BUG_ON(num_online_cpus() > 1);
1394 else
1395 pr_err("Non-boot CPUs are not disabled\n");
1396
1397
1398
1399
1400
1401
1402 cpu_hotplug_disabled++;
1403
1404 cpu_maps_update_done();
1405 return error;
1406}
1407
1408void __weak arch_thaw_secondary_cpus_begin(void)
1409{
1410}
1411
1412void __weak arch_thaw_secondary_cpus_end(void)
1413{
1414}
1415
1416void thaw_secondary_cpus(void)
1417{
1418 int cpu, error;
1419
1420
1421 cpu_maps_update_begin();
1422 __cpu_hotplug_enable();
1423 if (cpumask_empty(frozen_cpus))
1424 goto out;
1425
1426 pr_info("Enabling non-boot CPUs ...\n");
1427
1428 arch_thaw_secondary_cpus_begin();
1429
1430 for_each_cpu(cpu, frozen_cpus) {
1431 trace_suspend_resume(TPS("CPU_ON"), cpu, true);
1432 error = _cpu_up(cpu, 1, CPUHP_ONLINE);
1433 trace_suspend_resume(TPS("CPU_ON"), cpu, false);
1434 if (!error) {
1435 pr_info("CPU%d is up\n", cpu);
1436 continue;
1437 }
1438 pr_warn("Error taking CPU%d up: %d\n", cpu, error);
1439 }
1440
1441 arch_thaw_secondary_cpus_end();
1442
1443 cpumask_clear(frozen_cpus);
1444out:
1445 cpu_maps_update_done();
1446}
1447
1448static int __init alloc_frozen_cpus(void)
1449{
1450 if (!alloc_cpumask_var(&frozen_cpus, GFP_KERNEL|__GFP_ZERO))
1451 return -ENOMEM;
1452 return 0;
1453}
1454core_initcall(alloc_frozen_cpus);
1455
1456
1457
1458
1459
1460
1461
1462
1463
1464
1465
1466
1467static int
1468cpu_hotplug_pm_callback(struct notifier_block *nb,
1469 unsigned long action, void *ptr)
1470{
1471 switch (action) {
1472
1473 case PM_SUSPEND_PREPARE:
1474 case PM_HIBERNATION_PREPARE:
1475 cpu_hotplug_disable();
1476 break;
1477
1478 case PM_POST_SUSPEND:
1479 case PM_POST_HIBERNATION:
1480 cpu_hotplug_enable();
1481 break;
1482
1483 default:
1484 return NOTIFY_DONE;
1485 }
1486
1487 return NOTIFY_OK;
1488}
1489
1490
1491static int __init cpu_hotplug_pm_sync_init(void)
1492{
1493
1494
1495
1496
1497
1498 pm_notifier(cpu_hotplug_pm_callback, 0);
1499 return 0;
1500}
1501core_initcall(cpu_hotplug_pm_sync_init);
1502
1503#endif
1504
1505int __boot_cpu_id;
1506
1507#endif
1508
1509
1510static struct cpuhp_step cpuhp_hp_states[] = {
1511 [CPUHP_OFFLINE] = {
1512 .name = "offline",
1513 .startup.single = NULL,
1514 .teardown.single = NULL,
1515 },
1516#ifdef CONFIG_SMP
1517 [CPUHP_CREATE_THREADS]= {
1518 .name = "threads:prepare",
1519 .startup.single = smpboot_create_threads,
1520 .teardown.single = NULL,
1521 .cant_stop = true,
1522 },
1523 [CPUHP_PERF_PREPARE] = {
1524 .name = "perf:prepare",
1525 .startup.single = perf_event_init_cpu,
1526 .teardown.single = perf_event_exit_cpu,
1527 },
1528 [CPUHP_WORKQUEUE_PREP] = {
1529 .name = "workqueue:prepare",
1530 .startup.single = workqueue_prepare_cpu,
1531 .teardown.single = NULL,
1532 },
1533 [CPUHP_HRTIMERS_PREPARE] = {
1534 .name = "hrtimers:prepare",
1535 .startup.single = hrtimers_prepare_cpu,
1536 .teardown.single = hrtimers_dead_cpu,
1537 },
1538 [CPUHP_SMPCFD_PREPARE] = {
1539 .name = "smpcfd:prepare",
1540 .startup.single = smpcfd_prepare_cpu,
1541 .teardown.single = smpcfd_dead_cpu,
1542 },
1543 [CPUHP_RELAY_PREPARE] = {
1544 .name = "relay:prepare",
1545 .startup.single = relay_prepare_cpu,
1546 .teardown.single = NULL,
1547 },
1548 [CPUHP_SLAB_PREPARE] = {
1549 .name = "slab:prepare",
1550 .startup.single = slab_prepare_cpu,
1551 .teardown.single = slab_dead_cpu,
1552 },
1553 [CPUHP_RCUTREE_PREP] = {
1554 .name = "RCU/tree:prepare",
1555 .startup.single = rcutree_prepare_cpu,
1556 .teardown.single = rcutree_dead_cpu,
1557 },
1558
1559
1560
1561
1562
1563 [CPUHP_TIMERS_PREPARE] = {
1564 .name = "timers:prepare",
1565 .startup.single = timers_prepare_cpu,
1566 .teardown.single = timers_dead_cpu,
1567 },
1568
1569 [CPUHP_BRINGUP_CPU] = {
1570 .name = "cpu:bringup",
1571 .startup.single = bringup_cpu,
1572 .teardown.single = finish_cpu,
1573 .cant_stop = true,
1574 },
1575
1576 [CPUHP_AP_IDLE_DEAD] = {
1577 .name = "idle:dead",
1578 },
1579
1580
1581
1582
1583 [CPUHP_AP_OFFLINE] = {
1584 .name = "ap:offline",
1585 .cant_stop = true,
1586 },
1587
1588 [CPUHP_AP_SCHED_STARTING] = {
1589 .name = "sched:starting",
1590 .startup.single = sched_cpu_starting,
1591 .teardown.single = sched_cpu_dying,
1592 },
1593 [CPUHP_AP_RCUTREE_DYING] = {
1594 .name = "RCU/tree:dying",
1595 .startup.single = NULL,
1596 .teardown.single = rcutree_dying_cpu,
1597 },
1598 [CPUHP_AP_SMPCFD_DYING] = {
1599 .name = "smpcfd:dying",
1600 .startup.single = NULL,
1601 .teardown.single = smpcfd_dying_cpu,
1602 },
1603
1604
1605 [CPUHP_AP_ONLINE] = {
1606 .name = "ap:online",
1607 },
1608
1609
1610
1611
1612 [CPUHP_TEARDOWN_CPU] = {
1613 .name = "cpu:teardown",
1614 .startup.single = NULL,
1615 .teardown.single = takedown_cpu,
1616 .cant_stop = true,
1617 },
1618
1619 [CPUHP_AP_SCHED_WAIT_EMPTY] = {
1620 .name = "sched:waitempty",
1621 .startup.single = NULL,
1622 .teardown.single = sched_cpu_wait_empty,
1623 },
1624
1625
1626 [CPUHP_AP_SMPBOOT_THREADS] = {
1627 .name = "smpboot/threads:online",
1628 .startup.single = smpboot_unpark_threads,
1629 .teardown.single = smpboot_park_threads,
1630 },
1631 [CPUHP_AP_IRQ_AFFINITY_ONLINE] = {
1632 .name = "irq/affinity:online",
1633 .startup.single = irq_affinity_online_cpu,
1634 .teardown.single = NULL,
1635 },
1636 [CPUHP_AP_PERF_ONLINE] = {
1637 .name = "perf:online",
1638 .startup.single = perf_event_init_cpu,
1639 .teardown.single = perf_event_exit_cpu,
1640 },
1641 [CPUHP_AP_WATCHDOG_ONLINE] = {
1642 .name = "lockup_detector:online",
1643 .startup.single = lockup_detector_online_cpu,
1644 .teardown.single = lockup_detector_offline_cpu,
1645 },
1646 [CPUHP_AP_WORKQUEUE_ONLINE] = {
1647 .name = "workqueue:online",
1648 .startup.single = workqueue_online_cpu,
1649 .teardown.single = workqueue_offline_cpu,
1650 },
1651 [CPUHP_AP_RCUTREE_ONLINE] = {
1652 .name = "RCU/tree:online",
1653 .startup.single = rcutree_online_cpu,
1654 .teardown.single = rcutree_offline_cpu,
1655 },
1656#endif
1657
1658
1659
1660
1661#ifdef CONFIG_SMP
1662
1663 [CPUHP_AP_ACTIVE] = {
1664 .name = "sched:active",
1665 .startup.single = sched_cpu_activate,
1666 .teardown.single = sched_cpu_deactivate,
1667 },
1668#endif
1669
1670
1671 [CPUHP_ONLINE] = {
1672 .name = "online",
1673 .startup.single = NULL,
1674 .teardown.single = NULL,
1675 },
1676};
1677
1678
1679static int cpuhp_cb_check(enum cpuhp_state state)
1680{
1681 if (state <= CPUHP_OFFLINE || state >= CPUHP_ONLINE)
1682 return -EINVAL;
1683 return 0;
1684}
1685
1686
1687
1688
1689
1690
1691static int cpuhp_reserve_state(enum cpuhp_state state)
1692{
1693 enum cpuhp_state i, end;
1694 struct cpuhp_step *step;
1695
1696 switch (state) {
1697 case CPUHP_AP_ONLINE_DYN:
1698 step = cpuhp_hp_states + CPUHP_AP_ONLINE_DYN;
1699 end = CPUHP_AP_ONLINE_DYN_END;
1700 break;
1701 case CPUHP_BP_PREPARE_DYN:
1702 step = cpuhp_hp_states + CPUHP_BP_PREPARE_DYN;
1703 end = CPUHP_BP_PREPARE_DYN_END;
1704 break;
1705 default:
1706 return -EINVAL;
1707 }
1708
1709 for (i = state; i <= end; i++, step++) {
1710 if (!step->name)
1711 return i;
1712 }
1713 WARN(1, "No more dynamic states available for CPU hotplug\n");
1714 return -ENOSPC;
1715}
1716
1717static int cpuhp_store_callbacks(enum cpuhp_state state, const char *name,
1718 int (*startup)(unsigned int cpu),
1719 int (*teardown)(unsigned int cpu),
1720 bool multi_instance)
1721{
1722
1723 struct cpuhp_step *sp;
1724 int ret = 0;
1725
1726
1727
1728
1729
1730
1731
1732
1733
1734
1735 if (name && (state == CPUHP_AP_ONLINE_DYN ||
1736 state == CPUHP_BP_PREPARE_DYN)) {
1737 ret = cpuhp_reserve_state(state);
1738 if (ret < 0)
1739 return ret;
1740 state = ret;
1741 }
1742 sp = cpuhp_get_step(state);
1743 if (name && sp->name)
1744 return -EBUSY;
1745
1746 sp->startup.single = startup;
1747 sp->teardown.single = teardown;
1748 sp->name = name;
1749 sp->multi_instance = multi_instance;
1750 INIT_HLIST_HEAD(&sp->list);
1751 return ret;
1752}
1753
1754static void *cpuhp_get_teardown_cb(enum cpuhp_state state)
1755{
1756 return cpuhp_get_step(state)->teardown.single;
1757}
1758
1759
1760
1761
1762
1763static int cpuhp_issue_call(int cpu, enum cpuhp_state state, bool bringup,
1764 struct hlist_node *node)
1765{
1766 struct cpuhp_step *sp = cpuhp_get_step(state);
1767 int ret;
1768
1769
1770
1771
1772
1773 if ((bringup && !sp->startup.single) ||
1774 (!bringup && !sp->teardown.single))
1775 return 0;
1776
1777
1778
1779
1780#ifdef CONFIG_SMP
1781 if (cpuhp_is_ap_state(state))
1782 ret = cpuhp_invoke_ap_callback(cpu, state, bringup, node);
1783 else
1784 ret = cpuhp_invoke_callback(cpu, state, bringup, node, NULL);
1785#else
1786 ret = cpuhp_invoke_callback(cpu, state, bringup, node, NULL);
1787#endif
1788 BUG_ON(ret && !bringup);
1789 return ret;
1790}
1791
1792
1793
1794
1795
1796
1797static void cpuhp_rollback_install(int failedcpu, enum cpuhp_state state,
1798 struct hlist_node *node)
1799{
1800 int cpu;
1801
1802
1803 for_each_present_cpu(cpu) {
1804 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1805 int cpustate = st->state;
1806
1807 if (cpu >= failedcpu)
1808 break;
1809
1810
1811 if (cpustate >= state)
1812 cpuhp_issue_call(cpu, state, false, node);
1813 }
1814}
1815
1816int __cpuhp_state_add_instance_cpuslocked(enum cpuhp_state state,
1817 struct hlist_node *node,
1818 bool invoke)
1819{
1820 struct cpuhp_step *sp;
1821 int cpu;
1822 int ret;
1823
1824 lockdep_assert_cpus_held();
1825
1826 sp = cpuhp_get_step(state);
1827 if (sp->multi_instance == false)
1828 return -EINVAL;
1829
1830 mutex_lock(&cpuhp_state_mutex);
1831
1832 if (!invoke || !sp->startup.multi)
1833 goto add_node;
1834
1835
1836
1837
1838
1839 for_each_present_cpu(cpu) {
1840 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1841 int cpustate = st->state;
1842
1843 if (cpustate < state)
1844 continue;
1845
1846 ret = cpuhp_issue_call(cpu, state, true, node);
1847 if (ret) {
1848 if (sp->teardown.multi)
1849 cpuhp_rollback_install(cpu, state, node);
1850 goto unlock;
1851 }
1852 }
1853add_node:
1854 ret = 0;
1855 hlist_add_head(node, &sp->list);
1856unlock:
1857 mutex_unlock(&cpuhp_state_mutex);
1858 return ret;
1859}
1860
1861int __cpuhp_state_add_instance(enum cpuhp_state state, struct hlist_node *node,
1862 bool invoke)
1863{
1864 int ret;
1865
1866 cpus_read_lock();
1867 ret = __cpuhp_state_add_instance_cpuslocked(state, node, invoke);
1868 cpus_read_unlock();
1869 return ret;
1870}
1871EXPORT_SYMBOL_GPL(__cpuhp_state_add_instance);
1872
1873
1874
1875
1876
1877
1878
1879
1880
1881
1882
1883
1884
1885
1886
1887
1888
1889
1890int __cpuhp_setup_state_cpuslocked(enum cpuhp_state state,
1891 const char *name, bool invoke,
1892 int (*startup)(unsigned int cpu),
1893 int (*teardown)(unsigned int cpu),
1894 bool multi_instance)
1895{
1896 int cpu, ret = 0;
1897 bool dynstate;
1898
1899 lockdep_assert_cpus_held();
1900
1901 if (cpuhp_cb_check(state) || !name)
1902 return -EINVAL;
1903
1904 mutex_lock(&cpuhp_state_mutex);
1905
1906 ret = cpuhp_store_callbacks(state, name, startup, teardown,
1907 multi_instance);
1908
1909 dynstate = state == CPUHP_AP_ONLINE_DYN;
1910 if (ret > 0 && dynstate) {
1911 state = ret;
1912 ret = 0;
1913 }
1914
1915 if (ret || !invoke || !startup)
1916 goto out;
1917
1918
1919
1920
1921
1922 for_each_present_cpu(cpu) {
1923 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1924 int cpustate = st->state;
1925
1926 if (cpustate < state)
1927 continue;
1928
1929 ret = cpuhp_issue_call(cpu, state, true, NULL);
1930 if (ret) {
1931 if (teardown)
1932 cpuhp_rollback_install(cpu, state, NULL);
1933 cpuhp_store_callbacks(state, NULL, NULL, NULL, false);
1934 goto out;
1935 }
1936 }
1937out:
1938 mutex_unlock(&cpuhp_state_mutex);
1939
1940
1941
1942
1943 if (!ret && dynstate)
1944 return state;
1945 return ret;
1946}
1947EXPORT_SYMBOL(__cpuhp_setup_state_cpuslocked);
1948
1949int __cpuhp_setup_state(enum cpuhp_state state,
1950 const char *name, bool invoke,
1951 int (*startup)(unsigned int cpu),
1952 int (*teardown)(unsigned int cpu),
1953 bool multi_instance)
1954{
1955 int ret;
1956
1957 cpus_read_lock();
1958 ret = __cpuhp_setup_state_cpuslocked(state, name, invoke, startup,
1959 teardown, multi_instance);
1960 cpus_read_unlock();
1961 return ret;
1962}
1963EXPORT_SYMBOL(__cpuhp_setup_state);
1964
1965int __cpuhp_state_remove_instance(enum cpuhp_state state,
1966 struct hlist_node *node, bool invoke)
1967{
1968 struct cpuhp_step *sp = cpuhp_get_step(state);
1969 int cpu;
1970
1971 BUG_ON(cpuhp_cb_check(state));
1972
1973 if (!sp->multi_instance)
1974 return -EINVAL;
1975
1976 cpus_read_lock();
1977 mutex_lock(&cpuhp_state_mutex);
1978
1979 if (!invoke || !cpuhp_get_teardown_cb(state))
1980 goto remove;
1981
1982
1983
1984
1985
1986 for_each_present_cpu(cpu) {
1987 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1988 int cpustate = st->state;
1989
1990 if (cpustate >= state)
1991 cpuhp_issue_call(cpu, state, false, node);
1992 }
1993
1994remove:
1995 hlist_del(node);
1996 mutex_unlock(&cpuhp_state_mutex);
1997 cpus_read_unlock();
1998
1999 return 0;
2000}
2001EXPORT_SYMBOL_GPL(__cpuhp_state_remove_instance);
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
2013void __cpuhp_remove_state_cpuslocked(enum cpuhp_state state, bool invoke)
2014{
2015 struct cpuhp_step *sp = cpuhp_get_step(state);
2016 int cpu;
2017
2018 BUG_ON(cpuhp_cb_check(state));
2019
2020 lockdep_assert_cpus_held();
2021
2022 mutex_lock(&cpuhp_state_mutex);
2023 if (sp->multi_instance) {
2024 WARN(!hlist_empty(&sp->list),
2025 "Error: Removing state %d which has instances left.\n",
2026 state);
2027 goto remove;
2028 }
2029
2030 if (!invoke || !cpuhp_get_teardown_cb(state))
2031 goto remove;
2032
2033
2034
2035
2036
2037
2038 for_each_present_cpu(cpu) {
2039 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
2040 int cpustate = st->state;
2041
2042 if (cpustate >= state)
2043 cpuhp_issue_call(cpu, state, false, NULL);
2044 }
2045remove:
2046 cpuhp_store_callbacks(state, NULL, NULL, NULL, false);
2047 mutex_unlock(&cpuhp_state_mutex);
2048}
2049EXPORT_SYMBOL(__cpuhp_remove_state_cpuslocked);
2050
2051void __cpuhp_remove_state(enum cpuhp_state state, bool invoke)
2052{
2053 cpus_read_lock();
2054 __cpuhp_remove_state_cpuslocked(state, invoke);
2055 cpus_read_unlock();
2056}
2057EXPORT_SYMBOL(__cpuhp_remove_state);
2058
2059#ifdef CONFIG_HOTPLUG_SMT
2060static void cpuhp_offline_cpu_device(unsigned int cpu)
2061{
2062 struct device *dev = get_cpu_device(cpu);
2063
2064 dev->offline = true;
2065
2066 kobject_uevent(&dev->kobj, KOBJ_OFFLINE);
2067}
2068
2069static void cpuhp_online_cpu_device(unsigned int cpu)
2070{
2071 struct device *dev = get_cpu_device(cpu);
2072
2073 dev->offline = false;
2074
2075 kobject_uevent(&dev->kobj, KOBJ_ONLINE);
2076}
2077
2078int cpuhp_smt_disable(enum cpuhp_smt_control ctrlval)
2079{
2080 int cpu, ret = 0;
2081
2082 cpu_maps_update_begin();
2083 for_each_online_cpu(cpu) {
2084 if (topology_is_primary_thread(cpu))
2085 continue;
2086 ret = cpu_down_maps_locked(cpu, CPUHP_OFFLINE);
2087 if (ret)
2088 break;
2089
2090
2091
2092
2093
2094
2095
2096
2097
2098
2099
2100
2101
2102 cpuhp_offline_cpu_device(cpu);
2103 }
2104 if (!ret)
2105 cpu_smt_control = ctrlval;
2106 cpu_maps_update_done();
2107 return ret;
2108}
2109
2110int cpuhp_smt_enable(void)
2111{
2112 int cpu, ret = 0;
2113
2114 cpu_maps_update_begin();
2115 cpu_smt_control = CPU_SMT_ENABLED;
2116 for_each_present_cpu(cpu) {
2117
2118 if (cpu_online(cpu) || !node_online(cpu_to_node(cpu)))
2119 continue;
2120 ret = _cpu_up(cpu, 0, CPUHP_ONLINE);
2121 if (ret)
2122 break;
2123
2124 cpuhp_online_cpu_device(cpu);
2125 }
2126 cpu_maps_update_done();
2127 return ret;
2128}
2129#endif
2130
2131#if defined(CONFIG_SYSFS) && defined(CONFIG_HOTPLUG_CPU)
2132static ssize_t show_cpuhp_state(struct device *dev,
2133 struct device_attribute *attr, char *buf)
2134{
2135 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id);
2136
2137 return sprintf(buf, "%d\n", st->state);
2138}
2139static DEVICE_ATTR(state, 0444, show_cpuhp_state, NULL);
2140
2141static ssize_t write_cpuhp_target(struct device *dev,
2142 struct device_attribute *attr,
2143 const char *buf, size_t count)
2144{
2145 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id);
2146 struct cpuhp_step *sp;
2147 int target, ret;
2148
2149 ret = kstrtoint(buf, 10, &target);
2150 if (ret)
2151 return ret;
2152
2153#ifdef CONFIG_CPU_HOTPLUG_STATE_CONTROL
2154 if (target < CPUHP_OFFLINE || target > CPUHP_ONLINE)
2155 return -EINVAL;
2156#else
2157 if (target != CPUHP_OFFLINE && target != CPUHP_ONLINE)
2158 return -EINVAL;
2159#endif
2160
2161 ret = lock_device_hotplug_sysfs();
2162 if (ret)
2163 return ret;
2164
2165 mutex_lock(&cpuhp_state_mutex);
2166 sp = cpuhp_get_step(target);
2167 ret = !sp->name || sp->cant_stop ? -EINVAL : 0;
2168 mutex_unlock(&cpuhp_state_mutex);
2169 if (ret)
2170 goto out;
2171
2172 if (st->state < target)
2173 ret = cpu_up(dev->id, target);
2174 else
2175 ret = cpu_down(dev->id, target);
2176out:
2177 unlock_device_hotplug();
2178 return ret ? ret : count;
2179}
2180
2181static ssize_t show_cpuhp_target(struct device *dev,
2182 struct device_attribute *attr, char *buf)
2183{
2184 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id);
2185
2186 return sprintf(buf, "%d\n", st->target);
2187}
2188static DEVICE_ATTR(target, 0644, show_cpuhp_target, write_cpuhp_target);
2189
2190
2191static ssize_t write_cpuhp_fail(struct device *dev,
2192 struct device_attribute *attr,
2193 const char *buf, size_t count)
2194{
2195 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id);
2196 struct cpuhp_step *sp;
2197 int fail, ret;
2198
2199 ret = kstrtoint(buf, 10, &fail);
2200 if (ret)
2201 return ret;
2202
2203 if (fail < CPUHP_OFFLINE || fail > CPUHP_ONLINE)
2204 return -EINVAL;
2205
2206
2207
2208
2209 if (cpuhp_is_atomic_state(fail))
2210 return -EINVAL;
2211
2212
2213
2214
2215 mutex_lock(&cpuhp_state_mutex);
2216 sp = cpuhp_get_step(fail);
2217 if (!sp->startup.single && !sp->teardown.single)
2218 ret = -EINVAL;
2219 mutex_unlock(&cpuhp_state_mutex);
2220 if (ret)
2221 return ret;
2222
2223 st->fail = fail;
2224
2225 return count;
2226}
2227
2228static ssize_t show_cpuhp_fail(struct device *dev,
2229 struct device_attribute *attr, char *buf)
2230{
2231 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id);
2232
2233 return sprintf(buf, "%d\n", st->fail);
2234}
2235
2236static DEVICE_ATTR(fail, 0644, show_cpuhp_fail, write_cpuhp_fail);
2237
2238static struct attribute *cpuhp_cpu_attrs[] = {
2239 &dev_attr_state.attr,
2240 &dev_attr_target.attr,
2241 &dev_attr_fail.attr,
2242 NULL
2243};
2244
2245static const struct attribute_group cpuhp_cpu_attr_group = {
2246 .attrs = cpuhp_cpu_attrs,
2247 .name = "hotplug",
2248 NULL
2249};
2250
2251static ssize_t show_cpuhp_states(struct device *dev,
2252 struct device_attribute *attr, char *buf)
2253{
2254 ssize_t cur, res = 0;
2255 int i;
2256
2257 mutex_lock(&cpuhp_state_mutex);
2258 for (i = CPUHP_OFFLINE; i <= CPUHP_ONLINE; i++) {
2259 struct cpuhp_step *sp = cpuhp_get_step(i);
2260
2261 if (sp->name) {
2262 cur = sprintf(buf, "%3d: %s\n", i, sp->name);
2263 buf += cur;
2264 res += cur;
2265 }
2266 }
2267 mutex_unlock(&cpuhp_state_mutex);
2268 return res;
2269}
2270static DEVICE_ATTR(states, 0444, show_cpuhp_states, NULL);
2271
2272static struct attribute *cpuhp_cpu_root_attrs[] = {
2273 &dev_attr_states.attr,
2274 NULL
2275};
2276
2277static const struct attribute_group cpuhp_cpu_root_attr_group = {
2278 .attrs = cpuhp_cpu_root_attrs,
2279 .name = "hotplug",
2280 NULL
2281};
2282
2283#ifdef CONFIG_HOTPLUG_SMT
2284
2285static ssize_t
2286__store_smt_control(struct device *dev, struct device_attribute *attr,
2287 const char *buf, size_t count)
2288{
2289 int ctrlval, ret;
2290
2291 if (sysfs_streq(buf, "on"))
2292 ctrlval = CPU_SMT_ENABLED;
2293 else if (sysfs_streq(buf, "off"))
2294 ctrlval = CPU_SMT_DISABLED;
2295 else if (sysfs_streq(buf, "forceoff"))
2296 ctrlval = CPU_SMT_FORCE_DISABLED;
2297 else
2298 return -EINVAL;
2299
2300 if (cpu_smt_control == CPU_SMT_FORCE_DISABLED)
2301 return -EPERM;
2302
2303 if (cpu_smt_control == CPU_SMT_NOT_SUPPORTED)
2304 return -ENODEV;
2305
2306 ret = lock_device_hotplug_sysfs();
2307 if (ret)
2308 return ret;
2309
2310 if (ctrlval != cpu_smt_control) {
2311 switch (ctrlval) {
2312 case CPU_SMT_ENABLED:
2313 ret = cpuhp_smt_enable();
2314 break;
2315 case CPU_SMT_DISABLED:
2316 case CPU_SMT_FORCE_DISABLED:
2317 ret = cpuhp_smt_disable(ctrlval);
2318 break;
2319 }
2320 }
2321
2322 unlock_device_hotplug();
2323 return ret ? ret : count;
2324}
2325
2326#else
2327static ssize_t
2328__store_smt_control(struct device *dev, struct device_attribute *attr,
2329 const char *buf, size_t count)
2330{
2331 return -ENODEV;
2332}
2333#endif
2334
2335static const char *smt_states[] = {
2336 [CPU_SMT_ENABLED] = "on",
2337 [CPU_SMT_DISABLED] = "off",
2338 [CPU_SMT_FORCE_DISABLED] = "forceoff",
2339 [CPU_SMT_NOT_SUPPORTED] = "notsupported",
2340 [CPU_SMT_NOT_IMPLEMENTED] = "notimplemented",
2341};
2342
2343static ssize_t
2344show_smt_control(struct device *dev, struct device_attribute *attr, char *buf)
2345{
2346 const char *state = smt_states[cpu_smt_control];
2347
2348 return snprintf(buf, PAGE_SIZE - 2, "%s\n", state);
2349}
2350
2351static ssize_t
2352store_smt_control(struct device *dev, struct device_attribute *attr,
2353 const char *buf, size_t count)
2354{
2355 return __store_smt_control(dev, attr, buf, count);
2356}
2357static DEVICE_ATTR(control, 0644, show_smt_control, store_smt_control);
2358
2359static ssize_t
2360show_smt_active(struct device *dev, struct device_attribute *attr, char *buf)
2361{
2362 return snprintf(buf, PAGE_SIZE - 2, "%d\n", sched_smt_active());
2363}
2364static DEVICE_ATTR(active, 0444, show_smt_active, NULL);
2365
2366static struct attribute *cpuhp_smt_attrs[] = {
2367 &dev_attr_control.attr,
2368 &dev_attr_active.attr,
2369 NULL
2370};
2371
2372static const struct attribute_group cpuhp_smt_attr_group = {
2373 .attrs = cpuhp_smt_attrs,
2374 .name = "smt",
2375 NULL
2376};
2377
2378static int __init cpu_smt_sysfs_init(void)
2379{
2380 return sysfs_create_group(&cpu_subsys.dev_root->kobj,
2381 &cpuhp_smt_attr_group);
2382}
2383
2384static int __init cpuhp_sysfs_init(void)
2385{
2386 int cpu, ret;
2387
2388 ret = cpu_smt_sysfs_init();
2389 if (ret)
2390 return ret;
2391
2392 ret = sysfs_create_group(&cpu_subsys.dev_root->kobj,
2393 &cpuhp_cpu_root_attr_group);
2394 if (ret)
2395 return ret;
2396
2397 for_each_possible_cpu(cpu) {
2398 struct device *dev = get_cpu_device(cpu);
2399
2400 if (!dev)
2401 continue;
2402 ret = sysfs_create_group(&dev->kobj, &cpuhp_cpu_attr_group);
2403 if (ret)
2404 return ret;
2405 }
2406 return 0;
2407}
2408device_initcall(cpuhp_sysfs_init);
2409#endif
2410
2411
2412
2413
2414
2415
2416
2417
2418
2419
2420#define MASK_DECLARE_1(x) [x+1][0] = (1UL << (x))
2421#define MASK_DECLARE_2(x) MASK_DECLARE_1(x), MASK_DECLARE_1(x+1)
2422#define MASK_DECLARE_4(x) MASK_DECLARE_2(x), MASK_DECLARE_2(x+2)
2423#define MASK_DECLARE_8(x) MASK_DECLARE_4(x), MASK_DECLARE_4(x+4)
2424
2425const unsigned long cpu_bit_bitmap[BITS_PER_LONG+1][BITS_TO_LONGS(NR_CPUS)] = {
2426
2427 MASK_DECLARE_8(0), MASK_DECLARE_8(8),
2428 MASK_DECLARE_8(16), MASK_DECLARE_8(24),
2429#if BITS_PER_LONG > 32
2430 MASK_DECLARE_8(32), MASK_DECLARE_8(40),
2431 MASK_DECLARE_8(48), MASK_DECLARE_8(56),
2432#endif
2433};
2434EXPORT_SYMBOL_GPL(cpu_bit_bitmap);
2435
2436const DECLARE_BITMAP(cpu_all_bits, NR_CPUS) = CPU_BITS_ALL;
2437EXPORT_SYMBOL(cpu_all_bits);
2438
2439#ifdef CONFIG_INIT_ALL_POSSIBLE
2440struct cpumask __cpu_possible_mask __read_mostly
2441 = {CPU_BITS_ALL};
2442#else
2443struct cpumask __cpu_possible_mask __read_mostly;
2444#endif
2445EXPORT_SYMBOL(__cpu_possible_mask);
2446
2447struct cpumask __cpu_online_mask __read_mostly;
2448EXPORT_SYMBOL(__cpu_online_mask);
2449
2450struct cpumask __cpu_present_mask __read_mostly;
2451EXPORT_SYMBOL(__cpu_present_mask);
2452
2453struct cpumask __cpu_active_mask __read_mostly;
2454EXPORT_SYMBOL(__cpu_active_mask);
2455
2456atomic_t __num_online_cpus __read_mostly;
2457EXPORT_SYMBOL(__num_online_cpus);
2458
2459void init_cpu_present(const struct cpumask *src)
2460{
2461 cpumask_copy(&__cpu_present_mask, src);
2462}
2463
2464void init_cpu_possible(const struct cpumask *src)
2465{
2466 cpumask_copy(&__cpu_possible_mask, src);
2467}
2468
2469void init_cpu_online(const struct cpumask *src)
2470{
2471 cpumask_copy(&__cpu_online_mask, src);
2472}
2473
2474void set_cpu_online(unsigned int cpu, bool online)
2475{
2476
2477
2478
2479
2480
2481
2482
2483
2484
2485
2486 if (online) {
2487 if (!cpumask_test_and_set_cpu(cpu, &__cpu_online_mask))
2488 atomic_inc(&__num_online_cpus);
2489 } else {
2490 if (cpumask_test_and_clear_cpu(cpu, &__cpu_online_mask))
2491 atomic_dec(&__num_online_cpus);
2492 }
2493}
2494
2495
2496
2497
2498void __init boot_cpu_init(void)
2499{
2500 int cpu = smp_processor_id();
2501
2502
2503 set_cpu_online(cpu, true);
2504 set_cpu_active(cpu, true);
2505 set_cpu_present(cpu, true);
2506 set_cpu_possible(cpu, true);
2507
2508#ifdef CONFIG_SMP
2509 __boot_cpu_id = cpu;
2510#endif
2511}
2512
2513
2514
2515
2516void __init boot_cpu_hotplug_init(void)
2517{
2518#ifdef CONFIG_SMP
2519 cpumask_set_cpu(smp_processor_id(), &cpus_booted_once_mask);
2520#endif
2521 this_cpu_write(cpuhp_state.state, CPUHP_ONLINE);
2522}
2523
2524
2525
2526
2527
2528enum cpu_mitigations {
2529 CPU_MITIGATIONS_OFF,
2530 CPU_MITIGATIONS_AUTO,
2531 CPU_MITIGATIONS_AUTO_NOSMT,
2532};
2533
2534static enum cpu_mitigations cpu_mitigations __ro_after_init =
2535 CPU_MITIGATIONS_AUTO;
2536
2537static int __init mitigations_parse_cmdline(char *arg)
2538{
2539 if (!strcmp(arg, "off"))
2540 cpu_mitigations = CPU_MITIGATIONS_OFF;
2541 else if (!strcmp(arg, "auto"))
2542 cpu_mitigations = CPU_MITIGATIONS_AUTO;
2543 else if (!strcmp(arg, "auto,nosmt"))
2544 cpu_mitigations = CPU_MITIGATIONS_AUTO_NOSMT;
2545 else
2546 pr_crit("Unsupported mitigations=%s, system may still be vulnerable\n",
2547 arg);
2548
2549 return 0;
2550}
2551early_param("mitigations", mitigations_parse_cmdline);
2552
2553
2554bool cpu_mitigations_off(void)
2555{
2556 return cpu_mitigations == CPU_MITIGATIONS_OFF;
2557}
2558EXPORT_SYMBOL_GPL(cpu_mitigations_off);
2559
2560
2561bool cpu_mitigations_auto_nosmt(void)
2562{
2563 return cpu_mitigations == CPU_MITIGATIONS_AUTO_NOSMT;
2564}
2565EXPORT_SYMBOL_GPL(cpu_mitigations_auto_nosmt);
2566