• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /* CPU control.
2  * (C) 2001, 2002, 2003, 2004 Rusty Russell
3  *
4  * This code is licenced under the GPL.
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/scs.h>
35 #include <linux/percpu-rwsem.h>
36 #include <linux/cpuset.h>
37 #include <linux/random.h>
38 #include <linux/cc_platform.h>
39 
40 #include <trace/events/power.h>
41 #define CREATE_TRACE_POINTS
42 #include <trace/events/cpuhp.h>
43 
44 #include "smpboot.h"
45 
46 /**
47  * struct cpuhp_cpu_state - Per cpu hotplug state storage
48  * @state:	The current cpu state
49  * @target:	The target state
50  * @fail:	Current CPU hotplug callback state
51  * @thread:	Pointer to the hotplug thread
52  * @should_run:	Thread should execute
53  * @rollback:	Perform a rollback
54  * @single:	Single callback invocation
55  * @bringup:	Single callback bringup or teardown selector
56  * @cpu:	CPU number
57  * @node:	Remote CPU node; for multi-instance, do a
58  *		single entry callback for install/remove
59  * @last:	For multi-instance rollback, remember how far we got
60  * @cb_state:	The state for a single callback (install/uninstall)
61  * @result:	Result of the operation
62  * @done_up:	Signal completion to the issuer of the task for cpu-up
63  * @done_down:	Signal completion to the issuer of the task for cpu-down
64  */
65 struct cpuhp_cpu_state {
66 	enum cpuhp_state	state;
67 	enum cpuhp_state	target;
68 	enum cpuhp_state	fail;
69 #ifdef CONFIG_SMP
70 	struct task_struct	*thread;
71 	bool			should_run;
72 	bool			rollback;
73 	bool			single;
74 	bool			bringup;
75 	struct hlist_node	*node;
76 	struct hlist_node	*last;
77 	enum cpuhp_state	cb_state;
78 	int			result;
79 	struct completion	done_up;
80 	struct completion	done_down;
81 #endif
82 };
83 
84 static DEFINE_PER_CPU(struct cpuhp_cpu_state, cpuhp_state) = {
85 	.fail = CPUHP_INVALID,
86 };
87 
88 #ifdef CONFIG_SMP
89 cpumask_t cpus_booted_once_mask;
90 #endif
91 
92 #if defined(CONFIG_LOCKDEP) && defined(CONFIG_SMP)
93 static struct lockdep_map cpuhp_state_up_map =
94 	STATIC_LOCKDEP_MAP_INIT("cpuhp_state-up", &cpuhp_state_up_map);
95 static struct lockdep_map cpuhp_state_down_map =
96 	STATIC_LOCKDEP_MAP_INIT("cpuhp_state-down", &cpuhp_state_down_map);
97 
98 
cpuhp_lock_acquire(bool bringup)99 static inline void cpuhp_lock_acquire(bool bringup)
100 {
101 	lock_map_acquire(bringup ? &cpuhp_state_up_map : &cpuhp_state_down_map);
102 }
103 
cpuhp_lock_release(bool bringup)104 static inline void cpuhp_lock_release(bool bringup)
105 {
106 	lock_map_release(bringup ? &cpuhp_state_up_map : &cpuhp_state_down_map);
107 }
108 #else
109 
cpuhp_lock_acquire(bool bringup)110 static inline void cpuhp_lock_acquire(bool bringup) { }
cpuhp_lock_release(bool bringup)111 static inline void cpuhp_lock_release(bool bringup) { }
112 
113 #endif
114 
115 /**
116  * struct cpuhp_step - Hotplug state machine step
117  * @name:	Name of the step
118  * @startup:	Startup function of the step
119  * @teardown:	Teardown function of the step
120  * @cant_stop:	Bringup/teardown can't be stopped at this step
121  * @multi_instance:	State has multiple instances which get added afterwards
122  */
123 struct cpuhp_step {
124 	const char		*name;
125 	union {
126 		int		(*single)(unsigned int cpu);
127 		int		(*multi)(unsigned int cpu,
128 					 struct hlist_node *node);
129 	} startup;
130 	union {
131 		int		(*single)(unsigned int cpu);
132 		int		(*multi)(unsigned int cpu,
133 					 struct hlist_node *node);
134 	} teardown;
135 	/* private: */
136 	struct hlist_head	list;
137 	/* public: */
138 	bool			cant_stop;
139 	bool			multi_instance;
140 };
141 
142 static DEFINE_MUTEX(cpuhp_state_mutex);
143 static struct cpuhp_step cpuhp_hp_states[];
144 
cpuhp_get_step(enum cpuhp_state state)145 static struct cpuhp_step *cpuhp_get_step(enum cpuhp_state state)
146 {
147 	return cpuhp_hp_states + state;
148 }
149 
cpuhp_step_empty(bool bringup,struct cpuhp_step * step)150 static bool cpuhp_step_empty(bool bringup, struct cpuhp_step *step)
151 {
152 	return bringup ? !step->startup.single : !step->teardown.single;
153 }
154 
155 /**
156  * cpuhp_invoke_callback - Invoke the callbacks for a given state
157  * @cpu:	The cpu for which the callback should be invoked
158  * @state:	The state to do callbacks for
159  * @bringup:	True if the bringup callback should be invoked
160  * @node:	For multi-instance, do a single entry callback for install/remove
161  * @lastp:	For multi-instance rollback, remember how far we got
162  *
163  * Called from cpu hotplug and from the state register machinery.
164  *
165  * Return: %0 on success or a negative errno code
166  */
cpuhp_invoke_callback(unsigned int cpu,enum cpuhp_state state,bool bringup,struct hlist_node * node,struct hlist_node ** lastp)167 static int cpuhp_invoke_callback(unsigned int cpu, enum cpuhp_state state,
168 				 bool bringup, struct hlist_node *node,
169 				 struct hlist_node **lastp)
170 {
171 	struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
172 	struct cpuhp_step *step = cpuhp_get_step(state);
173 	int (*cbm)(unsigned int cpu, struct hlist_node *node);
174 	int (*cb)(unsigned int cpu);
175 	int ret, cnt;
176 
177 	if (st->fail == state) {
178 		st->fail = CPUHP_INVALID;
179 		return -EAGAIN;
180 	}
181 
182 	if (cpuhp_step_empty(bringup, step)) {
183 		WARN_ON_ONCE(1);
184 		return 0;
185 	}
186 
187 	if (!step->multi_instance) {
188 		WARN_ON_ONCE(lastp && *lastp);
189 		cb = bringup ? step->startup.single : step->teardown.single;
190 
191 		trace_cpuhp_enter(cpu, st->target, state, cb);
192 		ret = cb(cpu);
193 		trace_cpuhp_exit(cpu, st->state, state, ret);
194 		return ret;
195 	}
196 	cbm = bringup ? step->startup.multi : step->teardown.multi;
197 
198 	/* Single invocation for instance add/remove */
199 	if (node) {
200 		WARN_ON_ONCE(lastp && *lastp);
201 		trace_cpuhp_multi_enter(cpu, st->target, state, cbm, node);
202 		ret = cbm(cpu, node);
203 		trace_cpuhp_exit(cpu, st->state, state, ret);
204 		return ret;
205 	}
206 
207 	/* State transition. Invoke on all instances */
208 	cnt = 0;
209 	hlist_for_each(node, &step->list) {
210 		if (lastp && node == *lastp)
211 			break;
212 
213 		trace_cpuhp_multi_enter(cpu, st->target, state, cbm, node);
214 		ret = cbm(cpu, node);
215 		trace_cpuhp_exit(cpu, st->state, state, ret);
216 		if (ret) {
217 			if (!lastp)
218 				goto err;
219 
220 			*lastp = node;
221 			return ret;
222 		}
223 		cnt++;
224 	}
225 	if (lastp)
226 		*lastp = NULL;
227 	return 0;
228 err:
229 	/* Rollback the instances if one failed */
230 	cbm = !bringup ? step->startup.multi : step->teardown.multi;
231 	if (!cbm)
232 		return ret;
233 
234 	hlist_for_each(node, &step->list) {
235 		if (!cnt--)
236 			break;
237 
238 		trace_cpuhp_multi_enter(cpu, st->target, state, cbm, node);
239 		ret = cbm(cpu, node);
240 		trace_cpuhp_exit(cpu, st->state, state, ret);
241 		/*
242 		 * Rollback must not fail,
243 		 */
244 		WARN_ON_ONCE(ret);
245 	}
246 	return ret;
247 }
248 
249 #ifdef CONFIG_SMP
cpuhp_is_ap_state(enum cpuhp_state state)250 static bool cpuhp_is_ap_state(enum cpuhp_state state)
251 {
252 	/*
253 	 * The extra check for CPUHP_TEARDOWN_CPU is only for documentation
254 	 * purposes as that state is handled explicitly in cpu_down.
255 	 */
256 	return state > CPUHP_BRINGUP_CPU && state != CPUHP_TEARDOWN_CPU;
257 }
258 
wait_for_ap_thread(struct cpuhp_cpu_state * st,bool bringup)259 static inline void wait_for_ap_thread(struct cpuhp_cpu_state *st, bool bringup)
260 {
261 	struct completion *done = bringup ? &st->done_up : &st->done_down;
262 	wait_for_completion(done);
263 }
264 
complete_ap_thread(struct cpuhp_cpu_state * st,bool bringup)265 static inline void complete_ap_thread(struct cpuhp_cpu_state *st, bool bringup)
266 {
267 	struct completion *done = bringup ? &st->done_up : &st->done_down;
268 	complete(done);
269 }
270 
271 /*
272  * The former STARTING/DYING states, ran with IRQs disabled and must not fail.
273  */
cpuhp_is_atomic_state(enum cpuhp_state state)274 static bool cpuhp_is_atomic_state(enum cpuhp_state state)
275 {
276 	return CPUHP_AP_IDLE_DEAD <= state && state < CPUHP_AP_ONLINE;
277 }
278 
279 /* Serializes the updates to cpu_online_mask, cpu_present_mask */
280 static DEFINE_MUTEX(cpu_add_remove_lock);
281 bool cpuhp_tasks_frozen;
282 EXPORT_SYMBOL_GPL(cpuhp_tasks_frozen);
283 
284 /*
285  * The following two APIs (cpu_maps_update_begin/done) must be used when
286  * attempting to serialize the updates to cpu_online_mask & cpu_present_mask.
287  */
cpu_maps_update_begin(void)288 void cpu_maps_update_begin(void)
289 {
290 	mutex_lock(&cpu_add_remove_lock);
291 }
292 
cpu_maps_update_done(void)293 void cpu_maps_update_done(void)
294 {
295 	mutex_unlock(&cpu_add_remove_lock);
296 }
297 
298 /*
299  * If set, cpu_up and cpu_down will return -EBUSY and do nothing.
300  * Should always be manipulated under cpu_add_remove_lock
301  */
302 static int cpu_hotplug_disabled;
303 
304 #ifdef CONFIG_HOTPLUG_CPU
305 
306 DEFINE_STATIC_PERCPU_RWSEM(cpu_hotplug_lock);
307 
cpus_read_lock(void)308 void cpus_read_lock(void)
309 {
310 	percpu_down_read(&cpu_hotplug_lock);
311 }
312 EXPORT_SYMBOL_GPL(cpus_read_lock);
313 
cpus_read_trylock(void)314 int cpus_read_trylock(void)
315 {
316 	return percpu_down_read_trylock(&cpu_hotplug_lock);
317 }
318 EXPORT_SYMBOL_GPL(cpus_read_trylock);
319 
cpus_read_unlock(void)320 void cpus_read_unlock(void)
321 {
322 	percpu_up_read(&cpu_hotplug_lock);
323 }
324 EXPORT_SYMBOL_GPL(cpus_read_unlock);
325 
cpus_write_lock(void)326 void cpus_write_lock(void)
327 {
328 	percpu_down_write(&cpu_hotplug_lock);
329 }
330 
cpus_write_unlock(void)331 void cpus_write_unlock(void)
332 {
333 	percpu_up_write(&cpu_hotplug_lock);
334 }
335 
lockdep_assert_cpus_held(void)336 void lockdep_assert_cpus_held(void)
337 {
338 	/*
339 	 * We can't have hotplug operations before userspace starts running,
340 	 * and some init codepaths will knowingly not take the hotplug lock.
341 	 * This is all valid, so mute lockdep until it makes sense to report
342 	 * unheld locks.
343 	 */
344 	if (system_state < SYSTEM_RUNNING)
345 		return;
346 
347 	percpu_rwsem_assert_held(&cpu_hotplug_lock);
348 }
349 
350 #ifdef CONFIG_LOCKDEP
lockdep_is_cpus_held(void)351 int lockdep_is_cpus_held(void)
352 {
353 	return percpu_rwsem_is_held(&cpu_hotplug_lock);
354 }
355 #endif
356 
lockdep_acquire_cpus_lock(void)357 static void lockdep_acquire_cpus_lock(void)
358 {
359 	rwsem_acquire(&cpu_hotplug_lock.dep_map, 0, 0, _THIS_IP_);
360 }
361 
lockdep_release_cpus_lock(void)362 static void lockdep_release_cpus_lock(void)
363 {
364 	rwsem_release(&cpu_hotplug_lock.dep_map, _THIS_IP_);
365 }
366 
367 /*
368  * Wait for currently running CPU hotplug operations to complete (if any) and
369  * disable future CPU hotplug (from sysfs). The 'cpu_add_remove_lock' protects
370  * the 'cpu_hotplug_disabled' flag. The same lock is also acquired by the
371  * hotplug path before performing hotplug operations. So acquiring that lock
372  * guarantees mutual exclusion from any currently running hotplug operations.
373  */
cpu_hotplug_disable(void)374 void cpu_hotplug_disable(void)
375 {
376 	cpu_maps_update_begin();
377 	cpu_hotplug_disabled++;
378 	cpu_maps_update_done();
379 }
380 EXPORT_SYMBOL_GPL(cpu_hotplug_disable);
381 
__cpu_hotplug_enable(void)382 static void __cpu_hotplug_enable(void)
383 {
384 	if (WARN_ONCE(!cpu_hotplug_disabled, "Unbalanced cpu hotplug enable\n"))
385 		return;
386 	cpu_hotplug_disabled--;
387 }
388 
cpu_hotplug_enable(void)389 void cpu_hotplug_enable(void)
390 {
391 	cpu_maps_update_begin();
392 	__cpu_hotplug_enable();
393 	cpu_maps_update_done();
394 }
395 EXPORT_SYMBOL_GPL(cpu_hotplug_enable);
396 
397 #else
398 
lockdep_acquire_cpus_lock(void)399 static void lockdep_acquire_cpus_lock(void)
400 {
401 }
402 
lockdep_release_cpus_lock(void)403 static void lockdep_release_cpus_lock(void)
404 {
405 }
406 
407 #endif	/* CONFIG_HOTPLUG_CPU */
408 
409 /*
410  * Architectures that need SMT-specific errata handling during SMT hotplug
411  * should override this.
412  */
arch_smt_update(void)413 void __weak arch_smt_update(void) { }
414 
415 #ifdef CONFIG_HOTPLUG_SMT
416 enum cpuhp_smt_control cpu_smt_control __read_mostly = CPU_SMT_ENABLED;
417 
cpu_smt_disable(bool force)418 void __init cpu_smt_disable(bool force)
419 {
420 	if (!cpu_smt_possible())
421 		return;
422 
423 	if (force) {
424 		pr_info("SMT: Force disabled\n");
425 		cpu_smt_control = CPU_SMT_FORCE_DISABLED;
426 	} else {
427 		pr_info("SMT: disabled\n");
428 		cpu_smt_control = CPU_SMT_DISABLED;
429 	}
430 }
431 
432 /*
433  * The decision whether SMT is supported can only be done after the full
434  * CPU identification. Called from architecture code.
435  */
cpu_smt_check_topology(void)436 void __init cpu_smt_check_topology(void)
437 {
438 	if (!topology_smt_supported())
439 		cpu_smt_control = CPU_SMT_NOT_SUPPORTED;
440 }
441 
smt_cmdline_disable(char * str)442 static int __init smt_cmdline_disable(char *str)
443 {
444 	cpu_smt_disable(str && !strcmp(str, "force"));
445 	return 0;
446 }
447 early_param("nosmt", smt_cmdline_disable);
448 
449 /*
450  * For Archicture supporting partial SMT states check if the thread is allowed.
451  * Otherwise this has already been checked through cpu_smt_max_threads when
452  * setting the SMT level.
453  */
cpu_smt_thread_allowed(unsigned int cpu)454 static inline bool cpu_smt_thread_allowed(unsigned int cpu)
455 {
456 #ifdef CONFIG_SMT_NUM_THREADS_DYNAMIC
457 	return topology_smt_thread_allowed(cpu);
458 #else
459 	return true;
460 #endif
461 }
462 
cpu_bootable(unsigned int cpu)463 static inline bool cpu_bootable(unsigned int cpu)
464 {
465 	if (cpu_smt_control == CPU_SMT_ENABLED && cpu_smt_thread_allowed(cpu))
466 		return true;
467 
468 	/* All CPUs are bootable if controls are not configured */
469 	if (cpu_smt_control == CPU_SMT_NOT_IMPLEMENTED)
470 		return true;
471 
472 	/* All CPUs are bootable if CPU is not SMT capable */
473 	if (cpu_smt_control == CPU_SMT_NOT_SUPPORTED)
474 		return true;
475 
476 	if (topology_is_primary_thread(cpu))
477 		return true;
478 
479 	/*
480 	 * On x86 it's required to boot all logical CPUs at least once so
481 	 * that the init code can get a chance to set CR4.MCE on each
482 	 * CPU. Otherwise, a broadcasted MCE observing CR4.MCE=0b on any
483 	 * core will shutdown the machine.
484 	 */
485 	return !cpumask_test_cpu(cpu, &cpus_booted_once_mask);
486 }
487 
488 /* Returns true if SMT is not supported of forcefully (irreversibly) disabled */
cpu_smt_possible(void)489 bool cpu_smt_possible(void)
490 {
491 	return cpu_smt_control != CPU_SMT_FORCE_DISABLED &&
492 		cpu_smt_control != CPU_SMT_NOT_SUPPORTED;
493 }
494 EXPORT_SYMBOL_GPL(cpu_smt_possible);
495 #else
cpu_bootable(unsigned int cpu)496 static inline bool cpu_bootable(unsigned int cpu) { return true; }
497 #endif
498 
499 static inline enum cpuhp_state
cpuhp_set_state(int cpu,struct cpuhp_cpu_state * st,enum cpuhp_state target)500 cpuhp_set_state(int cpu, struct cpuhp_cpu_state *st, enum cpuhp_state target)
501 {
502 	enum cpuhp_state prev_state = st->state;
503 	bool bringup = st->state < target;
504 
505 	st->rollback = false;
506 	st->last = NULL;
507 
508 	st->target = target;
509 	st->single = false;
510 	st->bringup = bringup;
511 	if (cpu_dying(cpu) != !bringup)
512 		set_cpu_dying(cpu, !bringup);
513 
514 	return prev_state;
515 }
516 
517 static inline void
cpuhp_reset_state(int cpu,struct cpuhp_cpu_state * st,enum cpuhp_state prev_state)518 cpuhp_reset_state(int cpu, struct cpuhp_cpu_state *st,
519 		  enum cpuhp_state prev_state)
520 {
521 	bool bringup = !st->bringup;
522 
523 	st->target = prev_state;
524 
525 	/*
526 	 * Already rolling back. No need invert the bringup value or to change
527 	 * the current state.
528 	 */
529 	if (st->rollback)
530 		return;
531 
532 	st->rollback = true;
533 
534 	/*
535 	 * If we have st->last we need to undo partial multi_instance of this
536 	 * state first. Otherwise start undo at the previous state.
537 	 */
538 	if (!st->last) {
539 		if (st->bringup)
540 			st->state--;
541 		else
542 			st->state++;
543 	}
544 
545 	st->bringup = bringup;
546 	if (cpu_dying(cpu) != !bringup)
547 		set_cpu_dying(cpu, !bringup);
548 }
549 
550 /* Regular hotplug invocation of the AP hotplug thread */
__cpuhp_kick_ap(struct cpuhp_cpu_state * st)551 static void __cpuhp_kick_ap(struct cpuhp_cpu_state *st)
552 {
553 	if (!st->single && st->state == st->target)
554 		return;
555 
556 	st->result = 0;
557 	/*
558 	 * Make sure the above stores are visible before should_run becomes
559 	 * true. Paired with the mb() above in cpuhp_thread_fun()
560 	 */
561 	smp_mb();
562 	st->should_run = true;
563 	wake_up_process(st->thread);
564 	wait_for_ap_thread(st, st->bringup);
565 }
566 
cpuhp_kick_ap(int cpu,struct cpuhp_cpu_state * st,enum cpuhp_state target)567 static int cpuhp_kick_ap(int cpu, struct cpuhp_cpu_state *st,
568 			 enum cpuhp_state target)
569 {
570 	enum cpuhp_state prev_state;
571 	int ret;
572 
573 	prev_state = cpuhp_set_state(cpu, st, target);
574 	__cpuhp_kick_ap(st);
575 	if ((ret = st->result)) {
576 		cpuhp_reset_state(cpu, st, prev_state);
577 		__cpuhp_kick_ap(st);
578 	}
579 
580 	return ret;
581 }
582 
bringup_wait_for_ap(unsigned int cpu)583 static int bringup_wait_for_ap(unsigned int cpu)
584 {
585 	struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
586 
587 	/* Wait for the CPU to reach CPUHP_AP_ONLINE_IDLE */
588 	wait_for_ap_thread(st, true);
589 	if (WARN_ON_ONCE((!cpu_online(cpu))))
590 		return -ECANCELED;
591 
592 	/* Unpark the hotplug thread of the target cpu */
593 	kthread_unpark(st->thread);
594 
595 	/*
596 	 * SMT soft disabling on X86 requires to bring the CPU out of the
597 	 * BIOS 'wait for SIPI' state in order to set the CR4.MCE bit.  The
598 	 * CPU marked itself as booted_once in notify_cpu_starting() so the
599 	 * cpu_bootable() check will now return false if this is not the
600 	 * primary sibling.
601 	 */
602 	if (!cpu_bootable(cpu))
603 		return -ECANCELED;
604 
605 	if (st->target <= CPUHP_AP_ONLINE_IDLE)
606 		return 0;
607 
608 	return cpuhp_kick_ap(cpu, st, st->target);
609 }
610 
bringup_cpu(unsigned int cpu)611 static int bringup_cpu(unsigned int cpu)
612 {
613 	struct task_struct *idle = idle_thread_get(cpu);
614 	int ret;
615 
616 	/*
617 	 * Reset stale stack state from the last time this CPU was online.
618 	 */
619 	scs_task_reset(idle);
620 	kasan_unpoison_task_stack(idle);
621 
622 	/*
623 	 * Some architectures have to walk the irq descriptors to
624 	 * setup the vector space for the cpu which comes online.
625 	 * Prevent irq alloc/free across the bringup.
626 	 */
627 	irq_lock_sparse();
628 
629 	/* Arch-specific enabling code. */
630 	ret = __cpu_up(cpu, idle);
631 	irq_unlock_sparse();
632 	if (ret)
633 		return ret;
634 	return bringup_wait_for_ap(cpu);
635 }
636 
finish_cpu(unsigned int cpu)637 static int finish_cpu(unsigned int cpu)
638 {
639 	struct task_struct *idle = idle_thread_get(cpu);
640 	struct mm_struct *mm = idle->active_mm;
641 
642 	/*
643 	 * idle_task_exit() will have switched to &init_mm, now
644 	 * clean up any remaining active_mm state.
645 	 */
646 	if (mm != &init_mm)
647 		idle->active_mm = &init_mm;
648 	mmdrop(mm);
649 	return 0;
650 }
651 
652 /*
653  * Hotplug state machine related functions
654  */
655 
656 /*
657  * Get the next state to run. Empty ones will be skipped. Returns true if a
658  * state must be run.
659  *
660  * st->state will be modified ahead of time, to match state_to_run, as if it
661  * has already ran.
662  */
cpuhp_next_state(bool bringup,enum cpuhp_state * state_to_run,struct cpuhp_cpu_state * st,enum cpuhp_state target)663 static bool cpuhp_next_state(bool bringup,
664 			     enum cpuhp_state *state_to_run,
665 			     struct cpuhp_cpu_state *st,
666 			     enum cpuhp_state target)
667 {
668 	do {
669 		if (bringup) {
670 			if (st->state >= target)
671 				return false;
672 
673 			*state_to_run = ++st->state;
674 		} else {
675 			if (st->state <= target)
676 				return false;
677 
678 			*state_to_run = st->state--;
679 		}
680 
681 		if (!cpuhp_step_empty(bringup, cpuhp_get_step(*state_to_run)))
682 			break;
683 	} while (true);
684 
685 	return true;
686 }
687 
__cpuhp_invoke_callback_range(bool bringup,unsigned int cpu,struct cpuhp_cpu_state * st,enum cpuhp_state target,bool nofail)688 static int __cpuhp_invoke_callback_range(bool bringup,
689 					 unsigned int cpu,
690 					 struct cpuhp_cpu_state *st,
691 					 enum cpuhp_state target,
692 					 bool nofail)
693 {
694 	enum cpuhp_state state;
695 	int ret = 0;
696 
697 	while (cpuhp_next_state(bringup, &state, st, target)) {
698 		int err;
699 
700 		err = cpuhp_invoke_callback(cpu, state, bringup, NULL, NULL);
701 		if (!err)
702 			continue;
703 
704 		if (nofail) {
705 			pr_warn("CPU %u %s state %s (%d) failed (%d)\n",
706 				cpu, bringup ? "UP" : "DOWN",
707 				cpuhp_get_step(st->state)->name,
708 				st->state, err);
709 			ret = -1;
710 		} else {
711 			ret = err;
712 			break;
713 		}
714 	}
715 
716 	return ret;
717 }
718 
cpuhp_invoke_callback_range(bool bringup,unsigned int cpu,struct cpuhp_cpu_state * st,enum cpuhp_state target)719 static inline int cpuhp_invoke_callback_range(bool bringup,
720 					      unsigned int cpu,
721 					      struct cpuhp_cpu_state *st,
722 					      enum cpuhp_state target)
723 {
724 	return __cpuhp_invoke_callback_range(bringup, cpu, st, target, false);
725 }
726 
cpuhp_invoke_callback_range_nofail(bool bringup,unsigned int cpu,struct cpuhp_cpu_state * st,enum cpuhp_state target)727 static inline void cpuhp_invoke_callback_range_nofail(bool bringup,
728 						      unsigned int cpu,
729 						      struct cpuhp_cpu_state *st,
730 						      enum cpuhp_state target)
731 {
732 	__cpuhp_invoke_callback_range(bringup, cpu, st, target, true);
733 }
734 
can_rollback_cpu(struct cpuhp_cpu_state * st)735 static inline bool can_rollback_cpu(struct cpuhp_cpu_state *st)
736 {
737 	if (IS_ENABLED(CONFIG_HOTPLUG_CPU))
738 		return true;
739 	/*
740 	 * When CPU hotplug is disabled, then taking the CPU down is not
741 	 * possible because takedown_cpu() and the architecture and
742 	 * subsystem specific mechanisms are not available. So the CPU
743 	 * which would be completely unplugged again needs to stay around
744 	 * in the current state.
745 	 */
746 	return st->state <= CPUHP_BRINGUP_CPU;
747 }
748 
cpuhp_up_callbacks(unsigned int cpu,struct cpuhp_cpu_state * st,enum cpuhp_state target)749 static int cpuhp_up_callbacks(unsigned int cpu, struct cpuhp_cpu_state *st,
750 			      enum cpuhp_state target)
751 {
752 	enum cpuhp_state prev_state = st->state;
753 	int ret = 0;
754 
755 	ret = cpuhp_invoke_callback_range(true, cpu, st, target);
756 	if (ret) {
757 		pr_debug("CPU UP failed (%d) CPU %u state %s (%d)\n",
758 			 ret, cpu, cpuhp_get_step(st->state)->name,
759 			 st->state);
760 
761 		cpuhp_reset_state(cpu, st, prev_state);
762 		if (can_rollback_cpu(st))
763 			WARN_ON(cpuhp_invoke_callback_range(false, cpu, st,
764 							    prev_state));
765 	}
766 	return ret;
767 }
768 
769 /*
770  * The cpu hotplug threads manage the bringup and teardown of the cpus
771  */
cpuhp_should_run(unsigned int cpu)772 static int cpuhp_should_run(unsigned int cpu)
773 {
774 	struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
775 
776 	return st->should_run;
777 }
778 
779 /*
780  * Execute teardown/startup callbacks on the plugged cpu. Also used to invoke
781  * callbacks when a state gets [un]installed at runtime.
782  *
783  * Each invocation of this function by the smpboot thread does a single AP
784  * state callback.
785  *
786  * It has 3 modes of operation:
787  *  - single: runs st->cb_state
788  *  - up:     runs ++st->state, while st->state < st->target
789  *  - down:   runs st->state--, while st->state > st->target
790  *
791  * When complete or on error, should_run is cleared and the completion is fired.
792  */
cpuhp_thread_fun(unsigned int cpu)793 static void cpuhp_thread_fun(unsigned int cpu)
794 {
795 	struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
796 	bool bringup = st->bringup;
797 	enum cpuhp_state state;
798 
799 	if (WARN_ON_ONCE(!st->should_run))
800 		return;
801 
802 	/*
803 	 * ACQUIRE for the cpuhp_should_run() load of ->should_run. Ensures
804 	 * that if we see ->should_run we also see the rest of the state.
805 	 */
806 	smp_mb();
807 
808 	/*
809 	 * The BP holds the hotplug lock, but we're now running on the AP,
810 	 * ensure that anybody asserting the lock is held, will actually find
811 	 * it so.
812 	 */
813 	lockdep_acquire_cpus_lock();
814 	cpuhp_lock_acquire(bringup);
815 
816 	if (st->single) {
817 		state = st->cb_state;
818 		st->should_run = false;
819 	} else {
820 		st->should_run = cpuhp_next_state(bringup, &state, st, st->target);
821 		if (!st->should_run)
822 			goto end;
823 	}
824 
825 	WARN_ON_ONCE(!cpuhp_is_ap_state(state));
826 
827 	if (cpuhp_is_atomic_state(state)) {
828 		local_irq_disable();
829 		st->result = cpuhp_invoke_callback(cpu, state, bringup, st->node, &st->last);
830 		local_irq_enable();
831 
832 		/*
833 		 * STARTING/DYING must not fail!
834 		 */
835 		WARN_ON_ONCE(st->result);
836 	} else {
837 		st->result = cpuhp_invoke_callback(cpu, state, bringup, st->node, &st->last);
838 	}
839 
840 	if (st->result) {
841 		/*
842 		 * If we fail on a rollback, we're up a creek without no
843 		 * paddle, no way forward, no way back. We loose, thanks for
844 		 * playing.
845 		 */
846 		WARN_ON_ONCE(st->rollback);
847 		st->should_run = false;
848 	}
849 
850 end:
851 	cpuhp_lock_release(bringup);
852 	lockdep_release_cpus_lock();
853 
854 	if (!st->should_run)
855 		complete_ap_thread(st, bringup);
856 }
857 
858 /* Invoke a single callback on a remote cpu */
859 static int
cpuhp_invoke_ap_callback(int cpu,enum cpuhp_state state,bool bringup,struct hlist_node * node)860 cpuhp_invoke_ap_callback(int cpu, enum cpuhp_state state, bool bringup,
861 			 struct hlist_node *node)
862 {
863 	struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
864 	int ret;
865 
866 	if (!cpu_online(cpu))
867 		return 0;
868 
869 	cpuhp_lock_acquire(false);
870 	cpuhp_lock_release(false);
871 
872 	cpuhp_lock_acquire(true);
873 	cpuhp_lock_release(true);
874 
875 	/*
876 	 * If we are up and running, use the hotplug thread. For early calls
877 	 * we invoke the thread function directly.
878 	 */
879 	if (!st->thread)
880 		return cpuhp_invoke_callback(cpu, state, bringup, node, NULL);
881 
882 	st->rollback = false;
883 	st->last = NULL;
884 
885 	st->node = node;
886 	st->bringup = bringup;
887 	st->cb_state = state;
888 	st->single = true;
889 
890 	__cpuhp_kick_ap(st);
891 
892 	/*
893 	 * If we failed and did a partial, do a rollback.
894 	 */
895 	if ((ret = st->result) && st->last) {
896 		st->rollback = true;
897 		st->bringup = !bringup;
898 
899 		__cpuhp_kick_ap(st);
900 	}
901 
902 	/*
903 	 * Clean up the leftovers so the next hotplug operation wont use stale
904 	 * data.
905 	 */
906 	st->node = st->last = NULL;
907 	return ret;
908 }
909 
cpuhp_kick_ap_work(unsigned int cpu)910 static int cpuhp_kick_ap_work(unsigned int cpu)
911 {
912 	struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
913 	enum cpuhp_state prev_state = st->state;
914 	int ret;
915 
916 	cpuhp_lock_acquire(false);
917 	cpuhp_lock_release(false);
918 
919 	cpuhp_lock_acquire(true);
920 	cpuhp_lock_release(true);
921 
922 	trace_cpuhp_enter(cpu, st->target, prev_state, cpuhp_kick_ap_work);
923 	ret = cpuhp_kick_ap(cpu, st, st->target);
924 	trace_cpuhp_exit(cpu, st->state, prev_state, ret);
925 
926 	return ret;
927 }
928 
929 static struct smp_hotplug_thread cpuhp_threads = {
930 	.store			= &cpuhp_state.thread,
931 	.thread_should_run	= cpuhp_should_run,
932 	.thread_fn		= cpuhp_thread_fun,
933 	.thread_comm		= "cpuhp/%u",
934 	.selfparking		= true,
935 };
936 
cpuhp_init_state(void)937 static __init void cpuhp_init_state(void)
938 {
939 	struct cpuhp_cpu_state *st;
940 	int cpu;
941 
942 	for_each_possible_cpu(cpu) {
943 		st = per_cpu_ptr(&cpuhp_state, cpu);
944 		init_completion(&st->done_up);
945 		init_completion(&st->done_down);
946 	}
947 }
948 
cpuhp_threads_init(void)949 void __init cpuhp_threads_init(void)
950 {
951 	cpuhp_init_state();
952 	BUG_ON(smpboot_register_percpu_thread(&cpuhp_threads));
953 	kthread_unpark(this_cpu_read(cpuhp_state.thread));
954 }
955 
956 /*
957  *
958  * Serialize hotplug trainwrecks outside of the cpu_hotplug_lock
959  * protected region.
960  *
961  * The operation is still serialized against concurrent CPU hotplug via
962  * cpu_add_remove_lock, i.e. CPU map protection.  But it is _not_
963  * serialized against other hotplug related activity like adding or
964  * removing of state callbacks and state instances, which invoke either the
965  * startup or the teardown callback of the affected state.
966  *
967  * This is required for subsystems which are unfixable vs. CPU hotplug and
968  * evade lock inversion problems by scheduling work which has to be
969  * completed _before_ cpu_up()/_cpu_down() returns.
970  *
971  * Don't even think about adding anything to this for any new code or even
972  * drivers. It's only purpose is to keep existing lock order trainwrecks
973  * working.
974  *
975  * For cpu_down() there might be valid reasons to finish cleanups which are
976  * not required to be done under cpu_hotplug_lock, but that's a different
977  * story and would be not invoked via this.
978  */
cpu_up_down_serialize_trainwrecks(bool tasks_frozen)979 static void cpu_up_down_serialize_trainwrecks(bool tasks_frozen)
980 {
981 	/*
982 	 * cpusets delegate hotplug operations to a worker to "solve" the
983 	 * lock order problems. Wait for the worker, but only if tasks are
984 	 * _not_ frozen (suspend, hibernate) as that would wait forever.
985 	 *
986 	 * The wait is required because otherwise the hotplug operation
987 	 * returns with inconsistent state, which could even be observed in
988 	 * user space when a new CPU is brought up. The CPU plug uevent
989 	 * would be delivered and user space reacting on it would fail to
990 	 * move tasks to the newly plugged CPU up to the point where the
991 	 * work has finished because up to that point the newly plugged CPU
992 	 * is not assignable in cpusets/cgroups. On unplug that's not
993 	 * necessarily a visible issue, but it is still inconsistent state,
994 	 * which is the real problem which needs to be "fixed". This can't
995 	 * prevent the transient state between scheduling the work and
996 	 * returning from waiting for it.
997 	 */
998 	if (!tasks_frozen)
999 		cpuset_wait_for_hotplug();
1000 }
1001 
1002 #ifdef CONFIG_HOTPLUG_CPU
1003 #ifndef arch_clear_mm_cpumask_cpu
1004 #define arch_clear_mm_cpumask_cpu(cpu, mm) cpumask_clear_cpu(cpu, mm_cpumask(mm))
1005 #endif
1006 
1007 /**
1008  * clear_tasks_mm_cpumask - Safely clear tasks' mm_cpumask for a CPU
1009  * @cpu: a CPU id
1010  *
1011  * This function walks all processes, finds a valid mm struct for each one and
1012  * then clears a corresponding bit in mm's cpumask.  While this all sounds
1013  * trivial, there are various non-obvious corner cases, which this function
1014  * tries to solve in a safe manner.
1015  *
1016  * Also note that the function uses a somewhat relaxed locking scheme, so it may
1017  * be called only for an already offlined CPU.
1018  */
clear_tasks_mm_cpumask(int cpu)1019 void clear_tasks_mm_cpumask(int cpu)
1020 {
1021 	struct task_struct *p;
1022 
1023 	/*
1024 	 * This function is called after the cpu is taken down and marked
1025 	 * offline, so its not like new tasks will ever get this cpu set in
1026 	 * their mm mask. -- Peter Zijlstra
1027 	 * Thus, we may use rcu_read_lock() here, instead of grabbing
1028 	 * full-fledged tasklist_lock.
1029 	 */
1030 	WARN_ON(cpu_online(cpu));
1031 	rcu_read_lock();
1032 	for_each_process(p) {
1033 		struct task_struct *t;
1034 
1035 		/*
1036 		 * Main thread might exit, but other threads may still have
1037 		 * a valid mm. Find one.
1038 		 */
1039 		t = find_lock_task_mm(p);
1040 		if (!t)
1041 			continue;
1042 		arch_clear_mm_cpumask_cpu(cpu, t->mm);
1043 		task_unlock(t);
1044 	}
1045 	rcu_read_unlock();
1046 }
1047 
1048 /* Take this CPU down. */
take_cpu_down(void * _param)1049 static int take_cpu_down(void *_param)
1050 {
1051 	struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
1052 	enum cpuhp_state target = max((int)st->target, CPUHP_AP_OFFLINE);
1053 	int err, cpu = smp_processor_id();
1054 
1055 	/* Ensure this CPU doesn't handle any more interrupts. */
1056 	err = __cpu_disable();
1057 	if (err < 0)
1058 		return err;
1059 
1060 	/*
1061 	 * Must be called from CPUHP_TEARDOWN_CPU, which means, as we are going
1062 	 * down, that the current state is CPUHP_TEARDOWN_CPU - 1.
1063 	 */
1064 	WARN_ON(st->state != (CPUHP_TEARDOWN_CPU - 1));
1065 
1066 	/*
1067 	 * Invoke the former CPU_DYING callbacks. DYING must not fail!
1068 	 */
1069 	cpuhp_invoke_callback_range_nofail(false, cpu, st, target);
1070 
1071 	/* Give up timekeeping duties */
1072 	tick_handover_do_timer();
1073 	/* Remove CPU from timer broadcasting */
1074 	tick_offline_cpu(cpu);
1075 	/* Park the stopper thread */
1076 	stop_machine_park(cpu);
1077 	return 0;
1078 }
1079 
takedown_cpu(unsigned int cpu)1080 static int takedown_cpu(unsigned int cpu)
1081 {
1082 	struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1083 	int err;
1084 
1085 	/* Park the smpboot threads */
1086 	kthread_park(st->thread);
1087 
1088 	/*
1089 	 * Prevent irq alloc/free while the dying cpu reorganizes the
1090 	 * interrupt affinities.
1091 	 */
1092 	irq_lock_sparse();
1093 
1094 	/*
1095 	 * So now all preempt/rcu users must observe !cpu_active().
1096 	 */
1097 	err = stop_machine_cpuslocked(take_cpu_down, NULL, cpumask_of(cpu));
1098 	if (err) {
1099 		/* CPU refused to die */
1100 		irq_unlock_sparse();
1101 		/* Unpark the hotplug thread so we can rollback there */
1102 		kthread_unpark(st->thread);
1103 		return err;
1104 	}
1105 	BUG_ON(cpu_online(cpu));
1106 
1107 	/*
1108 	 * The teardown callback for CPUHP_AP_SCHED_STARTING will have removed
1109 	 * all runnable tasks from the CPU, there's only the idle task left now
1110 	 * that the migration thread is done doing the stop_machine thing.
1111 	 *
1112 	 * Wait for the stop thread to go away.
1113 	 */
1114 	wait_for_ap_thread(st, false);
1115 	BUG_ON(st->state != CPUHP_AP_IDLE_DEAD);
1116 
1117 	/* Interrupts are moved away from the dying cpu, reenable alloc/free */
1118 	irq_unlock_sparse();
1119 
1120 	hotplug_cpu__broadcast_tick_pull(cpu);
1121 	/* This actually kills the CPU. */
1122 	__cpu_die(cpu);
1123 
1124 	tick_cleanup_dead_cpu(cpu);
1125 	rcutree_migrate_callbacks(cpu);
1126 	return 0;
1127 }
1128 
cpuhp_complete_idle_dead(void * arg)1129 static void cpuhp_complete_idle_dead(void *arg)
1130 {
1131 	struct cpuhp_cpu_state *st = arg;
1132 
1133 	complete_ap_thread(st, false);
1134 }
1135 
cpuhp_report_idle_dead(void)1136 void cpuhp_report_idle_dead(void)
1137 {
1138 	struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
1139 
1140 	BUG_ON(st->state != CPUHP_AP_OFFLINE);
1141 	rcu_report_dead(smp_processor_id());
1142 	st->state = CPUHP_AP_IDLE_DEAD;
1143 	/*
1144 	 * We cannot call complete after rcu_report_dead() so we delegate it
1145 	 * to an online cpu.
1146 	 */
1147 	smp_call_function_single(cpumask_first(cpu_online_mask),
1148 				 cpuhp_complete_idle_dead, st, 0);
1149 }
1150 
cpuhp_down_callbacks(unsigned int cpu,struct cpuhp_cpu_state * st,enum cpuhp_state target)1151 static int cpuhp_down_callbacks(unsigned int cpu, struct cpuhp_cpu_state *st,
1152 				enum cpuhp_state target)
1153 {
1154 	enum cpuhp_state prev_state = st->state;
1155 	int ret = 0;
1156 
1157 	ret = cpuhp_invoke_callback_range(false, cpu, st, target);
1158 	if (ret) {
1159 		pr_debug("CPU DOWN failed (%d) CPU %u state %s (%d)\n",
1160 			 ret, cpu, cpuhp_get_step(st->state)->name,
1161 			 st->state);
1162 
1163 		cpuhp_reset_state(cpu, st, prev_state);
1164 
1165 		if (st->state < prev_state)
1166 			WARN_ON(cpuhp_invoke_callback_range(true, cpu, st,
1167 							    prev_state));
1168 	}
1169 
1170 	return ret;
1171 }
1172 
1173 /* Requires cpu_add_remove_lock to be held */
_cpu_down(unsigned int cpu,int tasks_frozen,enum cpuhp_state target)1174 static int __ref _cpu_down(unsigned int cpu, int tasks_frozen,
1175 			   enum cpuhp_state target)
1176 {
1177 	struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1178 	int prev_state, ret = 0;
1179 
1180 	if (num_online_cpus() == 1)
1181 		return -EBUSY;
1182 
1183 	if (!cpu_present(cpu))
1184 		return -EINVAL;
1185 
1186 	cpus_write_lock();
1187 
1188 	cpuhp_tasks_frozen = tasks_frozen;
1189 
1190 	prev_state = cpuhp_set_state(cpu, st, target);
1191 	/*
1192 	 * If the current CPU state is in the range of the AP hotplug thread,
1193 	 * then we need to kick the thread.
1194 	 */
1195 	if (st->state > CPUHP_TEARDOWN_CPU) {
1196 		st->target = max((int)target, CPUHP_TEARDOWN_CPU);
1197 		ret = cpuhp_kick_ap_work(cpu);
1198 		/*
1199 		 * The AP side has done the error rollback already. Just
1200 		 * return the error code..
1201 		 */
1202 		if (ret)
1203 			goto out;
1204 
1205 		/*
1206 		 * We might have stopped still in the range of the AP hotplug
1207 		 * thread. Nothing to do anymore.
1208 		 */
1209 		if (st->state > CPUHP_TEARDOWN_CPU)
1210 			goto out;
1211 
1212 		st->target = target;
1213 	}
1214 	/*
1215 	 * The AP brought itself down to CPUHP_TEARDOWN_CPU. So we need
1216 	 * to do the further cleanups.
1217 	 */
1218 	ret = cpuhp_down_callbacks(cpu, st, target);
1219 	if (ret && st->state < prev_state) {
1220 		if (st->state == CPUHP_TEARDOWN_CPU) {
1221 			cpuhp_reset_state(cpu, st, prev_state);
1222 			__cpuhp_kick_ap(st);
1223 		} else {
1224 			WARN(1, "DEAD callback error for CPU%d", cpu);
1225 		}
1226 	}
1227 
1228 out:
1229 	cpus_write_unlock();
1230 	/*
1231 	 * Do post unplug cleanup. This is still protected against
1232 	 * concurrent CPU hotplug via cpu_add_remove_lock.
1233 	 */
1234 	lockup_detector_cleanup();
1235 	arch_smt_update();
1236 	cpu_up_down_serialize_trainwrecks(tasks_frozen);
1237 	return ret;
1238 }
1239 
1240 struct cpu_down_work {
1241 	unsigned int		cpu;
1242 	enum cpuhp_state	target;
1243 };
1244 
__cpu_down_maps_locked(void * arg)1245 static long __cpu_down_maps_locked(void *arg)
1246 {
1247 	struct cpu_down_work *work = arg;
1248 
1249 	return _cpu_down(work->cpu, 0, work->target);
1250 }
1251 
cpu_down_maps_locked(unsigned int cpu,enum cpuhp_state target)1252 static int cpu_down_maps_locked(unsigned int cpu, enum cpuhp_state target)
1253 {
1254 	struct cpu_down_work work = { .cpu = cpu, .target = target, };
1255 
1256 	/*
1257 	 * If the platform does not support hotplug, report it explicitly to
1258 	 * differentiate it from a transient offlining failure.
1259 	 */
1260 	if (cc_platform_has(CC_ATTR_HOTPLUG_DISABLED))
1261 		return -EOPNOTSUPP;
1262 	if (cpu_hotplug_disabled)
1263 		return -EBUSY;
1264 
1265 	/*
1266 	 * Ensure that the control task does not run on the to be offlined
1267 	 * CPU to prevent a deadlock against cfs_b->period_timer.
1268 	 * Also keep at least one housekeeping cpu onlined to avoid generating
1269 	 * an empty sched_domain span.
1270 	 */
1271 	for_each_cpu_and(cpu, cpu_online_mask, housekeeping_cpumask(HK_TYPE_DOMAIN)) {
1272 		if (cpu != work.cpu)
1273 			return work_on_cpu(cpu, __cpu_down_maps_locked, &work);
1274 	}
1275 	return -EBUSY;
1276 }
1277 
cpu_down(unsigned int cpu,enum cpuhp_state target)1278 static int cpu_down(unsigned int cpu, enum cpuhp_state target)
1279 {
1280 	int err;
1281 
1282 	cpu_maps_update_begin();
1283 	err = cpu_down_maps_locked(cpu, target);
1284 	cpu_maps_update_done();
1285 	return err;
1286 }
1287 
1288 /**
1289  * cpu_device_down - Bring down a cpu device
1290  * @dev: Pointer to the cpu device to offline
1291  *
1292  * This function is meant to be used by device core cpu subsystem only.
1293  *
1294  * Other subsystems should use remove_cpu() instead.
1295  *
1296  * Return: %0 on success or a negative errno code
1297  */
cpu_device_down(struct device * dev)1298 int cpu_device_down(struct device *dev)
1299 {
1300 	return cpu_down(dev->id, CPUHP_OFFLINE);
1301 }
1302 
remove_cpu(unsigned int cpu)1303 int remove_cpu(unsigned int cpu)
1304 {
1305 	int ret;
1306 
1307 	lock_device_hotplug();
1308 	ret = device_offline(get_cpu_device(cpu));
1309 	unlock_device_hotplug();
1310 
1311 	return ret;
1312 }
1313 EXPORT_SYMBOL_GPL(remove_cpu);
1314 
smp_shutdown_nonboot_cpus(unsigned int primary_cpu)1315 void smp_shutdown_nonboot_cpus(unsigned int primary_cpu)
1316 {
1317 	unsigned int cpu;
1318 	int error;
1319 
1320 	cpu_maps_update_begin();
1321 
1322 	/*
1323 	 * Make certain the cpu I'm about to reboot on is online.
1324 	 *
1325 	 * This is inline to what migrate_to_reboot_cpu() already do.
1326 	 */
1327 	if (!cpu_online(primary_cpu))
1328 		primary_cpu = cpumask_first(cpu_online_mask);
1329 
1330 	for_each_online_cpu(cpu) {
1331 		if (cpu == primary_cpu)
1332 			continue;
1333 
1334 		error = cpu_down_maps_locked(cpu, CPUHP_OFFLINE);
1335 		if (error) {
1336 			pr_err("Failed to offline CPU%d - error=%d",
1337 				cpu, error);
1338 			break;
1339 		}
1340 	}
1341 
1342 	/*
1343 	 * Ensure all but the reboot CPU are offline.
1344 	 */
1345 	BUG_ON(num_online_cpus() > 1);
1346 
1347 	/*
1348 	 * Make sure the CPUs won't be enabled by someone else after this
1349 	 * point. Kexec will reboot to a new kernel shortly resetting
1350 	 * everything along the way.
1351 	 */
1352 	cpu_hotplug_disabled++;
1353 
1354 	cpu_maps_update_done();
1355 }
1356 
1357 #else
1358 #define takedown_cpu		NULL
1359 #endif /*CONFIG_HOTPLUG_CPU*/
1360 
1361 /**
1362  * notify_cpu_starting(cpu) - Invoke the callbacks on the starting CPU
1363  * @cpu: cpu that just started
1364  *
1365  * It must be called by the arch code on the new cpu, before the new cpu
1366  * enables interrupts and before the "boot" cpu returns from __cpu_up().
1367  */
notify_cpu_starting(unsigned int cpu)1368 void notify_cpu_starting(unsigned int cpu)
1369 {
1370 	struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1371 	enum cpuhp_state target = min((int)st->target, CPUHP_AP_ONLINE);
1372 
1373 	rcu_cpu_starting(cpu);	/* Enables RCU usage on this CPU. */
1374 	cpumask_set_cpu(cpu, &cpus_booted_once_mask);
1375 
1376 	/*
1377 	 * STARTING must not fail!
1378 	 */
1379 	cpuhp_invoke_callback_range_nofail(true, cpu, st, target);
1380 }
1381 
1382 /*
1383  * Called from the idle task. Wake up the controlling task which brings the
1384  * hotplug thread of the upcoming CPU up and then delegates the rest of the
1385  * online bringup to the hotplug thread.
1386  */
cpuhp_online_idle(enum cpuhp_state state)1387 void cpuhp_online_idle(enum cpuhp_state state)
1388 {
1389 	struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
1390 
1391 	/* Happens for the boot cpu */
1392 	if (state != CPUHP_AP_ONLINE_IDLE)
1393 		return;
1394 
1395 	/*
1396 	 * Unpart the stopper thread before we start the idle loop (and start
1397 	 * scheduling); this ensures the stopper task is always available.
1398 	 */
1399 	stop_machine_unpark(smp_processor_id());
1400 
1401 	st->state = CPUHP_AP_ONLINE_IDLE;
1402 	complete_ap_thread(st, true);
1403 }
1404 
1405 /* Requires cpu_add_remove_lock to be held */
_cpu_up(unsigned int cpu,int tasks_frozen,enum cpuhp_state target)1406 static int _cpu_up(unsigned int cpu, int tasks_frozen, enum cpuhp_state target)
1407 {
1408 	struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1409 	struct task_struct *idle;
1410 	int ret = 0;
1411 
1412 	cpus_write_lock();
1413 
1414 	if (!cpu_present(cpu)) {
1415 		ret = -EINVAL;
1416 		goto out;
1417 	}
1418 
1419 	/*
1420 	 * The caller of cpu_up() might have raced with another
1421 	 * caller. Nothing to do.
1422 	 */
1423 	if (st->state >= target)
1424 		goto out;
1425 
1426 	if (st->state == CPUHP_OFFLINE) {
1427 		/* Let it fail before we try to bring the cpu up */
1428 		idle = idle_thread_get(cpu);
1429 		if (IS_ERR(idle)) {
1430 			ret = PTR_ERR(idle);
1431 			goto out;
1432 		}
1433 	}
1434 
1435 	cpuhp_tasks_frozen = tasks_frozen;
1436 
1437 	cpuhp_set_state(cpu, st, target);
1438 	/*
1439 	 * If the current CPU state is in the range of the AP hotplug thread,
1440 	 * then we need to kick the thread once more.
1441 	 */
1442 	if (st->state > CPUHP_BRINGUP_CPU) {
1443 		ret = cpuhp_kick_ap_work(cpu);
1444 		/*
1445 		 * The AP side has done the error rollback already. Just
1446 		 * return the error code..
1447 		 */
1448 		if (ret)
1449 			goto out;
1450 	}
1451 
1452 	/*
1453 	 * Try to reach the target state. We max out on the BP at
1454 	 * CPUHP_BRINGUP_CPU. After that the AP hotplug thread is
1455 	 * responsible for bringing it up to the target state.
1456 	 */
1457 	target = min((int)target, CPUHP_BRINGUP_CPU);
1458 	ret = cpuhp_up_callbacks(cpu, st, target);
1459 out:
1460 	cpus_write_unlock();
1461 	arch_smt_update();
1462 	cpu_up_down_serialize_trainwrecks(tasks_frozen);
1463 	return ret;
1464 }
1465 
cpu_up(unsigned int cpu,enum cpuhp_state target)1466 static int cpu_up(unsigned int cpu, enum cpuhp_state target)
1467 {
1468 	int err = 0;
1469 
1470 	if (!cpu_possible(cpu)) {
1471 		pr_err("can't online cpu %d because it is not configured as may-hotadd at boot time\n",
1472 		       cpu);
1473 #if defined(CONFIG_IA64)
1474 		pr_err("please check additional_cpus= boot parameter\n");
1475 #endif
1476 		return -EINVAL;
1477 	}
1478 
1479 	err = try_online_node(cpu_to_node(cpu));
1480 	if (err)
1481 		return err;
1482 
1483 	cpu_maps_update_begin();
1484 
1485 	if (cpu_hotplug_disabled) {
1486 		err = -EBUSY;
1487 		goto out;
1488 	}
1489 	if (!cpu_bootable(cpu)) {
1490 		err = -EPERM;
1491 		goto out;
1492 	}
1493 
1494 	err = _cpu_up(cpu, 0, target);
1495 out:
1496 	cpu_maps_update_done();
1497 	return err;
1498 }
1499 
1500 /**
1501  * cpu_device_up - Bring up a cpu device
1502  * @dev: Pointer to the cpu device to online
1503  *
1504  * This function is meant to be used by device core cpu subsystem only.
1505  *
1506  * Other subsystems should use add_cpu() instead.
1507  *
1508  * Return: %0 on success or a negative errno code
1509  */
cpu_device_up(struct device * dev)1510 int cpu_device_up(struct device *dev)
1511 {
1512 	return cpu_up(dev->id, CPUHP_ONLINE);
1513 }
1514 
add_cpu(unsigned int cpu)1515 int add_cpu(unsigned int cpu)
1516 {
1517 	int ret;
1518 
1519 	lock_device_hotplug();
1520 	ret = device_online(get_cpu_device(cpu));
1521 	unlock_device_hotplug();
1522 
1523 	return ret;
1524 }
1525 EXPORT_SYMBOL_GPL(add_cpu);
1526 
1527 /**
1528  * bringup_hibernate_cpu - Bring up the CPU that we hibernated on
1529  * @sleep_cpu: The cpu we hibernated on and should be brought up.
1530  *
1531  * On some architectures like arm64, we can hibernate on any CPU, but on
1532  * wake up the CPU we hibernated on might be offline as a side effect of
1533  * using maxcpus= for example.
1534  *
1535  * Return: %0 on success or a negative errno code
1536  */
bringup_hibernate_cpu(unsigned int sleep_cpu)1537 int bringup_hibernate_cpu(unsigned int sleep_cpu)
1538 {
1539 	int ret;
1540 
1541 	if (!cpu_online(sleep_cpu)) {
1542 		pr_info("Hibernated on a CPU that is offline! Bringing CPU up.\n");
1543 		ret = cpu_up(sleep_cpu, CPUHP_ONLINE);
1544 		if (ret) {
1545 			pr_err("Failed to bring hibernate-CPU up!\n");
1546 			return ret;
1547 		}
1548 	}
1549 	return 0;
1550 }
1551 
bringup_nonboot_cpus(unsigned int setup_max_cpus)1552 void bringup_nonboot_cpus(unsigned int setup_max_cpus)
1553 {
1554 	unsigned int cpu;
1555 
1556 	for_each_present_cpu(cpu) {
1557 		if (num_online_cpus() >= setup_max_cpus)
1558 			break;
1559 		if (!cpu_online(cpu))
1560 			cpu_up(cpu, CPUHP_ONLINE);
1561 	}
1562 }
1563 
1564 #ifdef CONFIG_PM_SLEEP_SMP
1565 static cpumask_var_t frozen_cpus;
1566 
freeze_secondary_cpus(int primary)1567 int freeze_secondary_cpus(int primary)
1568 {
1569 	int cpu, error = 0;
1570 
1571 	cpu_maps_update_begin();
1572 	if (primary == -1) {
1573 		primary = cpumask_first(cpu_online_mask);
1574 		if (!housekeeping_cpu(primary, HK_TYPE_TIMER))
1575 			primary = housekeeping_any_cpu(HK_TYPE_TIMER);
1576 	} else {
1577 		if (!cpu_online(primary))
1578 			primary = cpumask_first(cpu_online_mask);
1579 	}
1580 
1581 	/*
1582 	 * We take down all of the non-boot CPUs in one shot to avoid races
1583 	 * with the userspace trying to use the CPU hotplug at the same time
1584 	 */
1585 	cpumask_clear(frozen_cpus);
1586 
1587 	pr_info("Disabling non-boot CPUs ...\n");
1588 	for_each_online_cpu(cpu) {
1589 		if (cpu == primary)
1590 			continue;
1591 
1592 		if (pm_wakeup_pending()) {
1593 			pr_info("Wakeup pending. Abort CPU freeze\n");
1594 			error = -EBUSY;
1595 			break;
1596 		}
1597 
1598 		trace_suspend_resume(TPS("CPU_OFF"), cpu, true);
1599 		error = _cpu_down(cpu, 1, CPUHP_OFFLINE);
1600 		trace_suspend_resume(TPS("CPU_OFF"), cpu, false);
1601 		if (!error)
1602 			cpumask_set_cpu(cpu, frozen_cpus);
1603 		else {
1604 			pr_err("Error taking CPU%d down: %d\n", cpu, error);
1605 			break;
1606 		}
1607 	}
1608 
1609 	if (!error)
1610 		BUG_ON(num_online_cpus() > 1);
1611 	else
1612 		pr_err("Non-boot CPUs are not disabled\n");
1613 
1614 	/*
1615 	 * Make sure the CPUs won't be enabled by someone else. We need to do
1616 	 * this even in case of failure as all freeze_secondary_cpus() users are
1617 	 * supposed to do thaw_secondary_cpus() on the failure path.
1618 	 */
1619 	cpu_hotplug_disabled++;
1620 
1621 	cpu_maps_update_done();
1622 	return error;
1623 }
1624 
arch_thaw_secondary_cpus_begin(void)1625 void __weak arch_thaw_secondary_cpus_begin(void)
1626 {
1627 }
1628 
arch_thaw_secondary_cpus_end(void)1629 void __weak arch_thaw_secondary_cpus_end(void)
1630 {
1631 }
1632 
thaw_secondary_cpus(void)1633 void thaw_secondary_cpus(void)
1634 {
1635 	int cpu, error;
1636 
1637 	/* Allow everyone to use the CPU hotplug again */
1638 	cpu_maps_update_begin();
1639 	__cpu_hotplug_enable();
1640 	if (cpumask_empty(frozen_cpus))
1641 		goto out;
1642 
1643 	pr_info("Enabling non-boot CPUs ...\n");
1644 
1645 	arch_thaw_secondary_cpus_begin();
1646 
1647 	for_each_cpu(cpu, frozen_cpus) {
1648 		trace_suspend_resume(TPS("CPU_ON"), cpu, true);
1649 		error = _cpu_up(cpu, 1, CPUHP_ONLINE);
1650 		trace_suspend_resume(TPS("CPU_ON"), cpu, false);
1651 		if (!error) {
1652 			pr_info("CPU%d is up\n", cpu);
1653 			continue;
1654 		}
1655 		pr_warn("Error taking CPU%d up: %d\n", cpu, error);
1656 	}
1657 
1658 	arch_thaw_secondary_cpus_end();
1659 
1660 	cpumask_clear(frozen_cpus);
1661 out:
1662 	cpu_maps_update_done();
1663 }
1664 
alloc_frozen_cpus(void)1665 static int __init alloc_frozen_cpus(void)
1666 {
1667 	if (!alloc_cpumask_var(&frozen_cpus, GFP_KERNEL|__GFP_ZERO))
1668 		return -ENOMEM;
1669 	return 0;
1670 }
1671 core_initcall(alloc_frozen_cpus);
1672 
1673 /*
1674  * When callbacks for CPU hotplug notifications are being executed, we must
1675  * ensure that the state of the system with respect to the tasks being frozen
1676  * or not, as reported by the notification, remains unchanged *throughout the
1677  * duration* of the execution of the callbacks.
1678  * Hence we need to prevent the freezer from racing with regular CPU hotplug.
1679  *
1680  * This synchronization is implemented by mutually excluding regular CPU
1681  * hotplug and Suspend/Hibernate call paths by hooking onto the Suspend/
1682  * Hibernate notifications.
1683  */
1684 static int
cpu_hotplug_pm_callback(struct notifier_block * nb,unsigned long action,void * ptr)1685 cpu_hotplug_pm_callback(struct notifier_block *nb,
1686 			unsigned long action, void *ptr)
1687 {
1688 	switch (action) {
1689 
1690 	case PM_SUSPEND_PREPARE:
1691 	case PM_HIBERNATION_PREPARE:
1692 		cpu_hotplug_disable();
1693 		break;
1694 
1695 	case PM_POST_SUSPEND:
1696 	case PM_POST_HIBERNATION:
1697 		cpu_hotplug_enable();
1698 		break;
1699 
1700 	default:
1701 		return NOTIFY_DONE;
1702 	}
1703 
1704 	return NOTIFY_OK;
1705 }
1706 
1707 
cpu_hotplug_pm_sync_init(void)1708 static int __init cpu_hotplug_pm_sync_init(void)
1709 {
1710 	/*
1711 	 * cpu_hotplug_pm_callback has higher priority than x86
1712 	 * bsp_pm_callback which depends on cpu_hotplug_pm_callback
1713 	 * to disable cpu hotplug to avoid cpu hotplug race.
1714 	 */
1715 	pm_notifier(cpu_hotplug_pm_callback, 0);
1716 	return 0;
1717 }
1718 core_initcall(cpu_hotplug_pm_sync_init);
1719 
1720 #endif /* CONFIG_PM_SLEEP_SMP */
1721 
1722 int __boot_cpu_id;
1723 
1724 #endif /* CONFIG_SMP */
1725 
1726 /* Boot processor state steps */
1727 static struct cpuhp_step cpuhp_hp_states[] = {
1728 	[CPUHP_OFFLINE] = {
1729 		.name			= "offline",
1730 		.startup.single		= NULL,
1731 		.teardown.single	= NULL,
1732 	},
1733 #ifdef CONFIG_SMP
1734 	[CPUHP_CREATE_THREADS]= {
1735 		.name			= "threads:prepare",
1736 		.startup.single		= smpboot_create_threads,
1737 		.teardown.single	= NULL,
1738 		.cant_stop		= true,
1739 	},
1740 	[CPUHP_PERF_PREPARE] = {
1741 		.name			= "perf:prepare",
1742 		.startup.single		= perf_event_init_cpu,
1743 		.teardown.single	= perf_event_exit_cpu,
1744 	},
1745 	[CPUHP_RANDOM_PREPARE] = {
1746 		.name			= "random:prepare",
1747 		.startup.single		= random_prepare_cpu,
1748 		.teardown.single	= NULL,
1749 	},
1750 	[CPUHP_WORKQUEUE_PREP] = {
1751 		.name			= "workqueue:prepare",
1752 		.startup.single		= workqueue_prepare_cpu,
1753 		.teardown.single	= NULL,
1754 	},
1755 	[CPUHP_HRTIMERS_PREPARE] = {
1756 		.name			= "hrtimers:prepare",
1757 		.startup.single		= hrtimers_prepare_cpu,
1758 		.teardown.single	= hrtimers_dead_cpu,
1759 	},
1760 	[CPUHP_SMPCFD_PREPARE] = {
1761 		.name			= "smpcfd:prepare",
1762 		.startup.single		= smpcfd_prepare_cpu,
1763 		.teardown.single	= smpcfd_dead_cpu,
1764 	},
1765 	[CPUHP_RELAY_PREPARE] = {
1766 		.name			= "relay:prepare",
1767 		.startup.single		= relay_prepare_cpu,
1768 		.teardown.single	= NULL,
1769 	},
1770 	[CPUHP_SLAB_PREPARE] = {
1771 		.name			= "slab:prepare",
1772 		.startup.single		= slab_prepare_cpu,
1773 		.teardown.single	= slab_dead_cpu,
1774 	},
1775 	[CPUHP_RCUTREE_PREP] = {
1776 		.name			= "RCU/tree:prepare",
1777 		.startup.single		= rcutree_prepare_cpu,
1778 		.teardown.single	= rcutree_dead_cpu,
1779 	},
1780 	/*
1781 	 * On the tear-down path, timers_dead_cpu() must be invoked
1782 	 * before blk_mq_queue_reinit_notify() from notify_dead(),
1783 	 * otherwise a RCU stall occurs.
1784 	 */
1785 	[CPUHP_TIMERS_PREPARE] = {
1786 		.name			= "timers:prepare",
1787 		.startup.single		= timers_prepare_cpu,
1788 		.teardown.single	= timers_dead_cpu,
1789 	},
1790 	/* Kicks the plugged cpu into life */
1791 	[CPUHP_BRINGUP_CPU] = {
1792 		.name			= "cpu:bringup",
1793 		.startup.single		= bringup_cpu,
1794 		.teardown.single	= finish_cpu,
1795 		.cant_stop		= true,
1796 	},
1797 	/* Final state before CPU kills itself */
1798 	[CPUHP_AP_IDLE_DEAD] = {
1799 		.name			= "idle:dead",
1800 	},
1801 	/*
1802 	 * Last state before CPU enters the idle loop to die. Transient state
1803 	 * for synchronization.
1804 	 */
1805 	[CPUHP_AP_OFFLINE] = {
1806 		.name			= "ap:offline",
1807 		.cant_stop		= true,
1808 	},
1809 	/* First state is scheduler control. Interrupts are disabled */
1810 	[CPUHP_AP_SCHED_STARTING] = {
1811 		.name			= "sched:starting",
1812 		.startup.single		= sched_cpu_starting,
1813 		.teardown.single	= sched_cpu_dying,
1814 	},
1815 	[CPUHP_AP_RCUTREE_DYING] = {
1816 		.name			= "RCU/tree:dying",
1817 		.startup.single		= NULL,
1818 		.teardown.single	= rcutree_dying_cpu,
1819 	},
1820 	[CPUHP_AP_SMPCFD_DYING] = {
1821 		.name			= "smpcfd:dying",
1822 		.startup.single		= NULL,
1823 		.teardown.single	= smpcfd_dying_cpu,
1824 	},
1825 	/* Entry state on starting. Interrupts enabled from here on. Transient
1826 	 * state for synchronsization */
1827 	[CPUHP_AP_ONLINE] = {
1828 		.name			= "ap:online",
1829 	},
1830 	/*
1831 	 * Handled on control processor until the plugged processor manages
1832 	 * this itself.
1833 	 */
1834 	[CPUHP_TEARDOWN_CPU] = {
1835 		.name			= "cpu:teardown",
1836 		.startup.single		= NULL,
1837 		.teardown.single	= takedown_cpu,
1838 		.cant_stop		= true,
1839 	},
1840 
1841 	[CPUHP_AP_SCHED_WAIT_EMPTY] = {
1842 		.name			= "sched:waitempty",
1843 		.startup.single		= NULL,
1844 		.teardown.single	= sched_cpu_wait_empty,
1845 	},
1846 
1847 	/* Handle smpboot threads park/unpark */
1848 	[CPUHP_AP_SMPBOOT_THREADS] = {
1849 		.name			= "smpboot/threads:online",
1850 		.startup.single		= smpboot_unpark_threads,
1851 		.teardown.single	= smpboot_park_threads,
1852 	},
1853 	[CPUHP_AP_IRQ_AFFINITY_ONLINE] = {
1854 		.name			= "irq/affinity:online",
1855 		.startup.single		= irq_affinity_online_cpu,
1856 		.teardown.single	= NULL,
1857 	},
1858 	[CPUHP_AP_PERF_ONLINE] = {
1859 		.name			= "perf:online",
1860 		.startup.single		= perf_event_init_cpu,
1861 		.teardown.single	= perf_event_exit_cpu,
1862 	},
1863 	[CPUHP_AP_WATCHDOG_ONLINE] = {
1864 		.name			= "lockup_detector:online",
1865 		.startup.single		= lockup_detector_online_cpu,
1866 		.teardown.single	= lockup_detector_offline_cpu,
1867 	},
1868 	[CPUHP_AP_WORKQUEUE_ONLINE] = {
1869 		.name			= "workqueue:online",
1870 		.startup.single		= workqueue_online_cpu,
1871 		.teardown.single	= workqueue_offline_cpu,
1872 	},
1873 	[CPUHP_AP_RANDOM_ONLINE] = {
1874 		.name			= "random:online",
1875 		.startup.single		= random_online_cpu,
1876 		.teardown.single	= NULL,
1877 	},
1878 	[CPUHP_AP_RCUTREE_ONLINE] = {
1879 		.name			= "RCU/tree:online",
1880 		.startup.single		= rcutree_online_cpu,
1881 		.teardown.single	= rcutree_offline_cpu,
1882 	},
1883 #endif
1884 	/*
1885 	 * The dynamically registered state space is here
1886 	 */
1887 
1888 #ifdef CONFIG_SMP
1889 	/* Last state is scheduler control setting the cpu active */
1890 	[CPUHP_AP_ACTIVE] = {
1891 		.name			= "sched:active",
1892 		.startup.single		= sched_cpu_activate,
1893 		.teardown.single	= sched_cpu_deactivate,
1894 	},
1895 #endif
1896 
1897 	/* CPU is fully up and running. */
1898 	[CPUHP_ONLINE] = {
1899 		.name			= "online",
1900 		.startup.single		= NULL,
1901 		.teardown.single	= NULL,
1902 	},
1903 };
1904 
1905 /* Sanity check for callbacks */
cpuhp_cb_check(enum cpuhp_state state)1906 static int cpuhp_cb_check(enum cpuhp_state state)
1907 {
1908 	if (state <= CPUHP_OFFLINE || state >= CPUHP_ONLINE)
1909 		return -EINVAL;
1910 	return 0;
1911 }
1912 
1913 /*
1914  * Returns a free for dynamic slot assignment of the Online state. The states
1915  * are protected by the cpuhp_slot_states mutex and an empty slot is identified
1916  * by having no name assigned.
1917  */
cpuhp_reserve_state(enum cpuhp_state state)1918 static int cpuhp_reserve_state(enum cpuhp_state state)
1919 {
1920 	enum cpuhp_state i, end;
1921 	struct cpuhp_step *step;
1922 
1923 	switch (state) {
1924 	case CPUHP_AP_ONLINE_DYN:
1925 		step = cpuhp_hp_states + CPUHP_AP_ONLINE_DYN;
1926 		end = CPUHP_AP_ONLINE_DYN_END;
1927 		break;
1928 	case CPUHP_BP_PREPARE_DYN:
1929 		step = cpuhp_hp_states + CPUHP_BP_PREPARE_DYN;
1930 		end = CPUHP_BP_PREPARE_DYN_END;
1931 		break;
1932 	default:
1933 		return -EINVAL;
1934 	}
1935 
1936 	for (i = state; i <= end; i++, step++) {
1937 		if (!step->name)
1938 			return i;
1939 	}
1940 	WARN(1, "No more dynamic states available for CPU hotplug\n");
1941 	return -ENOSPC;
1942 }
1943 
cpuhp_store_callbacks(enum cpuhp_state state,const char * name,int (* startup)(unsigned int cpu),int (* teardown)(unsigned int cpu),bool multi_instance)1944 static int cpuhp_store_callbacks(enum cpuhp_state state, const char *name,
1945 				 int (*startup)(unsigned int cpu),
1946 				 int (*teardown)(unsigned int cpu),
1947 				 bool multi_instance)
1948 {
1949 	/* (Un)Install the callbacks for further cpu hotplug operations */
1950 	struct cpuhp_step *sp;
1951 	int ret = 0;
1952 
1953 	/*
1954 	 * If name is NULL, then the state gets removed.
1955 	 *
1956 	 * CPUHP_AP_ONLINE_DYN and CPUHP_BP_PREPARE_DYN are handed out on
1957 	 * the first allocation from these dynamic ranges, so the removal
1958 	 * would trigger a new allocation and clear the wrong (already
1959 	 * empty) state, leaving the callbacks of the to be cleared state
1960 	 * dangling, which causes wreckage on the next hotplug operation.
1961 	 */
1962 	if (name && (state == CPUHP_AP_ONLINE_DYN ||
1963 		     state == CPUHP_BP_PREPARE_DYN)) {
1964 		ret = cpuhp_reserve_state(state);
1965 		if (ret < 0)
1966 			return ret;
1967 		state = ret;
1968 	}
1969 	sp = cpuhp_get_step(state);
1970 	if (name && sp->name)
1971 		return -EBUSY;
1972 
1973 	sp->startup.single = startup;
1974 	sp->teardown.single = teardown;
1975 	sp->name = name;
1976 	sp->multi_instance = multi_instance;
1977 	INIT_HLIST_HEAD(&sp->list);
1978 	return ret;
1979 }
1980 
cpuhp_get_teardown_cb(enum cpuhp_state state)1981 static void *cpuhp_get_teardown_cb(enum cpuhp_state state)
1982 {
1983 	return cpuhp_get_step(state)->teardown.single;
1984 }
1985 
1986 /*
1987  * Call the startup/teardown function for a step either on the AP or
1988  * on the current CPU.
1989  */
cpuhp_issue_call(int cpu,enum cpuhp_state state,bool bringup,struct hlist_node * node)1990 static int cpuhp_issue_call(int cpu, enum cpuhp_state state, bool bringup,
1991 			    struct hlist_node *node)
1992 {
1993 	struct cpuhp_step *sp = cpuhp_get_step(state);
1994 	int ret;
1995 
1996 	/*
1997 	 * If there's nothing to do, we done.
1998 	 * Relies on the union for multi_instance.
1999 	 */
2000 	if (cpuhp_step_empty(bringup, sp))
2001 		return 0;
2002 	/*
2003 	 * The non AP bound callbacks can fail on bringup. On teardown
2004 	 * e.g. module removal we crash for now.
2005 	 */
2006 #ifdef CONFIG_SMP
2007 	if (cpuhp_is_ap_state(state))
2008 		ret = cpuhp_invoke_ap_callback(cpu, state, bringup, node);
2009 	else
2010 		ret = cpuhp_invoke_callback(cpu, state, bringup, node, NULL);
2011 #else
2012 	ret = cpuhp_invoke_callback(cpu, state, bringup, node, NULL);
2013 #endif
2014 	BUG_ON(ret && !bringup);
2015 	return ret;
2016 }
2017 
2018 /*
2019  * Called from __cpuhp_setup_state on a recoverable failure.
2020  *
2021  * Note: The teardown callbacks for rollback are not allowed to fail!
2022  */
cpuhp_rollback_install(int failedcpu,enum cpuhp_state state,struct hlist_node * node)2023 static void cpuhp_rollback_install(int failedcpu, enum cpuhp_state state,
2024 				   struct hlist_node *node)
2025 {
2026 	int cpu;
2027 
2028 	/* Roll back the already executed steps on the other cpus */
2029 	for_each_present_cpu(cpu) {
2030 		struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
2031 		int cpustate = st->state;
2032 
2033 		if (cpu >= failedcpu)
2034 			break;
2035 
2036 		/* Did we invoke the startup call on that cpu ? */
2037 		if (cpustate >= state)
2038 			cpuhp_issue_call(cpu, state, false, node);
2039 	}
2040 }
2041 
__cpuhp_state_add_instance_cpuslocked(enum cpuhp_state state,struct hlist_node * node,bool invoke)2042 int __cpuhp_state_add_instance_cpuslocked(enum cpuhp_state state,
2043 					  struct hlist_node *node,
2044 					  bool invoke)
2045 {
2046 	struct cpuhp_step *sp;
2047 	int cpu;
2048 	int ret;
2049 
2050 	lockdep_assert_cpus_held();
2051 
2052 	sp = cpuhp_get_step(state);
2053 	if (sp->multi_instance == false)
2054 		return -EINVAL;
2055 
2056 	mutex_lock(&cpuhp_state_mutex);
2057 
2058 	if (!invoke || !sp->startup.multi)
2059 		goto add_node;
2060 
2061 	/*
2062 	 * Try to call the startup callback for each present cpu
2063 	 * depending on the hotplug state of the cpu.
2064 	 */
2065 	for_each_present_cpu(cpu) {
2066 		struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
2067 		int cpustate = st->state;
2068 
2069 		if (cpustate < state)
2070 			continue;
2071 
2072 		ret = cpuhp_issue_call(cpu, state, true, node);
2073 		if (ret) {
2074 			if (sp->teardown.multi)
2075 				cpuhp_rollback_install(cpu, state, node);
2076 			goto unlock;
2077 		}
2078 	}
2079 add_node:
2080 	ret = 0;
2081 	hlist_add_head(node, &sp->list);
2082 unlock:
2083 	mutex_unlock(&cpuhp_state_mutex);
2084 	return ret;
2085 }
2086 
__cpuhp_state_add_instance(enum cpuhp_state state,struct hlist_node * node,bool invoke)2087 int __cpuhp_state_add_instance(enum cpuhp_state state, struct hlist_node *node,
2088 			       bool invoke)
2089 {
2090 	int ret;
2091 
2092 	cpus_read_lock();
2093 	ret = __cpuhp_state_add_instance_cpuslocked(state, node, invoke);
2094 	cpus_read_unlock();
2095 	return ret;
2096 }
2097 EXPORT_SYMBOL_GPL(__cpuhp_state_add_instance);
2098 
2099 /**
2100  * __cpuhp_setup_state_cpuslocked - Setup the callbacks for an hotplug machine state
2101  * @state:		The state to setup
2102  * @name:		Name of the step
2103  * @invoke:		If true, the startup function is invoked for cpus where
2104  *			cpu state >= @state
2105  * @startup:		startup callback function
2106  * @teardown:		teardown callback function
2107  * @multi_instance:	State is set up for multiple instances which get
2108  *			added afterwards.
2109  *
2110  * The caller needs to hold cpus read locked while calling this function.
2111  * Return:
2112  *   On success:
2113  *      Positive state number if @state is CPUHP_AP_ONLINE_DYN;
2114  *      0 for all other states
2115  *   On failure: proper (negative) error code
2116  */
__cpuhp_setup_state_cpuslocked(enum cpuhp_state state,const char * name,bool invoke,int (* startup)(unsigned int cpu),int (* teardown)(unsigned int cpu),bool multi_instance)2117 int __cpuhp_setup_state_cpuslocked(enum cpuhp_state state,
2118 				   const char *name, bool invoke,
2119 				   int (*startup)(unsigned int cpu),
2120 				   int (*teardown)(unsigned int cpu),
2121 				   bool multi_instance)
2122 {
2123 	int cpu, ret = 0;
2124 	bool dynstate;
2125 
2126 	lockdep_assert_cpus_held();
2127 
2128 	if (cpuhp_cb_check(state) || !name)
2129 		return -EINVAL;
2130 
2131 	mutex_lock(&cpuhp_state_mutex);
2132 
2133 	ret = cpuhp_store_callbacks(state, name, startup, teardown,
2134 				    multi_instance);
2135 
2136 	dynstate = state == CPUHP_AP_ONLINE_DYN;
2137 	if (ret > 0 && dynstate) {
2138 		state = ret;
2139 		ret = 0;
2140 	}
2141 
2142 	if (ret || !invoke || !startup)
2143 		goto out;
2144 
2145 	/*
2146 	 * Try to call the startup callback for each present cpu
2147 	 * depending on the hotplug state of the cpu.
2148 	 */
2149 	for_each_present_cpu(cpu) {
2150 		struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
2151 		int cpustate = st->state;
2152 
2153 		if (cpustate < state)
2154 			continue;
2155 
2156 		ret = cpuhp_issue_call(cpu, state, true, NULL);
2157 		if (ret) {
2158 			if (teardown)
2159 				cpuhp_rollback_install(cpu, state, NULL);
2160 			cpuhp_store_callbacks(state, NULL, NULL, NULL, false);
2161 			goto out;
2162 		}
2163 	}
2164 out:
2165 	mutex_unlock(&cpuhp_state_mutex);
2166 	/*
2167 	 * If the requested state is CPUHP_AP_ONLINE_DYN, return the
2168 	 * dynamically allocated state in case of success.
2169 	 */
2170 	if (!ret && dynstate)
2171 		return state;
2172 	return ret;
2173 }
2174 EXPORT_SYMBOL(__cpuhp_setup_state_cpuslocked);
2175 
__cpuhp_setup_state(enum cpuhp_state state,const char * name,bool invoke,int (* startup)(unsigned int cpu),int (* teardown)(unsigned int cpu),bool multi_instance)2176 int __cpuhp_setup_state(enum cpuhp_state state,
2177 			const char *name, bool invoke,
2178 			int (*startup)(unsigned int cpu),
2179 			int (*teardown)(unsigned int cpu),
2180 			bool multi_instance)
2181 {
2182 	int ret;
2183 
2184 	cpus_read_lock();
2185 	ret = __cpuhp_setup_state_cpuslocked(state, name, invoke, startup,
2186 					     teardown, multi_instance);
2187 	cpus_read_unlock();
2188 	return ret;
2189 }
2190 EXPORT_SYMBOL(__cpuhp_setup_state);
2191 
__cpuhp_state_remove_instance(enum cpuhp_state state,struct hlist_node * node,bool invoke)2192 int __cpuhp_state_remove_instance(enum cpuhp_state state,
2193 				  struct hlist_node *node, bool invoke)
2194 {
2195 	struct cpuhp_step *sp = cpuhp_get_step(state);
2196 	int cpu;
2197 
2198 	BUG_ON(cpuhp_cb_check(state));
2199 
2200 	if (!sp->multi_instance)
2201 		return -EINVAL;
2202 
2203 	cpus_read_lock();
2204 	mutex_lock(&cpuhp_state_mutex);
2205 
2206 	if (!invoke || !cpuhp_get_teardown_cb(state))
2207 		goto remove;
2208 	/*
2209 	 * Call the teardown callback for each present cpu depending
2210 	 * on the hotplug state of the cpu. This function is not
2211 	 * allowed to fail currently!
2212 	 */
2213 	for_each_present_cpu(cpu) {
2214 		struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
2215 		int cpustate = st->state;
2216 
2217 		if (cpustate >= state)
2218 			cpuhp_issue_call(cpu, state, false, node);
2219 	}
2220 
2221 remove:
2222 	hlist_del(node);
2223 	mutex_unlock(&cpuhp_state_mutex);
2224 	cpus_read_unlock();
2225 
2226 	return 0;
2227 }
2228 EXPORT_SYMBOL_GPL(__cpuhp_state_remove_instance);
2229 
2230 /**
2231  * __cpuhp_remove_state_cpuslocked - Remove the callbacks for an hotplug machine state
2232  * @state:	The state to remove
2233  * @invoke:	If true, the teardown function is invoked for cpus where
2234  *		cpu state >= @state
2235  *
2236  * The caller needs to hold cpus read locked while calling this function.
2237  * The teardown callback is currently not allowed to fail. Think
2238  * about module removal!
2239  */
__cpuhp_remove_state_cpuslocked(enum cpuhp_state state,bool invoke)2240 void __cpuhp_remove_state_cpuslocked(enum cpuhp_state state, bool invoke)
2241 {
2242 	struct cpuhp_step *sp = cpuhp_get_step(state);
2243 	int cpu;
2244 
2245 	BUG_ON(cpuhp_cb_check(state));
2246 
2247 	lockdep_assert_cpus_held();
2248 
2249 	mutex_lock(&cpuhp_state_mutex);
2250 	if (sp->multi_instance) {
2251 		WARN(!hlist_empty(&sp->list),
2252 		     "Error: Removing state %d which has instances left.\n",
2253 		     state);
2254 		goto remove;
2255 	}
2256 
2257 	if (!invoke || !cpuhp_get_teardown_cb(state))
2258 		goto remove;
2259 
2260 	/*
2261 	 * Call the teardown callback for each present cpu depending
2262 	 * on the hotplug state of the cpu. This function is not
2263 	 * allowed to fail currently!
2264 	 */
2265 	for_each_present_cpu(cpu) {
2266 		struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
2267 		int cpustate = st->state;
2268 
2269 		if (cpustate >= state)
2270 			cpuhp_issue_call(cpu, state, false, NULL);
2271 	}
2272 remove:
2273 	cpuhp_store_callbacks(state, NULL, NULL, NULL, false);
2274 	mutex_unlock(&cpuhp_state_mutex);
2275 }
2276 EXPORT_SYMBOL(__cpuhp_remove_state_cpuslocked);
2277 
__cpuhp_remove_state(enum cpuhp_state state,bool invoke)2278 void __cpuhp_remove_state(enum cpuhp_state state, bool invoke)
2279 {
2280 	cpus_read_lock();
2281 	__cpuhp_remove_state_cpuslocked(state, invoke);
2282 	cpus_read_unlock();
2283 }
2284 EXPORT_SYMBOL(__cpuhp_remove_state);
2285 
2286 #ifdef CONFIG_HOTPLUG_SMT
cpuhp_offline_cpu_device(unsigned int cpu)2287 static void cpuhp_offline_cpu_device(unsigned int cpu)
2288 {
2289 	struct device *dev = get_cpu_device(cpu);
2290 
2291 	dev->offline = true;
2292 	/* Tell user space about the state change */
2293 	kobject_uevent(&dev->kobj, KOBJ_OFFLINE);
2294 }
2295 
cpuhp_online_cpu_device(unsigned int cpu)2296 static void cpuhp_online_cpu_device(unsigned int cpu)
2297 {
2298 	struct device *dev = get_cpu_device(cpu);
2299 
2300 	dev->offline = false;
2301 	/* Tell user space about the state change */
2302 	kobject_uevent(&dev->kobj, KOBJ_ONLINE);
2303 }
2304 
cpuhp_smt_disable(enum cpuhp_smt_control ctrlval)2305 int cpuhp_smt_disable(enum cpuhp_smt_control ctrlval)
2306 {
2307 	int cpu, ret = 0;
2308 
2309 	cpu_maps_update_begin();
2310 	for_each_online_cpu(cpu) {
2311 		if (topology_is_primary_thread(cpu))
2312 			continue;
2313 		/*
2314 		 * Disable can be called with CPU_SMT_ENABLED when changing
2315 		 * from a higher to lower number of SMT threads per core.
2316 		 */
2317 		if (ctrlval == CPU_SMT_ENABLED && cpu_smt_thread_allowed(cpu))
2318 			continue;
2319 		ret = cpu_down_maps_locked(cpu, CPUHP_OFFLINE);
2320 		if (ret)
2321 			break;
2322 		/*
2323 		 * As this needs to hold the cpu maps lock it's impossible
2324 		 * to call device_offline() because that ends up calling
2325 		 * cpu_down() which takes cpu maps lock. cpu maps lock
2326 		 * needs to be held as this might race against in kernel
2327 		 * abusers of the hotplug machinery (thermal management).
2328 		 *
2329 		 * So nothing would update device:offline state. That would
2330 		 * leave the sysfs entry stale and prevent onlining after
2331 		 * smt control has been changed to 'off' again. This is
2332 		 * called under the sysfs hotplug lock, so it is properly
2333 		 * serialized against the regular offline usage.
2334 		 */
2335 		cpuhp_offline_cpu_device(cpu);
2336 	}
2337 	if (!ret)
2338 		cpu_smt_control = ctrlval;
2339 	cpu_maps_update_done();
2340 	return ret;
2341 }
2342 
cpuhp_smt_enable(void)2343 int cpuhp_smt_enable(void)
2344 {
2345 	int cpu, ret = 0;
2346 
2347 	cpu_maps_update_begin();
2348 	cpu_smt_control = CPU_SMT_ENABLED;
2349 	for_each_present_cpu(cpu) {
2350 		/* Skip online CPUs and CPUs on offline nodes */
2351 		if (cpu_online(cpu) || !node_online(cpu_to_node(cpu)))
2352 			continue;
2353 		if (!cpu_smt_thread_allowed(cpu))
2354 			continue;
2355 		ret = _cpu_up(cpu, 0, CPUHP_ONLINE);
2356 		if (ret)
2357 			break;
2358 		/* See comment in cpuhp_smt_disable() */
2359 		cpuhp_online_cpu_device(cpu);
2360 	}
2361 	cpu_maps_update_done();
2362 	return ret;
2363 }
2364 #endif
2365 
2366 #if defined(CONFIG_SYSFS) && defined(CONFIG_HOTPLUG_CPU)
state_show(struct device * dev,struct device_attribute * attr,char * buf)2367 static ssize_t state_show(struct device *dev,
2368 			  struct device_attribute *attr, char *buf)
2369 {
2370 	struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id);
2371 
2372 	return sprintf(buf, "%d\n", st->state);
2373 }
2374 static DEVICE_ATTR_RO(state);
2375 
target_store(struct device * dev,struct device_attribute * attr,const char * buf,size_t count)2376 static ssize_t target_store(struct device *dev, struct device_attribute *attr,
2377 			    const char *buf, size_t count)
2378 {
2379 	struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id);
2380 	struct cpuhp_step *sp;
2381 	int target, ret;
2382 
2383 	ret = kstrtoint(buf, 10, &target);
2384 	if (ret)
2385 		return ret;
2386 
2387 #ifdef CONFIG_CPU_HOTPLUG_STATE_CONTROL
2388 	if (target < CPUHP_OFFLINE || target > CPUHP_ONLINE)
2389 		return -EINVAL;
2390 #else
2391 	if (target != CPUHP_OFFLINE && target != CPUHP_ONLINE)
2392 		return -EINVAL;
2393 #endif
2394 
2395 	ret = lock_device_hotplug_sysfs();
2396 	if (ret)
2397 		return ret;
2398 
2399 	mutex_lock(&cpuhp_state_mutex);
2400 	sp = cpuhp_get_step(target);
2401 	ret = !sp->name || sp->cant_stop ? -EINVAL : 0;
2402 	mutex_unlock(&cpuhp_state_mutex);
2403 	if (ret)
2404 		goto out;
2405 
2406 	if (st->state < target)
2407 		ret = cpu_up(dev->id, target);
2408 	else if (st->state > target)
2409 		ret = cpu_down(dev->id, target);
2410 	else if (WARN_ON(st->target != target))
2411 		st->target = target;
2412 out:
2413 	unlock_device_hotplug();
2414 	return ret ? ret : count;
2415 }
2416 
target_show(struct device * dev,struct device_attribute * attr,char * buf)2417 static ssize_t target_show(struct device *dev,
2418 			   struct device_attribute *attr, char *buf)
2419 {
2420 	struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id);
2421 
2422 	return sprintf(buf, "%d\n", st->target);
2423 }
2424 static DEVICE_ATTR_RW(target);
2425 
fail_store(struct device * dev,struct device_attribute * attr,const char * buf,size_t count)2426 static ssize_t fail_store(struct device *dev, struct device_attribute *attr,
2427 			  const char *buf, size_t count)
2428 {
2429 	struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id);
2430 	struct cpuhp_step *sp;
2431 	int fail, ret;
2432 
2433 	ret = kstrtoint(buf, 10, &fail);
2434 	if (ret)
2435 		return ret;
2436 
2437 	if (fail == CPUHP_INVALID) {
2438 		st->fail = fail;
2439 		return count;
2440 	}
2441 
2442 	if (fail < CPUHP_OFFLINE || fail > CPUHP_ONLINE)
2443 		return -EINVAL;
2444 
2445 	/*
2446 	 * Cannot fail STARTING/DYING callbacks.
2447 	 */
2448 	if (cpuhp_is_atomic_state(fail))
2449 		return -EINVAL;
2450 
2451 	/*
2452 	 * DEAD callbacks cannot fail...
2453 	 * ... neither can CPUHP_BRINGUP_CPU during hotunplug. The latter
2454 	 * triggering STARTING callbacks, a failure in this state would
2455 	 * hinder rollback.
2456 	 */
2457 	if (fail <= CPUHP_BRINGUP_CPU && st->state > CPUHP_BRINGUP_CPU)
2458 		return -EINVAL;
2459 
2460 	/*
2461 	 * Cannot fail anything that doesn't have callbacks.
2462 	 */
2463 	mutex_lock(&cpuhp_state_mutex);
2464 	sp = cpuhp_get_step(fail);
2465 	if (!sp->startup.single && !sp->teardown.single)
2466 		ret = -EINVAL;
2467 	mutex_unlock(&cpuhp_state_mutex);
2468 	if (ret)
2469 		return ret;
2470 
2471 	st->fail = fail;
2472 
2473 	return count;
2474 }
2475 
fail_show(struct device * dev,struct device_attribute * attr,char * buf)2476 static ssize_t fail_show(struct device *dev,
2477 			 struct device_attribute *attr, char *buf)
2478 {
2479 	struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id);
2480 
2481 	return sprintf(buf, "%d\n", st->fail);
2482 }
2483 
2484 static DEVICE_ATTR_RW(fail);
2485 
2486 static struct attribute *cpuhp_cpu_attrs[] = {
2487 	&dev_attr_state.attr,
2488 	&dev_attr_target.attr,
2489 	&dev_attr_fail.attr,
2490 	NULL
2491 };
2492 
2493 static const struct attribute_group cpuhp_cpu_attr_group = {
2494 	.attrs = cpuhp_cpu_attrs,
2495 	.name = "hotplug",
2496 	NULL
2497 };
2498 
states_show(struct device * dev,struct device_attribute * attr,char * buf)2499 static ssize_t states_show(struct device *dev,
2500 				 struct device_attribute *attr, char *buf)
2501 {
2502 	ssize_t cur, res = 0;
2503 	int i;
2504 
2505 	mutex_lock(&cpuhp_state_mutex);
2506 	for (i = CPUHP_OFFLINE; i <= CPUHP_ONLINE; i++) {
2507 		struct cpuhp_step *sp = cpuhp_get_step(i);
2508 
2509 		if (sp->name) {
2510 			cur = sprintf(buf, "%3d: %s\n", i, sp->name);
2511 			buf += cur;
2512 			res += cur;
2513 		}
2514 	}
2515 	mutex_unlock(&cpuhp_state_mutex);
2516 	return res;
2517 }
2518 static DEVICE_ATTR_RO(states);
2519 
2520 static struct attribute *cpuhp_cpu_root_attrs[] = {
2521 	&dev_attr_states.attr,
2522 	NULL
2523 };
2524 
2525 static const struct attribute_group cpuhp_cpu_root_attr_group = {
2526 	.attrs = cpuhp_cpu_root_attrs,
2527 	.name = "hotplug",
2528 	NULL
2529 };
2530 
2531 #ifdef CONFIG_HOTPLUG_SMT
2532 
2533 static ssize_t
__store_smt_control(struct device * dev,struct device_attribute * attr,const char * buf,size_t count)2534 __store_smt_control(struct device *dev, struct device_attribute *attr,
2535 		    const char *buf, size_t count)
2536 {
2537 	int ctrlval, ret;
2538 
2539 	if (sysfs_streq(buf, "on"))
2540 		ctrlval = CPU_SMT_ENABLED;
2541 	else if (sysfs_streq(buf, "off"))
2542 		ctrlval = CPU_SMT_DISABLED;
2543 	else if (sysfs_streq(buf, "forceoff"))
2544 		ctrlval = CPU_SMT_FORCE_DISABLED;
2545 	else
2546 		return -EINVAL;
2547 
2548 	if (cpu_smt_control == CPU_SMT_FORCE_DISABLED)
2549 		return -EPERM;
2550 
2551 	if (cpu_smt_control == CPU_SMT_NOT_SUPPORTED)
2552 		return -ENODEV;
2553 
2554 	ret = lock_device_hotplug_sysfs();
2555 	if (ret)
2556 		return ret;
2557 
2558 	if (ctrlval != cpu_smt_control) {
2559 		switch (ctrlval) {
2560 		case CPU_SMT_ENABLED:
2561 			ret = cpuhp_smt_enable();
2562 			break;
2563 		case CPU_SMT_DISABLED:
2564 		case CPU_SMT_FORCE_DISABLED:
2565 			ret = cpuhp_smt_disable(ctrlval);
2566 			break;
2567 		}
2568 	}
2569 
2570 	unlock_device_hotplug();
2571 	return ret ? ret : count;
2572 }
2573 
2574 #else /* !CONFIG_HOTPLUG_SMT */
2575 static ssize_t
__store_smt_control(struct device * dev,struct device_attribute * attr,const char * buf,size_t count)2576 __store_smt_control(struct device *dev, struct device_attribute *attr,
2577 		    const char *buf, size_t count)
2578 {
2579 	return -ENODEV;
2580 }
2581 #endif /* CONFIG_HOTPLUG_SMT */
2582 
2583 static const char *smt_states[] = {
2584 	[CPU_SMT_ENABLED]		= "on",
2585 	[CPU_SMT_DISABLED]		= "off",
2586 	[CPU_SMT_FORCE_DISABLED]	= "forceoff",
2587 	[CPU_SMT_NOT_SUPPORTED]		= "notsupported",
2588 	[CPU_SMT_NOT_IMPLEMENTED]	= "notimplemented",
2589 };
2590 
control_show(struct device * dev,struct device_attribute * attr,char * buf)2591 static ssize_t control_show(struct device *dev,
2592 			    struct device_attribute *attr, char *buf)
2593 {
2594 	const char *state = smt_states[cpu_smt_control];
2595 
2596 	return snprintf(buf, PAGE_SIZE - 2, "%s\n", state);
2597 }
2598 
control_store(struct device * dev,struct device_attribute * attr,const char * buf,size_t count)2599 static ssize_t control_store(struct device *dev, struct device_attribute *attr,
2600 			     const char *buf, size_t count)
2601 {
2602 	return __store_smt_control(dev, attr, buf, count);
2603 }
2604 static DEVICE_ATTR_RW(control);
2605 
active_show(struct device * dev,struct device_attribute * attr,char * buf)2606 static ssize_t active_show(struct device *dev,
2607 			   struct device_attribute *attr, char *buf)
2608 {
2609 	return snprintf(buf, PAGE_SIZE - 2, "%d\n", sched_smt_active());
2610 }
2611 static DEVICE_ATTR_RO(active);
2612 
2613 static struct attribute *cpuhp_smt_attrs[] = {
2614 	&dev_attr_control.attr,
2615 	&dev_attr_active.attr,
2616 	NULL
2617 };
2618 
2619 static const struct attribute_group cpuhp_smt_attr_group = {
2620 	.attrs = cpuhp_smt_attrs,
2621 	.name = "smt",
2622 	NULL
2623 };
2624 
cpu_smt_sysfs_init(void)2625 static int __init cpu_smt_sysfs_init(void)
2626 {
2627 	return sysfs_create_group(&cpu_subsys.dev_root->kobj,
2628 				  &cpuhp_smt_attr_group);
2629 }
2630 
cpuhp_sysfs_init(void)2631 static int __init cpuhp_sysfs_init(void)
2632 {
2633 	int cpu, ret;
2634 
2635 	ret = cpu_smt_sysfs_init();
2636 	if (ret)
2637 		return ret;
2638 
2639 	ret = sysfs_create_group(&cpu_subsys.dev_root->kobj,
2640 				 &cpuhp_cpu_root_attr_group);
2641 	if (ret)
2642 		return ret;
2643 
2644 	for_each_possible_cpu(cpu) {
2645 		struct device *dev = get_cpu_device(cpu);
2646 
2647 		if (!dev)
2648 			continue;
2649 		ret = sysfs_create_group(&dev->kobj, &cpuhp_cpu_attr_group);
2650 		if (ret)
2651 			return ret;
2652 	}
2653 	return 0;
2654 }
2655 device_initcall(cpuhp_sysfs_init);
2656 #endif /* CONFIG_SYSFS && CONFIG_HOTPLUG_CPU */
2657 
2658 /*
2659  * cpu_bit_bitmap[] is a special, "compressed" data structure that
2660  * represents all NR_CPUS bits binary values of 1<<nr.
2661  *
2662  * It is used by cpumask_of() to get a constant address to a CPU
2663  * mask value that has a single bit set only.
2664  */
2665 
2666 /* cpu_bit_bitmap[0] is empty - so we can back into it */
2667 #define MASK_DECLARE_1(x)	[x+1][0] = (1UL << (x))
2668 #define MASK_DECLARE_2(x)	MASK_DECLARE_1(x), MASK_DECLARE_1(x+1)
2669 #define MASK_DECLARE_4(x)	MASK_DECLARE_2(x), MASK_DECLARE_2(x+2)
2670 #define MASK_DECLARE_8(x)	MASK_DECLARE_4(x), MASK_DECLARE_4(x+4)
2671 
2672 const unsigned long cpu_bit_bitmap[BITS_PER_LONG+1][BITS_TO_LONGS(NR_CPUS)] = {
2673 
2674 	MASK_DECLARE_8(0),	MASK_DECLARE_8(8),
2675 	MASK_DECLARE_8(16),	MASK_DECLARE_8(24),
2676 #if BITS_PER_LONG > 32
2677 	MASK_DECLARE_8(32),	MASK_DECLARE_8(40),
2678 	MASK_DECLARE_8(48),	MASK_DECLARE_8(56),
2679 #endif
2680 };
2681 EXPORT_SYMBOL_GPL(cpu_bit_bitmap);
2682 
2683 const DECLARE_BITMAP(cpu_all_bits, NR_CPUS) = CPU_BITS_ALL;
2684 EXPORT_SYMBOL(cpu_all_bits);
2685 
2686 #ifdef CONFIG_INIT_ALL_POSSIBLE
2687 struct cpumask __cpu_possible_mask __read_mostly
2688 	= {CPU_BITS_ALL};
2689 #else
2690 struct cpumask __cpu_possible_mask __read_mostly;
2691 #endif
2692 EXPORT_SYMBOL(__cpu_possible_mask);
2693 
2694 struct cpumask __cpu_online_mask __read_mostly;
2695 EXPORT_SYMBOL(__cpu_online_mask);
2696 
2697 struct cpumask __cpu_present_mask __read_mostly;
2698 EXPORT_SYMBOL(__cpu_present_mask);
2699 
2700 struct cpumask __cpu_active_mask __read_mostly;
2701 EXPORT_SYMBOL(__cpu_active_mask);
2702 
2703 struct cpumask __cpu_dying_mask __read_mostly;
2704 EXPORT_SYMBOL(__cpu_dying_mask);
2705 
2706 atomic_t __num_online_cpus __read_mostly;
2707 EXPORT_SYMBOL(__num_online_cpus);
2708 
init_cpu_present(const struct cpumask * src)2709 void init_cpu_present(const struct cpumask *src)
2710 {
2711 	cpumask_copy(&__cpu_present_mask, src);
2712 }
2713 
init_cpu_possible(const struct cpumask * src)2714 void init_cpu_possible(const struct cpumask *src)
2715 {
2716 	cpumask_copy(&__cpu_possible_mask, src);
2717 }
2718 
init_cpu_online(const struct cpumask * src)2719 void init_cpu_online(const struct cpumask *src)
2720 {
2721 	cpumask_copy(&__cpu_online_mask, src);
2722 }
2723 
set_cpu_online(unsigned int cpu,bool online)2724 void set_cpu_online(unsigned int cpu, bool online)
2725 {
2726 	/*
2727 	 * atomic_inc/dec() is required to handle the horrid abuse of this
2728 	 * function by the reboot and kexec code which invoke it from
2729 	 * IPI/NMI broadcasts when shutting down CPUs. Invocation from
2730 	 * regular CPU hotplug is properly serialized.
2731 	 *
2732 	 * Note, that the fact that __num_online_cpus is of type atomic_t
2733 	 * does not protect readers which are not serialized against
2734 	 * concurrent hotplug operations.
2735 	 */
2736 	if (online) {
2737 		if (!cpumask_test_and_set_cpu(cpu, &__cpu_online_mask))
2738 			atomic_inc(&__num_online_cpus);
2739 	} else {
2740 		if (cpumask_test_and_clear_cpu(cpu, &__cpu_online_mask))
2741 			atomic_dec(&__num_online_cpus);
2742 	}
2743 }
2744 
2745 /*
2746  * Activate the first processor.
2747  */
boot_cpu_init(void)2748 void __init boot_cpu_init(void)
2749 {
2750 	int cpu = smp_processor_id();
2751 
2752 	/* Mark the boot cpu "present", "online" etc for SMP and UP case */
2753 	set_cpu_online(cpu, true);
2754 	set_cpu_active(cpu, true);
2755 	set_cpu_present(cpu, true);
2756 	set_cpu_possible(cpu, true);
2757 
2758 #ifdef CONFIG_SMP
2759 	__boot_cpu_id = cpu;
2760 #endif
2761 }
2762 
2763 /*
2764  * Must be called _AFTER_ setting up the per_cpu areas
2765  */
boot_cpu_hotplug_init(void)2766 void __init boot_cpu_hotplug_init(void)
2767 {
2768 #ifdef CONFIG_SMP
2769 	cpumask_set_cpu(smp_processor_id(), &cpus_booted_once_mask);
2770 #endif
2771 	this_cpu_write(cpuhp_state.state, CPUHP_ONLINE);
2772 }
2773 
2774 /*
2775  * These are used for a global "mitigations=" cmdline option for toggling
2776  * optional CPU mitigations.
2777  */
2778 enum cpu_mitigations {
2779 	CPU_MITIGATIONS_OFF,
2780 	CPU_MITIGATIONS_AUTO,
2781 	CPU_MITIGATIONS_AUTO_NOSMT,
2782 };
2783 
2784 static enum cpu_mitigations cpu_mitigations __ro_after_init =
2785 	CPU_MITIGATIONS_AUTO;
2786 
mitigations_parse_cmdline(char * arg)2787 static int __init mitigations_parse_cmdline(char *arg)
2788 {
2789 	if (!strcmp(arg, "off"))
2790 		cpu_mitigations = CPU_MITIGATIONS_OFF;
2791 	else if (!strcmp(arg, "auto"))
2792 		cpu_mitigations = CPU_MITIGATIONS_AUTO;
2793 	else if (!strcmp(arg, "auto,nosmt"))
2794 		cpu_mitigations = CPU_MITIGATIONS_AUTO_NOSMT;
2795 	else
2796 		pr_crit("Unsupported mitigations=%s, system may still be vulnerable\n",
2797 			arg);
2798 
2799 	return 0;
2800 }
2801 early_param("mitigations", mitigations_parse_cmdline);
2802 
2803 /* mitigations=off */
cpu_mitigations_off(void)2804 bool cpu_mitigations_off(void)
2805 {
2806 	return cpu_mitigations == CPU_MITIGATIONS_OFF;
2807 }
2808 EXPORT_SYMBOL_GPL(cpu_mitigations_off);
2809 
2810 /* mitigations=auto,nosmt */
cpu_mitigations_auto_nosmt(void)2811 bool cpu_mitigations_auto_nosmt(void)
2812 {
2813 	return cpu_mitigations == CPU_MITIGATIONS_AUTO_NOSMT;
2814 }
2815 EXPORT_SYMBOL_GPL(cpu_mitigations_auto_nosmt);
2816