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/delay.h>
21 #include <linux/export.h>
22 #include <linux/bug.h>
23 #include <linux/kthread.h>
24 #include <linux/stop_machine.h>
25 #include <linux/mutex.h>
26 #include <linux/gfp.h>
27 #include <linux/suspend.h>
28 #include <linux/lockdep.h>
29 #include <linux/tick.h>
30 #include <linux/irq.h>
31 #include <linux/nmi.h>
32 #include <linux/smpboot.h>
33 #include <linux/relay.h>
34 #include <linux/slab.h>
35 #include <linux/scs.h>
36 #include <linux/percpu-rwsem.h>
37 #include <linux/cpuset.h>
38 #include <linux/random.h>
39 #include <linux/cc_platform.h>
40 
41 #include <trace/events/power.h>
42 #define CREATE_TRACE_POINTS
43 #include <trace/events/cpuhp.h>
44 
45 #include "smpboot.h"
46 
47 /**
48  * struct cpuhp_cpu_state - Per cpu hotplug state storage
49  * @state:	The current cpu state
50  * @target:	The target state
51  * @fail:	Current CPU hotplug callback state
52  * @thread:	Pointer to the hotplug thread
53  * @should_run:	Thread should execute
54  * @rollback:	Perform a rollback
55  * @single:	Single callback invocation
56  * @bringup:	Single callback bringup or teardown selector
57  * @cpu:	CPU number
58  * @node:	Remote CPU node; for multi-instance, do a
59  *		single entry callback for install/remove
60  * @last:	For multi-instance rollback, remember how far we got
61  * @cb_state:	The state for a single callback (install/uninstall)
62  * @result:	Result of the operation
63  * @ap_sync_state:	State for AP synchronization
64  * @done_up:	Signal completion to the issuer of the task for cpu-up
65  * @done_down:	Signal completion to the issuer of the task for cpu-down
66  */
67 struct cpuhp_cpu_state {
68 	enum cpuhp_state	state;
69 	enum cpuhp_state	target;
70 	enum cpuhp_state	fail;
71 #ifdef CONFIG_SMP
72 	struct task_struct	*thread;
73 	bool			should_run;
74 	bool			rollback;
75 	bool			single;
76 	bool			bringup;
77 	struct hlist_node	*node;
78 	struct hlist_node	*last;
79 	enum cpuhp_state	cb_state;
80 	int			result;
81 	atomic_t		ap_sync_state;
82 	struct completion	done_up;
83 	struct completion	done_down;
84 #endif
85 };
86 
87 static DEFINE_PER_CPU(struct cpuhp_cpu_state, cpuhp_state) = {
88 	.fail = CPUHP_INVALID,
89 };
90 
91 #ifdef CONFIG_SMP
92 cpumask_t cpus_booted_once_mask;
93 #endif
94 
95 #if defined(CONFIG_LOCKDEP) && defined(CONFIG_SMP)
96 static struct lockdep_map cpuhp_state_up_map =
97 	STATIC_LOCKDEP_MAP_INIT("cpuhp_state-up", &cpuhp_state_up_map);
98 static struct lockdep_map cpuhp_state_down_map =
99 	STATIC_LOCKDEP_MAP_INIT("cpuhp_state-down", &cpuhp_state_down_map);
100 
101 
cpuhp_lock_acquire(bool bringup)102 static inline void cpuhp_lock_acquire(bool bringup)
103 {
104 	lock_map_acquire(bringup ? &cpuhp_state_up_map : &cpuhp_state_down_map);
105 }
106 
cpuhp_lock_release(bool bringup)107 static inline void cpuhp_lock_release(bool bringup)
108 {
109 	lock_map_release(bringup ? &cpuhp_state_up_map : &cpuhp_state_down_map);
110 }
111 #else
112 
cpuhp_lock_acquire(bool bringup)113 static inline void cpuhp_lock_acquire(bool bringup) { }
cpuhp_lock_release(bool bringup)114 static inline void cpuhp_lock_release(bool bringup) { }
115 
116 #endif
117 
118 /**
119  * struct cpuhp_step - Hotplug state machine step
120  * @name:	Name of the step
121  * @startup:	Startup function of the step
122  * @teardown:	Teardown function of the step
123  * @cant_stop:	Bringup/teardown can't be stopped at this step
124  * @multi_instance:	State has multiple instances which get added afterwards
125  */
126 struct cpuhp_step {
127 	const char		*name;
128 	union {
129 		int		(*single)(unsigned int cpu);
130 		int		(*multi)(unsigned int cpu,
131 					 struct hlist_node *node);
132 	} startup;
133 	union {
134 		int		(*single)(unsigned int cpu);
135 		int		(*multi)(unsigned int cpu,
136 					 struct hlist_node *node);
137 	} teardown;
138 	/* private: */
139 	struct hlist_head	list;
140 	/* public: */
141 	bool			cant_stop;
142 	bool			multi_instance;
143 };
144 
145 static DEFINE_MUTEX(cpuhp_state_mutex);
146 static struct cpuhp_step cpuhp_hp_states[];
147 
cpuhp_get_step(enum cpuhp_state state)148 static struct cpuhp_step *cpuhp_get_step(enum cpuhp_state state)
149 {
150 	return cpuhp_hp_states + state;
151 }
152 
cpuhp_step_empty(bool bringup,struct cpuhp_step * step)153 static bool cpuhp_step_empty(bool bringup, struct cpuhp_step *step)
154 {
155 	return bringup ? !step->startup.single : !step->teardown.single;
156 }
157 
158 /**
159  * cpuhp_invoke_callback - Invoke the callbacks for a given state
160  * @cpu:	The cpu for which the callback should be invoked
161  * @state:	The state to do callbacks for
162  * @bringup:	True if the bringup callback should be invoked
163  * @node:	For multi-instance, do a single entry callback for install/remove
164  * @lastp:	For multi-instance rollback, remember how far we got
165  *
166  * Called from cpu hotplug and from the state register machinery.
167  *
168  * Return: %0 on success or a negative errno code
169  */
cpuhp_invoke_callback(unsigned int cpu,enum cpuhp_state state,bool bringup,struct hlist_node * node,struct hlist_node ** lastp)170 static int cpuhp_invoke_callback(unsigned int cpu, enum cpuhp_state state,
171 				 bool bringup, struct hlist_node *node,
172 				 struct hlist_node **lastp)
173 {
174 	struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
175 	struct cpuhp_step *step = cpuhp_get_step(state);
176 	int (*cbm)(unsigned int cpu, struct hlist_node *node);
177 	int (*cb)(unsigned int cpu);
178 	int ret, cnt;
179 
180 	if (st->fail == state) {
181 		st->fail = CPUHP_INVALID;
182 		return -EAGAIN;
183 	}
184 
185 	if (cpuhp_step_empty(bringup, step)) {
186 		WARN_ON_ONCE(1);
187 		return 0;
188 	}
189 
190 	if (!step->multi_instance) {
191 		WARN_ON_ONCE(lastp && *lastp);
192 		cb = bringup ? step->startup.single : step->teardown.single;
193 
194 		trace_cpuhp_enter(cpu, st->target, state, cb);
195 		ret = cb(cpu);
196 		trace_cpuhp_exit(cpu, st->state, state, ret);
197 		return ret;
198 	}
199 	cbm = bringup ? step->startup.multi : step->teardown.multi;
200 
201 	/* Single invocation for instance add/remove */
202 	if (node) {
203 		WARN_ON_ONCE(lastp && *lastp);
204 		trace_cpuhp_multi_enter(cpu, st->target, state, cbm, node);
205 		ret = cbm(cpu, node);
206 		trace_cpuhp_exit(cpu, st->state, state, ret);
207 		return ret;
208 	}
209 
210 	/* State transition. Invoke on all instances */
211 	cnt = 0;
212 	hlist_for_each(node, &step->list) {
213 		if (lastp && node == *lastp)
214 			break;
215 
216 		trace_cpuhp_multi_enter(cpu, st->target, state, cbm, node);
217 		ret = cbm(cpu, node);
218 		trace_cpuhp_exit(cpu, st->state, state, ret);
219 		if (ret) {
220 			if (!lastp)
221 				goto err;
222 
223 			*lastp = node;
224 			return ret;
225 		}
226 		cnt++;
227 	}
228 	if (lastp)
229 		*lastp = NULL;
230 	return 0;
231 err:
232 	/* Rollback the instances if one failed */
233 	cbm = !bringup ? step->startup.multi : step->teardown.multi;
234 	if (!cbm)
235 		return ret;
236 
237 	hlist_for_each(node, &step->list) {
238 		if (!cnt--)
239 			break;
240 
241 		trace_cpuhp_multi_enter(cpu, st->target, state, cbm, node);
242 		ret = cbm(cpu, node);
243 		trace_cpuhp_exit(cpu, st->state, state, ret);
244 		/*
245 		 * Rollback must not fail,
246 		 */
247 		WARN_ON_ONCE(ret);
248 	}
249 	return ret;
250 }
251 
252 #ifdef CONFIG_SMP
cpuhp_is_ap_state(enum cpuhp_state state)253 static bool cpuhp_is_ap_state(enum cpuhp_state state)
254 {
255 	/*
256 	 * The extra check for CPUHP_TEARDOWN_CPU is only for documentation
257 	 * purposes as that state is handled explicitly in cpu_down.
258 	 */
259 	return state > CPUHP_BRINGUP_CPU && state != CPUHP_TEARDOWN_CPU;
260 }
261 
wait_for_ap_thread(struct cpuhp_cpu_state * st,bool bringup)262 static inline void wait_for_ap_thread(struct cpuhp_cpu_state *st, bool bringup)
263 {
264 	struct completion *done = bringup ? &st->done_up : &st->done_down;
265 	wait_for_completion(done);
266 }
267 
complete_ap_thread(struct cpuhp_cpu_state * st,bool bringup)268 static inline void complete_ap_thread(struct cpuhp_cpu_state *st, bool bringup)
269 {
270 	struct completion *done = bringup ? &st->done_up : &st->done_down;
271 	complete(done);
272 }
273 
274 /*
275  * The former STARTING/DYING states, ran with IRQs disabled and must not fail.
276  */
cpuhp_is_atomic_state(enum cpuhp_state state)277 static bool cpuhp_is_atomic_state(enum cpuhp_state state)
278 {
279 	return CPUHP_AP_IDLE_DEAD <= state && state < CPUHP_AP_ONLINE;
280 }
281 
282 /* Synchronization state management */
283 enum cpuhp_sync_state {
284 	SYNC_STATE_DEAD,
285 	SYNC_STATE_KICKED,
286 	SYNC_STATE_SHOULD_DIE,
287 	SYNC_STATE_ALIVE,
288 	SYNC_STATE_SHOULD_ONLINE,
289 	SYNC_STATE_ONLINE,
290 };
291 
292 #ifdef CONFIG_HOTPLUG_CORE_SYNC
293 /**
294  * cpuhp_ap_update_sync_state - Update synchronization state during bringup/teardown
295  * @state:	The synchronization state to set
296  *
297  * No synchronization point. Just update of the synchronization state, but implies
298  * a full barrier so that the AP changes are visible before the control CPU proceeds.
299  */
cpuhp_ap_update_sync_state(enum cpuhp_sync_state state)300 static inline void cpuhp_ap_update_sync_state(enum cpuhp_sync_state state)
301 {
302 	atomic_t *st = this_cpu_ptr(&cpuhp_state.ap_sync_state);
303 
304 	(void)atomic_xchg(st, state);
305 }
306 
arch_cpuhp_sync_state_poll(void)307 void __weak arch_cpuhp_sync_state_poll(void) { cpu_relax(); }
308 
cpuhp_wait_for_sync_state(unsigned int cpu,enum cpuhp_sync_state state,enum cpuhp_sync_state next_state)309 static bool cpuhp_wait_for_sync_state(unsigned int cpu, enum cpuhp_sync_state state,
310 				      enum cpuhp_sync_state next_state)
311 {
312 	atomic_t *st = per_cpu_ptr(&cpuhp_state.ap_sync_state, cpu);
313 	ktime_t now, end, start = ktime_get();
314 	int sync;
315 
316 	end = start + 10ULL * NSEC_PER_SEC;
317 
318 	sync = atomic_read(st);
319 	while (1) {
320 		if (sync == state) {
321 			if (!atomic_try_cmpxchg(st, &sync, next_state))
322 				continue;
323 			return true;
324 		}
325 
326 		now = ktime_get();
327 		if (now > end) {
328 			/* Timeout. Leave the state unchanged */
329 			return false;
330 		} else if (now - start < NSEC_PER_MSEC) {
331 			/* Poll for one millisecond */
332 			arch_cpuhp_sync_state_poll();
333 		} else {
334 			usleep_range_state(USEC_PER_MSEC, 2 * USEC_PER_MSEC, TASK_UNINTERRUPTIBLE);
335 		}
336 		sync = atomic_read(st);
337 	}
338 	return true;
339 }
340 #else  /* CONFIG_HOTPLUG_CORE_SYNC */
cpuhp_ap_update_sync_state(enum cpuhp_sync_state state)341 static inline void cpuhp_ap_update_sync_state(enum cpuhp_sync_state state) { }
342 #endif /* !CONFIG_HOTPLUG_CORE_SYNC */
343 
344 #ifdef CONFIG_HOTPLUG_CORE_SYNC_DEAD
345 /**
346  * cpuhp_ap_report_dead - Update synchronization state to DEAD
347  *
348  * No synchronization point. Just update of the synchronization state.
349  */
cpuhp_ap_report_dead(void)350 void cpuhp_ap_report_dead(void)
351 {
352 	cpuhp_ap_update_sync_state(SYNC_STATE_DEAD);
353 }
354 
arch_cpuhp_cleanup_dead_cpu(unsigned int cpu)355 void __weak arch_cpuhp_cleanup_dead_cpu(unsigned int cpu) { }
356 
357 /*
358  * Late CPU shutdown synchronization point. Cannot use cpuhp_state::done_down
359  * because the AP cannot issue complete() at this stage.
360  */
cpuhp_bp_sync_dead(unsigned int cpu)361 static void cpuhp_bp_sync_dead(unsigned int cpu)
362 {
363 	atomic_t *st = per_cpu_ptr(&cpuhp_state.ap_sync_state, cpu);
364 	int sync = atomic_read(st);
365 
366 	do {
367 		/* CPU can have reported dead already. Don't overwrite that! */
368 		if (sync == SYNC_STATE_DEAD)
369 			break;
370 	} while (!atomic_try_cmpxchg(st, &sync, SYNC_STATE_SHOULD_DIE));
371 
372 	if (cpuhp_wait_for_sync_state(cpu, SYNC_STATE_DEAD, SYNC_STATE_DEAD)) {
373 		/* CPU reached dead state. Invoke the cleanup function */
374 		arch_cpuhp_cleanup_dead_cpu(cpu);
375 		return;
376 	}
377 
378 	/* No further action possible. Emit message and give up. */
379 	pr_err("CPU%u failed to report dead state\n", cpu);
380 }
381 #else /* CONFIG_HOTPLUG_CORE_SYNC_DEAD */
cpuhp_bp_sync_dead(unsigned int cpu)382 static inline void cpuhp_bp_sync_dead(unsigned int cpu) { }
383 #endif /* !CONFIG_HOTPLUG_CORE_SYNC_DEAD */
384 
385 #ifdef CONFIG_HOTPLUG_CORE_SYNC_FULL
386 /**
387  * cpuhp_ap_sync_alive - Synchronize AP with the control CPU once it is alive
388  *
389  * Updates the AP synchronization state to SYNC_STATE_ALIVE and waits
390  * for the BP to release it.
391  */
cpuhp_ap_sync_alive(void)392 void cpuhp_ap_sync_alive(void)
393 {
394 	atomic_t *st = this_cpu_ptr(&cpuhp_state.ap_sync_state);
395 
396 	cpuhp_ap_update_sync_state(SYNC_STATE_ALIVE);
397 
398 	/* Wait for the control CPU to release it. */
399 	while (atomic_read(st) != SYNC_STATE_SHOULD_ONLINE)
400 		cpu_relax();
401 }
402 
cpuhp_can_boot_ap(unsigned int cpu)403 static bool cpuhp_can_boot_ap(unsigned int cpu)
404 {
405 	atomic_t *st = per_cpu_ptr(&cpuhp_state.ap_sync_state, cpu);
406 	int sync = atomic_read(st);
407 
408 again:
409 	switch (sync) {
410 	case SYNC_STATE_DEAD:
411 		/* CPU is properly dead */
412 		break;
413 	case SYNC_STATE_KICKED:
414 		/* CPU did not come up in previous attempt */
415 		break;
416 	case SYNC_STATE_ALIVE:
417 		/* CPU is stuck cpuhp_ap_sync_alive(). */
418 		break;
419 	default:
420 		/* CPU failed to report online or dead and is in limbo state. */
421 		return false;
422 	}
423 
424 	/* Prepare for booting */
425 	if (!atomic_try_cmpxchg(st, &sync, SYNC_STATE_KICKED))
426 		goto again;
427 
428 	return true;
429 }
430 
arch_cpuhp_cleanup_kick_cpu(unsigned int cpu)431 void __weak arch_cpuhp_cleanup_kick_cpu(unsigned int cpu) { }
432 
433 /*
434  * Early CPU bringup synchronization point. Cannot use cpuhp_state::done_up
435  * because the AP cannot issue complete() so early in the bringup.
436  */
cpuhp_bp_sync_alive(unsigned int cpu)437 static int cpuhp_bp_sync_alive(unsigned int cpu)
438 {
439 	int ret = 0;
440 
441 	if (!IS_ENABLED(CONFIG_HOTPLUG_CORE_SYNC_FULL))
442 		return 0;
443 
444 	if (!cpuhp_wait_for_sync_state(cpu, SYNC_STATE_ALIVE, SYNC_STATE_SHOULD_ONLINE)) {
445 		pr_err("CPU%u failed to report alive state\n", cpu);
446 		ret = -EIO;
447 	}
448 
449 	/* Let the architecture cleanup the kick alive mechanics. */
450 	arch_cpuhp_cleanup_kick_cpu(cpu);
451 	return ret;
452 }
453 #else /* CONFIG_HOTPLUG_CORE_SYNC_FULL */
cpuhp_bp_sync_alive(unsigned int cpu)454 static inline int cpuhp_bp_sync_alive(unsigned int cpu) { return 0; }
cpuhp_can_boot_ap(unsigned int cpu)455 static inline bool cpuhp_can_boot_ap(unsigned int cpu) { return true; }
456 #endif /* !CONFIG_HOTPLUG_CORE_SYNC_FULL */
457 
458 /* Serializes the updates to cpu_online_mask, cpu_present_mask */
459 static DEFINE_MUTEX(cpu_add_remove_lock);
460 bool cpuhp_tasks_frozen;
461 EXPORT_SYMBOL_GPL(cpuhp_tasks_frozen);
462 
463 /*
464  * The following two APIs (cpu_maps_update_begin/done) must be used when
465  * attempting to serialize the updates to cpu_online_mask & cpu_present_mask.
466  */
cpu_maps_update_begin(void)467 void cpu_maps_update_begin(void)
468 {
469 	mutex_lock(&cpu_add_remove_lock);
470 }
471 
cpu_maps_update_done(void)472 void cpu_maps_update_done(void)
473 {
474 	mutex_unlock(&cpu_add_remove_lock);
475 }
476 
477 /*
478  * If set, cpu_up and cpu_down will return -EBUSY and do nothing.
479  * Should always be manipulated under cpu_add_remove_lock
480  */
481 static int cpu_hotplug_disabled;
482 
483 #ifdef CONFIG_HOTPLUG_CPU
484 
485 DEFINE_STATIC_PERCPU_RWSEM(cpu_hotplug_lock);
486 
cpus_read_lock(void)487 void cpus_read_lock(void)
488 {
489 	percpu_down_read(&cpu_hotplug_lock);
490 }
491 EXPORT_SYMBOL_GPL(cpus_read_lock);
492 
cpus_read_trylock(void)493 int cpus_read_trylock(void)
494 {
495 	return percpu_down_read_trylock(&cpu_hotplug_lock);
496 }
497 EXPORT_SYMBOL_GPL(cpus_read_trylock);
498 
cpus_read_unlock(void)499 void cpus_read_unlock(void)
500 {
501 	percpu_up_read(&cpu_hotplug_lock);
502 }
503 EXPORT_SYMBOL_GPL(cpus_read_unlock);
504 
cpus_write_lock(void)505 void cpus_write_lock(void)
506 {
507 	percpu_down_write(&cpu_hotplug_lock);
508 }
509 
cpus_write_unlock(void)510 void cpus_write_unlock(void)
511 {
512 	percpu_up_write(&cpu_hotplug_lock);
513 }
514 
lockdep_assert_cpus_held(void)515 void lockdep_assert_cpus_held(void)
516 {
517 	/*
518 	 * We can't have hotplug operations before userspace starts running,
519 	 * and some init codepaths will knowingly not take the hotplug lock.
520 	 * This is all valid, so mute lockdep until it makes sense to report
521 	 * unheld locks.
522 	 */
523 	if (system_state < SYSTEM_RUNNING)
524 		return;
525 
526 	percpu_rwsem_assert_held(&cpu_hotplug_lock);
527 }
528 
529 #ifdef CONFIG_LOCKDEP
lockdep_is_cpus_held(void)530 int lockdep_is_cpus_held(void)
531 {
532 	return percpu_rwsem_is_held(&cpu_hotplug_lock);
533 }
534 #endif
535 
lockdep_acquire_cpus_lock(void)536 static void lockdep_acquire_cpus_lock(void)
537 {
538 	rwsem_acquire(&cpu_hotplug_lock.dep_map, 0, 0, _THIS_IP_);
539 }
540 
lockdep_release_cpus_lock(void)541 static void lockdep_release_cpus_lock(void)
542 {
543 	rwsem_release(&cpu_hotplug_lock.dep_map, _THIS_IP_);
544 }
545 
546 /*
547  * Wait for currently running CPU hotplug operations to complete (if any) and
548  * disable future CPU hotplug (from sysfs). The 'cpu_add_remove_lock' protects
549  * the 'cpu_hotplug_disabled' flag. The same lock is also acquired by the
550  * hotplug path before performing hotplug operations. So acquiring that lock
551  * guarantees mutual exclusion from any currently running hotplug operations.
552  */
cpu_hotplug_disable(void)553 void cpu_hotplug_disable(void)
554 {
555 	cpu_maps_update_begin();
556 	cpu_hotplug_disabled++;
557 	cpu_maps_update_done();
558 }
559 EXPORT_SYMBOL_GPL(cpu_hotplug_disable);
560 
__cpu_hotplug_enable(void)561 static void __cpu_hotplug_enable(void)
562 {
563 	if (WARN_ONCE(!cpu_hotplug_disabled, "Unbalanced cpu hotplug enable\n"))
564 		return;
565 	cpu_hotplug_disabled--;
566 }
567 
cpu_hotplug_enable(void)568 void cpu_hotplug_enable(void)
569 {
570 	cpu_maps_update_begin();
571 	__cpu_hotplug_enable();
572 	cpu_maps_update_done();
573 }
574 EXPORT_SYMBOL_GPL(cpu_hotplug_enable);
575 
576 #else
577 
lockdep_acquire_cpus_lock(void)578 static void lockdep_acquire_cpus_lock(void)
579 {
580 }
581 
lockdep_release_cpus_lock(void)582 static void lockdep_release_cpus_lock(void)
583 {
584 }
585 
586 #endif	/* CONFIG_HOTPLUG_CPU */
587 
588 /*
589  * Architectures that need SMT-specific errata handling during SMT hotplug
590  * should override this.
591  */
arch_smt_update(void)592 void __weak arch_smt_update(void) { }
593 
594 #ifdef CONFIG_HOTPLUG_SMT
595 
596 enum cpuhp_smt_control cpu_smt_control __read_mostly = CPU_SMT_ENABLED;
597 static unsigned int cpu_smt_max_threads __ro_after_init;
598 unsigned int cpu_smt_num_threads __read_mostly = UINT_MAX;
599 
cpu_smt_disable(bool force)600 void __init cpu_smt_disable(bool force)
601 {
602 	if (!cpu_smt_possible())
603 		return;
604 
605 	if (force) {
606 		pr_info("SMT: Force disabled\n");
607 		cpu_smt_control = CPU_SMT_FORCE_DISABLED;
608 	} else {
609 		pr_info("SMT: disabled\n");
610 		cpu_smt_control = CPU_SMT_DISABLED;
611 	}
612 	cpu_smt_num_threads = 1;
613 }
614 
615 /*
616  * The decision whether SMT is supported can only be done after the full
617  * CPU identification. Called from architecture code.
618  */
cpu_smt_set_num_threads(unsigned int num_threads,unsigned int max_threads)619 void __init cpu_smt_set_num_threads(unsigned int num_threads,
620 				    unsigned int max_threads)
621 {
622 	WARN_ON(!num_threads || (num_threads > max_threads));
623 
624 	if (max_threads == 1)
625 		cpu_smt_control = CPU_SMT_NOT_SUPPORTED;
626 
627 	cpu_smt_max_threads = max_threads;
628 
629 	/*
630 	 * If SMT has been disabled via the kernel command line or SMT is
631 	 * not supported, set cpu_smt_num_threads to 1 for consistency.
632 	 * If enabled, take the architecture requested number of threads
633 	 * to bring up into account.
634 	 */
635 	if (cpu_smt_control != CPU_SMT_ENABLED)
636 		cpu_smt_num_threads = 1;
637 	else if (num_threads < cpu_smt_num_threads)
638 		cpu_smt_num_threads = num_threads;
639 }
640 
smt_cmdline_disable(char * str)641 static int __init smt_cmdline_disable(char *str)
642 {
643 	cpu_smt_disable(str && !strcmp(str, "force"));
644 	return 0;
645 }
646 early_param("nosmt", smt_cmdline_disable);
647 
648 /*
649  * For Archicture supporting partial SMT states check if the thread is allowed.
650  * Otherwise this has already been checked through cpu_smt_max_threads when
651  * setting the SMT level.
652  */
cpu_smt_thread_allowed(unsigned int cpu)653 static inline bool cpu_smt_thread_allowed(unsigned int cpu)
654 {
655 #ifdef CONFIG_SMT_NUM_THREADS_DYNAMIC
656 	return topology_smt_thread_allowed(cpu);
657 #else
658 	return true;
659 #endif
660 }
661 
cpu_bootable(unsigned int cpu)662 static inline bool cpu_bootable(unsigned int cpu)
663 {
664 	if (cpu_smt_control == CPU_SMT_ENABLED && cpu_smt_thread_allowed(cpu))
665 		return true;
666 
667 	/* All CPUs are bootable if controls are not configured */
668 	if (cpu_smt_control == CPU_SMT_NOT_IMPLEMENTED)
669 		return true;
670 
671 	/* All CPUs are bootable if CPU is not SMT capable */
672 	if (cpu_smt_control == CPU_SMT_NOT_SUPPORTED)
673 		return true;
674 
675 	if (topology_is_primary_thread(cpu))
676 		return true;
677 
678 	/*
679 	 * On x86 it's required to boot all logical CPUs at least once so
680 	 * that the init code can get a chance to set CR4.MCE on each
681 	 * CPU. Otherwise, a broadcasted MCE observing CR4.MCE=0b on any
682 	 * core will shutdown the machine.
683 	 */
684 	return !cpumask_test_cpu(cpu, &cpus_booted_once_mask);
685 }
686 
687 /* Returns true if SMT is supported and not forcefully (irreversibly) disabled */
cpu_smt_possible(void)688 bool cpu_smt_possible(void)
689 {
690 	return cpu_smt_control != CPU_SMT_FORCE_DISABLED &&
691 		cpu_smt_control != CPU_SMT_NOT_SUPPORTED;
692 }
693 EXPORT_SYMBOL_GPL(cpu_smt_possible);
694 
695 #else
cpu_bootable(unsigned int cpu)696 static inline bool cpu_bootable(unsigned int cpu) { return true; }
697 #endif
698 
699 static inline enum cpuhp_state
cpuhp_set_state(int cpu,struct cpuhp_cpu_state * st,enum cpuhp_state target)700 cpuhp_set_state(int cpu, struct cpuhp_cpu_state *st, enum cpuhp_state target)
701 {
702 	enum cpuhp_state prev_state = st->state;
703 	bool bringup = st->state < target;
704 
705 	st->rollback = false;
706 	st->last = NULL;
707 
708 	st->target = target;
709 	st->single = false;
710 	st->bringup = bringup;
711 	if (cpu_dying(cpu) != !bringup)
712 		set_cpu_dying(cpu, !bringup);
713 
714 	return prev_state;
715 }
716 
717 static inline void
cpuhp_reset_state(int cpu,struct cpuhp_cpu_state * st,enum cpuhp_state prev_state)718 cpuhp_reset_state(int cpu, struct cpuhp_cpu_state *st,
719 		  enum cpuhp_state prev_state)
720 {
721 	bool bringup = !st->bringup;
722 
723 	st->target = prev_state;
724 
725 	/*
726 	 * Already rolling back. No need invert the bringup value or to change
727 	 * the current state.
728 	 */
729 	if (st->rollback)
730 		return;
731 
732 	st->rollback = true;
733 
734 	/*
735 	 * If we have st->last we need to undo partial multi_instance of this
736 	 * state first. Otherwise start undo at the previous state.
737 	 */
738 	if (!st->last) {
739 		if (st->bringup)
740 			st->state--;
741 		else
742 			st->state++;
743 	}
744 
745 	st->bringup = bringup;
746 	if (cpu_dying(cpu) != !bringup)
747 		set_cpu_dying(cpu, !bringup);
748 }
749 
750 /* Regular hotplug invocation of the AP hotplug thread */
__cpuhp_kick_ap(struct cpuhp_cpu_state * st)751 static void __cpuhp_kick_ap(struct cpuhp_cpu_state *st)
752 {
753 	if (!st->single && st->state == st->target)
754 		return;
755 
756 	st->result = 0;
757 	/*
758 	 * Make sure the above stores are visible before should_run becomes
759 	 * true. Paired with the mb() above in cpuhp_thread_fun()
760 	 */
761 	smp_mb();
762 	st->should_run = true;
763 	wake_up_process(st->thread);
764 	wait_for_ap_thread(st, st->bringup);
765 }
766 
cpuhp_kick_ap(int cpu,struct cpuhp_cpu_state * st,enum cpuhp_state target)767 static int cpuhp_kick_ap(int cpu, struct cpuhp_cpu_state *st,
768 			 enum cpuhp_state target)
769 {
770 	enum cpuhp_state prev_state;
771 	int ret;
772 
773 	prev_state = cpuhp_set_state(cpu, st, target);
774 	__cpuhp_kick_ap(st);
775 	if ((ret = st->result)) {
776 		cpuhp_reset_state(cpu, st, prev_state);
777 		__cpuhp_kick_ap(st);
778 	}
779 
780 	return ret;
781 }
782 
bringup_wait_for_ap_online(unsigned int cpu)783 static int bringup_wait_for_ap_online(unsigned int cpu)
784 {
785 	struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
786 
787 	/* Wait for the CPU to reach CPUHP_AP_ONLINE_IDLE */
788 	wait_for_ap_thread(st, true);
789 	if (WARN_ON_ONCE((!cpu_online(cpu))))
790 		return -ECANCELED;
791 
792 	/* Unpark the hotplug thread of the target cpu */
793 	kthread_unpark(st->thread);
794 
795 	/*
796 	 * SMT soft disabling on X86 requires to bring the CPU out of the
797 	 * BIOS 'wait for SIPI' state in order to set the CR4.MCE bit.  The
798 	 * CPU marked itself as booted_once in notify_cpu_starting() so the
799 	 * cpu_bootable() check will now return false if this is not the
800 	 * primary sibling.
801 	 */
802 	if (!cpu_bootable(cpu))
803 		return -ECANCELED;
804 	return 0;
805 }
806 
807 #ifdef CONFIG_HOTPLUG_SPLIT_STARTUP
cpuhp_kick_ap_alive(unsigned int cpu)808 static int cpuhp_kick_ap_alive(unsigned int cpu)
809 {
810 	if (!cpuhp_can_boot_ap(cpu))
811 		return -EAGAIN;
812 
813 	return arch_cpuhp_kick_ap_alive(cpu, idle_thread_get(cpu));
814 }
815 
cpuhp_bringup_ap(unsigned int cpu)816 static int cpuhp_bringup_ap(unsigned int cpu)
817 {
818 	struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
819 	int ret;
820 
821 	/*
822 	 * Some architectures have to walk the irq descriptors to
823 	 * setup the vector space for the cpu which comes online.
824 	 * Prevent irq alloc/free across the bringup.
825 	 */
826 	irq_lock_sparse();
827 
828 	ret = cpuhp_bp_sync_alive(cpu);
829 	if (ret)
830 		goto out_unlock;
831 
832 	ret = bringup_wait_for_ap_online(cpu);
833 	if (ret)
834 		goto out_unlock;
835 
836 	irq_unlock_sparse();
837 
838 	if (st->target <= CPUHP_AP_ONLINE_IDLE)
839 		return 0;
840 
841 	return cpuhp_kick_ap(cpu, st, st->target);
842 
843 out_unlock:
844 	irq_unlock_sparse();
845 	return ret;
846 }
847 #else
bringup_cpu(unsigned int cpu)848 static int bringup_cpu(unsigned int cpu)
849 {
850 	struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
851 	struct task_struct *idle = idle_thread_get(cpu);
852 	int ret;
853 
854 	if (!cpuhp_can_boot_ap(cpu))
855 		return -EAGAIN;
856 
857 	/*
858 	 * Some architectures have to walk the irq descriptors to
859 	 * setup the vector space for the cpu which comes online.
860 	 *
861 	 * Prevent irq alloc/free across the bringup by acquiring the
862 	 * sparse irq lock. Hold it until the upcoming CPU completes the
863 	 * startup in cpuhp_online_idle() which allows to avoid
864 	 * intermediate synchronization points in the architecture code.
865 	 */
866 	irq_lock_sparse();
867 
868 	ret = __cpu_up(cpu, idle);
869 	if (ret)
870 		goto out_unlock;
871 
872 	ret = cpuhp_bp_sync_alive(cpu);
873 	if (ret)
874 		goto out_unlock;
875 
876 	ret = bringup_wait_for_ap_online(cpu);
877 	if (ret)
878 		goto out_unlock;
879 
880 	irq_unlock_sparse();
881 
882 	if (st->target <= CPUHP_AP_ONLINE_IDLE)
883 		return 0;
884 
885 	return cpuhp_kick_ap(cpu, st, st->target);
886 
887 out_unlock:
888 	irq_unlock_sparse();
889 	return ret;
890 }
891 #endif
892 
finish_cpu(unsigned int cpu)893 static int finish_cpu(unsigned int cpu)
894 {
895 	struct task_struct *idle = idle_thread_get(cpu);
896 	struct mm_struct *mm = idle->active_mm;
897 
898 	/*
899 	 * idle_task_exit() will have switched to &init_mm, now
900 	 * clean up any remaining active_mm state.
901 	 */
902 	if (mm != &init_mm)
903 		idle->active_mm = &init_mm;
904 	mmdrop_lazy_tlb(mm);
905 	return 0;
906 }
907 
908 /*
909  * Hotplug state machine related functions
910  */
911 
912 /*
913  * Get the next state to run. Empty ones will be skipped. Returns true if a
914  * state must be run.
915  *
916  * st->state will be modified ahead of time, to match state_to_run, as if it
917  * has already ran.
918  */
cpuhp_next_state(bool bringup,enum cpuhp_state * state_to_run,struct cpuhp_cpu_state * st,enum cpuhp_state target)919 static bool cpuhp_next_state(bool bringup,
920 			     enum cpuhp_state *state_to_run,
921 			     struct cpuhp_cpu_state *st,
922 			     enum cpuhp_state target)
923 {
924 	do {
925 		if (bringup) {
926 			if (st->state >= target)
927 				return false;
928 
929 			*state_to_run = ++st->state;
930 		} else {
931 			if (st->state <= target)
932 				return false;
933 
934 			*state_to_run = st->state--;
935 		}
936 
937 		if (!cpuhp_step_empty(bringup, cpuhp_get_step(*state_to_run)))
938 			break;
939 	} while (true);
940 
941 	return true;
942 }
943 
__cpuhp_invoke_callback_range(bool bringup,unsigned int cpu,struct cpuhp_cpu_state * st,enum cpuhp_state target,bool nofail)944 static int __cpuhp_invoke_callback_range(bool bringup,
945 					 unsigned int cpu,
946 					 struct cpuhp_cpu_state *st,
947 					 enum cpuhp_state target,
948 					 bool nofail)
949 {
950 	enum cpuhp_state state;
951 	int ret = 0;
952 
953 	while (cpuhp_next_state(bringup, &state, st, target)) {
954 		int err;
955 
956 		err = cpuhp_invoke_callback(cpu, state, bringup, NULL, NULL);
957 		if (!err)
958 			continue;
959 
960 		if (nofail) {
961 			pr_warn("CPU %u %s state %s (%d) failed (%d)\n",
962 				cpu, bringup ? "UP" : "DOWN",
963 				cpuhp_get_step(st->state)->name,
964 				st->state, err);
965 			ret = -1;
966 		} else {
967 			ret = err;
968 			break;
969 		}
970 	}
971 
972 	return ret;
973 }
974 
cpuhp_invoke_callback_range(bool bringup,unsigned int cpu,struct cpuhp_cpu_state * st,enum cpuhp_state target)975 static inline int cpuhp_invoke_callback_range(bool bringup,
976 					      unsigned int cpu,
977 					      struct cpuhp_cpu_state *st,
978 					      enum cpuhp_state target)
979 {
980 	return __cpuhp_invoke_callback_range(bringup, cpu, st, target, false);
981 }
982 
cpuhp_invoke_callback_range_nofail(bool bringup,unsigned int cpu,struct cpuhp_cpu_state * st,enum cpuhp_state target)983 static inline void cpuhp_invoke_callback_range_nofail(bool bringup,
984 						      unsigned int cpu,
985 						      struct cpuhp_cpu_state *st,
986 						      enum cpuhp_state target)
987 {
988 	__cpuhp_invoke_callback_range(bringup, cpu, st, target, true);
989 }
990 
can_rollback_cpu(struct cpuhp_cpu_state * st)991 static inline bool can_rollback_cpu(struct cpuhp_cpu_state *st)
992 {
993 	if (IS_ENABLED(CONFIG_HOTPLUG_CPU))
994 		return true;
995 	/*
996 	 * When CPU hotplug is disabled, then taking the CPU down is not
997 	 * possible because takedown_cpu() and the architecture and
998 	 * subsystem specific mechanisms are not available. So the CPU
999 	 * which would be completely unplugged again needs to stay around
1000 	 * in the current state.
1001 	 */
1002 	return st->state <= CPUHP_BRINGUP_CPU;
1003 }
1004 
cpuhp_up_callbacks(unsigned int cpu,struct cpuhp_cpu_state * st,enum cpuhp_state target)1005 static int cpuhp_up_callbacks(unsigned int cpu, struct cpuhp_cpu_state *st,
1006 			      enum cpuhp_state target)
1007 {
1008 	enum cpuhp_state prev_state = st->state;
1009 	int ret = 0;
1010 
1011 	ret = cpuhp_invoke_callback_range(true, cpu, st, target);
1012 	if (ret) {
1013 		pr_debug("CPU UP failed (%d) CPU %u state %s (%d)\n",
1014 			 ret, cpu, cpuhp_get_step(st->state)->name,
1015 			 st->state);
1016 
1017 		cpuhp_reset_state(cpu, st, prev_state);
1018 		if (can_rollback_cpu(st))
1019 			WARN_ON(cpuhp_invoke_callback_range(false, cpu, st,
1020 							    prev_state));
1021 	}
1022 	return ret;
1023 }
1024 
1025 /*
1026  * The cpu hotplug threads manage the bringup and teardown of the cpus
1027  */
cpuhp_should_run(unsigned int cpu)1028 static int cpuhp_should_run(unsigned int cpu)
1029 {
1030 	struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
1031 
1032 	return st->should_run;
1033 }
1034 
1035 /*
1036  * Execute teardown/startup callbacks on the plugged cpu. Also used to invoke
1037  * callbacks when a state gets [un]installed at runtime.
1038  *
1039  * Each invocation of this function by the smpboot thread does a single AP
1040  * state callback.
1041  *
1042  * It has 3 modes of operation:
1043  *  - single: runs st->cb_state
1044  *  - up:     runs ++st->state, while st->state < st->target
1045  *  - down:   runs st->state--, while st->state > st->target
1046  *
1047  * When complete or on error, should_run is cleared and the completion is fired.
1048  */
cpuhp_thread_fun(unsigned int cpu)1049 static void cpuhp_thread_fun(unsigned int cpu)
1050 {
1051 	struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
1052 	bool bringup = st->bringup;
1053 	enum cpuhp_state state;
1054 
1055 	if (WARN_ON_ONCE(!st->should_run))
1056 		return;
1057 
1058 	/*
1059 	 * ACQUIRE for the cpuhp_should_run() load of ->should_run. Ensures
1060 	 * that if we see ->should_run we also see the rest of the state.
1061 	 */
1062 	smp_mb();
1063 
1064 	/*
1065 	 * The BP holds the hotplug lock, but we're now running on the AP,
1066 	 * ensure that anybody asserting the lock is held, will actually find
1067 	 * it so.
1068 	 */
1069 	lockdep_acquire_cpus_lock();
1070 	cpuhp_lock_acquire(bringup);
1071 
1072 	if (st->single) {
1073 		state = st->cb_state;
1074 		st->should_run = false;
1075 	} else {
1076 		st->should_run = cpuhp_next_state(bringup, &state, st, st->target);
1077 		if (!st->should_run)
1078 			goto end;
1079 	}
1080 
1081 	WARN_ON_ONCE(!cpuhp_is_ap_state(state));
1082 
1083 	if (cpuhp_is_atomic_state(state)) {
1084 		local_irq_disable();
1085 		st->result = cpuhp_invoke_callback(cpu, state, bringup, st->node, &st->last);
1086 		local_irq_enable();
1087 
1088 		/*
1089 		 * STARTING/DYING must not fail!
1090 		 */
1091 		WARN_ON_ONCE(st->result);
1092 	} else {
1093 		st->result = cpuhp_invoke_callback(cpu, state, bringup, st->node, &st->last);
1094 	}
1095 
1096 	if (st->result) {
1097 		/*
1098 		 * If we fail on a rollback, we're up a creek without no
1099 		 * paddle, no way forward, no way back. We loose, thanks for
1100 		 * playing.
1101 		 */
1102 		WARN_ON_ONCE(st->rollback);
1103 		st->should_run = false;
1104 	}
1105 
1106 end:
1107 	cpuhp_lock_release(bringup);
1108 	lockdep_release_cpus_lock();
1109 
1110 	if (!st->should_run)
1111 		complete_ap_thread(st, bringup);
1112 }
1113 
1114 /* Invoke a single callback on a remote cpu */
1115 static int
cpuhp_invoke_ap_callback(int cpu,enum cpuhp_state state,bool bringup,struct hlist_node * node)1116 cpuhp_invoke_ap_callback(int cpu, enum cpuhp_state state, bool bringup,
1117 			 struct hlist_node *node)
1118 {
1119 	struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1120 	int ret;
1121 
1122 	if (!cpu_online(cpu))
1123 		return 0;
1124 
1125 	cpuhp_lock_acquire(false);
1126 	cpuhp_lock_release(false);
1127 
1128 	cpuhp_lock_acquire(true);
1129 	cpuhp_lock_release(true);
1130 
1131 	/*
1132 	 * If we are up and running, use the hotplug thread. For early calls
1133 	 * we invoke the thread function directly.
1134 	 */
1135 	if (!st->thread)
1136 		return cpuhp_invoke_callback(cpu, state, bringup, node, NULL);
1137 
1138 	st->rollback = false;
1139 	st->last = NULL;
1140 
1141 	st->node = node;
1142 	st->bringup = bringup;
1143 	st->cb_state = state;
1144 	st->single = true;
1145 
1146 	__cpuhp_kick_ap(st);
1147 
1148 	/*
1149 	 * If we failed and did a partial, do a rollback.
1150 	 */
1151 	if ((ret = st->result) && st->last) {
1152 		st->rollback = true;
1153 		st->bringup = !bringup;
1154 
1155 		__cpuhp_kick_ap(st);
1156 	}
1157 
1158 	/*
1159 	 * Clean up the leftovers so the next hotplug operation wont use stale
1160 	 * data.
1161 	 */
1162 	st->node = st->last = NULL;
1163 	return ret;
1164 }
1165 
cpuhp_kick_ap_work(unsigned int cpu)1166 static int cpuhp_kick_ap_work(unsigned int cpu)
1167 {
1168 	struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1169 	enum cpuhp_state prev_state = st->state;
1170 	int ret;
1171 
1172 	cpuhp_lock_acquire(false);
1173 	cpuhp_lock_release(false);
1174 
1175 	cpuhp_lock_acquire(true);
1176 	cpuhp_lock_release(true);
1177 
1178 	trace_cpuhp_enter(cpu, st->target, prev_state, cpuhp_kick_ap_work);
1179 	ret = cpuhp_kick_ap(cpu, st, st->target);
1180 	trace_cpuhp_exit(cpu, st->state, prev_state, ret);
1181 
1182 	return ret;
1183 }
1184 
1185 static struct smp_hotplug_thread cpuhp_threads = {
1186 	.store			= &cpuhp_state.thread,
1187 	.thread_should_run	= cpuhp_should_run,
1188 	.thread_fn		= cpuhp_thread_fun,
1189 	.thread_comm		= "cpuhp/%u",
1190 	.selfparking		= true,
1191 };
1192 
cpuhp_init_state(void)1193 static __init void cpuhp_init_state(void)
1194 {
1195 	struct cpuhp_cpu_state *st;
1196 	int cpu;
1197 
1198 	for_each_possible_cpu(cpu) {
1199 		st = per_cpu_ptr(&cpuhp_state, cpu);
1200 		init_completion(&st->done_up);
1201 		init_completion(&st->done_down);
1202 	}
1203 }
1204 
cpuhp_threads_init(void)1205 void __init cpuhp_threads_init(void)
1206 {
1207 	cpuhp_init_state();
1208 	BUG_ON(smpboot_register_percpu_thread(&cpuhp_threads));
1209 	kthread_unpark(this_cpu_read(cpuhp_state.thread));
1210 }
1211 
1212 /*
1213  *
1214  * Serialize hotplug trainwrecks outside of the cpu_hotplug_lock
1215  * protected region.
1216  *
1217  * The operation is still serialized against concurrent CPU hotplug via
1218  * cpu_add_remove_lock, i.e. CPU map protection.  But it is _not_
1219  * serialized against other hotplug related activity like adding or
1220  * removing of state callbacks and state instances, which invoke either the
1221  * startup or the teardown callback of the affected state.
1222  *
1223  * This is required for subsystems which are unfixable vs. CPU hotplug and
1224  * evade lock inversion problems by scheduling work which has to be
1225  * completed _before_ cpu_up()/_cpu_down() returns.
1226  *
1227  * Don't even think about adding anything to this for any new code or even
1228  * drivers. It's only purpose is to keep existing lock order trainwrecks
1229  * working.
1230  *
1231  * For cpu_down() there might be valid reasons to finish cleanups which are
1232  * not required to be done under cpu_hotplug_lock, but that's a different
1233  * story and would be not invoked via this.
1234  */
cpu_up_down_serialize_trainwrecks(bool tasks_frozen)1235 static void cpu_up_down_serialize_trainwrecks(bool tasks_frozen)
1236 {
1237 	/*
1238 	 * cpusets delegate hotplug operations to a worker to "solve" the
1239 	 * lock order problems. Wait for the worker, but only if tasks are
1240 	 * _not_ frozen (suspend, hibernate) as that would wait forever.
1241 	 *
1242 	 * The wait is required because otherwise the hotplug operation
1243 	 * returns with inconsistent state, which could even be observed in
1244 	 * user space when a new CPU is brought up. The CPU plug uevent
1245 	 * would be delivered and user space reacting on it would fail to
1246 	 * move tasks to the newly plugged CPU up to the point where the
1247 	 * work has finished because up to that point the newly plugged CPU
1248 	 * is not assignable in cpusets/cgroups. On unplug that's not
1249 	 * necessarily a visible issue, but it is still inconsistent state,
1250 	 * which is the real problem which needs to be "fixed". This can't
1251 	 * prevent the transient state between scheduling the work and
1252 	 * returning from waiting for it.
1253 	 */
1254 	if (!tasks_frozen)
1255 		cpuset_wait_for_hotplug();
1256 }
1257 
1258 #ifdef CONFIG_HOTPLUG_CPU
1259 #ifndef arch_clear_mm_cpumask_cpu
1260 #define arch_clear_mm_cpumask_cpu(cpu, mm) cpumask_clear_cpu(cpu, mm_cpumask(mm))
1261 #endif
1262 
1263 /**
1264  * clear_tasks_mm_cpumask - Safely clear tasks' mm_cpumask for a CPU
1265  * @cpu: a CPU id
1266  *
1267  * This function walks all processes, finds a valid mm struct for each one and
1268  * then clears a corresponding bit in mm's cpumask.  While this all sounds
1269  * trivial, there are various non-obvious corner cases, which this function
1270  * tries to solve in a safe manner.
1271  *
1272  * Also note that the function uses a somewhat relaxed locking scheme, so it may
1273  * be called only for an already offlined CPU.
1274  */
clear_tasks_mm_cpumask(int cpu)1275 void clear_tasks_mm_cpumask(int cpu)
1276 {
1277 	struct task_struct *p;
1278 
1279 	/*
1280 	 * This function is called after the cpu is taken down and marked
1281 	 * offline, so its not like new tasks will ever get this cpu set in
1282 	 * their mm mask. -- Peter Zijlstra
1283 	 * Thus, we may use rcu_read_lock() here, instead of grabbing
1284 	 * full-fledged tasklist_lock.
1285 	 */
1286 	WARN_ON(cpu_online(cpu));
1287 	rcu_read_lock();
1288 	for_each_process(p) {
1289 		struct task_struct *t;
1290 
1291 		/*
1292 		 * Main thread might exit, but other threads may still have
1293 		 * a valid mm. Find one.
1294 		 */
1295 		t = find_lock_task_mm(p);
1296 		if (!t)
1297 			continue;
1298 		arch_clear_mm_cpumask_cpu(cpu, t->mm);
1299 		task_unlock(t);
1300 	}
1301 	rcu_read_unlock();
1302 }
1303 
1304 /* Take this CPU down. */
take_cpu_down(void * _param)1305 static int take_cpu_down(void *_param)
1306 {
1307 	struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
1308 	enum cpuhp_state target = max((int)st->target, CPUHP_AP_OFFLINE);
1309 	int err, cpu = smp_processor_id();
1310 
1311 	/* Ensure this CPU doesn't handle any more interrupts. */
1312 	err = __cpu_disable();
1313 	if (err < 0)
1314 		return err;
1315 
1316 	/*
1317 	 * Must be called from CPUHP_TEARDOWN_CPU, which means, as we are going
1318 	 * down, that the current state is CPUHP_TEARDOWN_CPU - 1.
1319 	 */
1320 	WARN_ON(st->state != (CPUHP_TEARDOWN_CPU - 1));
1321 
1322 	/*
1323 	 * Invoke the former CPU_DYING callbacks. DYING must not fail!
1324 	 */
1325 	cpuhp_invoke_callback_range_nofail(false, cpu, st, target);
1326 
1327 	/* Give up timekeeping duties */
1328 	tick_handover_do_timer();
1329 	/* Remove CPU from timer broadcasting */
1330 	tick_offline_cpu(cpu);
1331 	/* Park the stopper thread */
1332 	stop_machine_park(cpu);
1333 	return 0;
1334 }
1335 
takedown_cpu(unsigned int cpu)1336 static int takedown_cpu(unsigned int cpu)
1337 {
1338 	struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1339 	int err;
1340 
1341 	/* Park the smpboot threads */
1342 	kthread_park(st->thread);
1343 
1344 	/*
1345 	 * Prevent irq alloc/free while the dying cpu reorganizes the
1346 	 * interrupt affinities.
1347 	 */
1348 	irq_lock_sparse();
1349 
1350 	/*
1351 	 * So now all preempt/rcu users must observe !cpu_active().
1352 	 */
1353 	err = stop_machine_cpuslocked(take_cpu_down, NULL, cpumask_of(cpu));
1354 	if (err) {
1355 		/* CPU refused to die */
1356 		irq_unlock_sparse();
1357 		/* Unpark the hotplug thread so we can rollback there */
1358 		kthread_unpark(st->thread);
1359 		return err;
1360 	}
1361 	BUG_ON(cpu_online(cpu));
1362 
1363 	/*
1364 	 * The teardown callback for CPUHP_AP_SCHED_STARTING will have removed
1365 	 * all runnable tasks from the CPU, there's only the idle task left now
1366 	 * that the migration thread is done doing the stop_machine thing.
1367 	 *
1368 	 * Wait for the stop thread to go away.
1369 	 */
1370 	wait_for_ap_thread(st, false);
1371 	BUG_ON(st->state != CPUHP_AP_IDLE_DEAD);
1372 
1373 	/* Interrupts are moved away from the dying cpu, reenable alloc/free */
1374 	irq_unlock_sparse();
1375 
1376 	hotplug_cpu__broadcast_tick_pull(cpu);
1377 	/* This actually kills the CPU. */
1378 	__cpu_die(cpu);
1379 
1380 	cpuhp_bp_sync_dead(cpu);
1381 
1382 	tick_cleanup_dead_cpu(cpu);
1383 	rcutree_migrate_callbacks(cpu);
1384 	return 0;
1385 }
1386 
cpuhp_complete_idle_dead(void * arg)1387 static void cpuhp_complete_idle_dead(void *arg)
1388 {
1389 	struct cpuhp_cpu_state *st = arg;
1390 
1391 	complete_ap_thread(st, false);
1392 }
1393 
cpuhp_report_idle_dead(void)1394 void cpuhp_report_idle_dead(void)
1395 {
1396 	struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
1397 
1398 	BUG_ON(st->state != CPUHP_AP_OFFLINE);
1399 	rcu_report_dead(smp_processor_id());
1400 	st->state = CPUHP_AP_IDLE_DEAD;
1401 	/*
1402 	 * We cannot call complete after rcu_report_dead() so we delegate it
1403 	 * to an online cpu.
1404 	 */
1405 	smp_call_function_single(cpumask_first(cpu_online_mask),
1406 				 cpuhp_complete_idle_dead, st, 0);
1407 }
1408 
cpuhp_down_callbacks(unsigned int cpu,struct cpuhp_cpu_state * st,enum cpuhp_state target)1409 static int cpuhp_down_callbacks(unsigned int cpu, struct cpuhp_cpu_state *st,
1410 				enum cpuhp_state target)
1411 {
1412 	enum cpuhp_state prev_state = st->state;
1413 	int ret = 0;
1414 
1415 	ret = cpuhp_invoke_callback_range(false, cpu, st, target);
1416 	if (ret) {
1417 		pr_debug("CPU DOWN failed (%d) CPU %u state %s (%d)\n",
1418 			 ret, cpu, cpuhp_get_step(st->state)->name,
1419 			 st->state);
1420 
1421 		cpuhp_reset_state(cpu, st, prev_state);
1422 
1423 		if (st->state < prev_state)
1424 			WARN_ON(cpuhp_invoke_callback_range(true, cpu, st,
1425 							    prev_state));
1426 	}
1427 
1428 	return ret;
1429 }
1430 
1431 /* Requires cpu_add_remove_lock to be held */
_cpu_down(unsigned int cpu,int tasks_frozen,enum cpuhp_state target)1432 static int __ref _cpu_down(unsigned int cpu, int tasks_frozen,
1433 			   enum cpuhp_state target)
1434 {
1435 	struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1436 	int prev_state, ret = 0;
1437 
1438 	if (num_online_cpus() == 1)
1439 		return -EBUSY;
1440 
1441 	if (!cpu_present(cpu))
1442 		return -EINVAL;
1443 
1444 	cpus_write_lock();
1445 
1446 	cpuhp_tasks_frozen = tasks_frozen;
1447 
1448 	prev_state = cpuhp_set_state(cpu, st, target);
1449 	/*
1450 	 * If the current CPU state is in the range of the AP hotplug thread,
1451 	 * then we need to kick the thread.
1452 	 */
1453 	if (st->state > CPUHP_TEARDOWN_CPU) {
1454 		st->target = max((int)target, CPUHP_TEARDOWN_CPU);
1455 		ret = cpuhp_kick_ap_work(cpu);
1456 		/*
1457 		 * The AP side has done the error rollback already. Just
1458 		 * return the error code..
1459 		 */
1460 		if (ret)
1461 			goto out;
1462 
1463 		/*
1464 		 * We might have stopped still in the range of the AP hotplug
1465 		 * thread. Nothing to do anymore.
1466 		 */
1467 		if (st->state > CPUHP_TEARDOWN_CPU)
1468 			goto out;
1469 
1470 		st->target = target;
1471 	}
1472 	/*
1473 	 * The AP brought itself down to CPUHP_TEARDOWN_CPU. So we need
1474 	 * to do the further cleanups.
1475 	 */
1476 	ret = cpuhp_down_callbacks(cpu, st, target);
1477 	if (ret && st->state < prev_state) {
1478 		if (st->state == CPUHP_TEARDOWN_CPU) {
1479 			cpuhp_reset_state(cpu, st, prev_state);
1480 			__cpuhp_kick_ap(st);
1481 		} else {
1482 			WARN(1, "DEAD callback error for CPU%d", cpu);
1483 		}
1484 	}
1485 
1486 out:
1487 	cpus_write_unlock();
1488 	/*
1489 	 * Do post unplug cleanup. This is still protected against
1490 	 * concurrent CPU hotplug via cpu_add_remove_lock.
1491 	 */
1492 	lockup_detector_cleanup();
1493 	arch_smt_update();
1494 	cpu_up_down_serialize_trainwrecks(tasks_frozen);
1495 	return ret;
1496 }
1497 
1498 struct cpu_down_work {
1499 	unsigned int		cpu;
1500 	enum cpuhp_state	target;
1501 };
1502 
__cpu_down_maps_locked(void * arg)1503 static long __cpu_down_maps_locked(void *arg)
1504 {
1505 	struct cpu_down_work *work = arg;
1506 
1507 	return _cpu_down(work->cpu, 0, work->target);
1508 }
1509 
cpu_down_maps_locked(unsigned int cpu,enum cpuhp_state target)1510 static int cpu_down_maps_locked(unsigned int cpu, enum cpuhp_state target)
1511 {
1512 	struct cpu_down_work work = { .cpu = cpu, .target = target, };
1513 
1514 	/*
1515 	 * If the platform does not support hotplug, report it explicitly to
1516 	 * differentiate it from a transient offlining failure.
1517 	 */
1518 	if (cc_platform_has(CC_ATTR_HOTPLUG_DISABLED))
1519 		return -EOPNOTSUPP;
1520 	if (cpu_hotplug_disabled)
1521 		return -EBUSY;
1522 
1523 	/*
1524 	 * Ensure that the control task does not run on the to be offlined
1525 	 * CPU to prevent a deadlock against cfs_b->period_timer.
1526 	 * Also keep at least one housekeeping cpu onlined to avoid generating
1527 	 * an empty sched_domain span.
1528 	 */
1529 	for_each_cpu_and(cpu, cpu_online_mask, housekeeping_cpumask(HK_TYPE_DOMAIN)) {
1530 		if (cpu != work.cpu)
1531 			return work_on_cpu(cpu, __cpu_down_maps_locked, &work);
1532 	}
1533 	return -EBUSY;
1534 }
1535 
cpu_down(unsigned int cpu,enum cpuhp_state target)1536 static int cpu_down(unsigned int cpu, enum cpuhp_state target)
1537 {
1538 	int err;
1539 
1540 	cpu_maps_update_begin();
1541 	err = cpu_down_maps_locked(cpu, target);
1542 	cpu_maps_update_done();
1543 	return err;
1544 }
1545 
1546 /**
1547  * cpu_device_down - Bring down a cpu device
1548  * @dev: Pointer to the cpu device to offline
1549  *
1550  * This function is meant to be used by device core cpu subsystem only.
1551  *
1552  * Other subsystems should use remove_cpu() instead.
1553  *
1554  * Return: %0 on success or a negative errno code
1555  */
cpu_device_down(struct device * dev)1556 int cpu_device_down(struct device *dev)
1557 {
1558 	return cpu_down(dev->id, CPUHP_OFFLINE);
1559 }
1560 
remove_cpu(unsigned int cpu)1561 int remove_cpu(unsigned int cpu)
1562 {
1563 	int ret;
1564 
1565 	lock_device_hotplug();
1566 	ret = device_offline(get_cpu_device(cpu));
1567 	unlock_device_hotplug();
1568 
1569 	return ret;
1570 }
1571 EXPORT_SYMBOL_GPL(remove_cpu);
1572 
smp_shutdown_nonboot_cpus(unsigned int primary_cpu)1573 void smp_shutdown_nonboot_cpus(unsigned int primary_cpu)
1574 {
1575 	unsigned int cpu;
1576 	int error;
1577 
1578 	cpu_maps_update_begin();
1579 
1580 	/*
1581 	 * Make certain the cpu I'm about to reboot on is online.
1582 	 *
1583 	 * This is inline to what migrate_to_reboot_cpu() already do.
1584 	 */
1585 	if (!cpu_online(primary_cpu))
1586 		primary_cpu = cpumask_first(cpu_online_mask);
1587 
1588 	for_each_online_cpu(cpu) {
1589 		if (cpu == primary_cpu)
1590 			continue;
1591 
1592 		error = cpu_down_maps_locked(cpu, CPUHP_OFFLINE);
1593 		if (error) {
1594 			pr_err("Failed to offline CPU%d - error=%d",
1595 				cpu, error);
1596 			break;
1597 		}
1598 	}
1599 
1600 	/*
1601 	 * Ensure all but the reboot CPU are offline.
1602 	 */
1603 	BUG_ON(num_online_cpus() > 1);
1604 
1605 	/*
1606 	 * Make sure the CPUs won't be enabled by someone else after this
1607 	 * point. Kexec will reboot to a new kernel shortly resetting
1608 	 * everything along the way.
1609 	 */
1610 	cpu_hotplug_disabled++;
1611 
1612 	cpu_maps_update_done();
1613 }
1614 
1615 #else
1616 #define takedown_cpu		NULL
1617 #endif /*CONFIG_HOTPLUG_CPU*/
1618 
1619 /**
1620  * notify_cpu_starting(cpu) - Invoke the callbacks on the starting CPU
1621  * @cpu: cpu that just started
1622  *
1623  * It must be called by the arch code on the new cpu, before the new cpu
1624  * enables interrupts and before the "boot" cpu returns from __cpu_up().
1625  */
notify_cpu_starting(unsigned int cpu)1626 void notify_cpu_starting(unsigned int cpu)
1627 {
1628 	struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1629 	enum cpuhp_state target = min((int)st->target, CPUHP_AP_ONLINE);
1630 
1631 	rcu_cpu_starting(cpu);	/* Enables RCU usage on this CPU. */
1632 	cpumask_set_cpu(cpu, &cpus_booted_once_mask);
1633 
1634 	/*
1635 	 * STARTING must not fail!
1636 	 */
1637 	cpuhp_invoke_callback_range_nofail(true, cpu, st, target);
1638 }
1639 
1640 /*
1641  * Called from the idle task. Wake up the controlling task which brings the
1642  * hotplug thread of the upcoming CPU up and then delegates the rest of the
1643  * online bringup to the hotplug thread.
1644  */
cpuhp_online_idle(enum cpuhp_state state)1645 void cpuhp_online_idle(enum cpuhp_state state)
1646 {
1647 	struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
1648 
1649 	/* Happens for the boot cpu */
1650 	if (state != CPUHP_AP_ONLINE_IDLE)
1651 		return;
1652 
1653 	cpuhp_ap_update_sync_state(SYNC_STATE_ONLINE);
1654 
1655 	/*
1656 	 * Unpark the stopper thread before we start the idle loop (and start
1657 	 * scheduling); this ensures the stopper task is always available.
1658 	 */
1659 	stop_machine_unpark(smp_processor_id());
1660 
1661 	st->state = CPUHP_AP_ONLINE_IDLE;
1662 	complete_ap_thread(st, true);
1663 }
1664 
1665 /* Requires cpu_add_remove_lock to be held */
_cpu_up(unsigned int cpu,int tasks_frozen,enum cpuhp_state target)1666 static int _cpu_up(unsigned int cpu, int tasks_frozen, enum cpuhp_state target)
1667 {
1668 	struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1669 	struct task_struct *idle;
1670 	int ret = 0;
1671 
1672 	cpus_write_lock();
1673 
1674 	if (!cpu_present(cpu)) {
1675 		ret = -EINVAL;
1676 		goto out;
1677 	}
1678 
1679 	/*
1680 	 * The caller of cpu_up() might have raced with another
1681 	 * caller. Nothing to do.
1682 	 */
1683 	if (st->state >= target)
1684 		goto out;
1685 
1686 	if (st->state == CPUHP_OFFLINE) {
1687 		/* Let it fail before we try to bring the cpu up */
1688 		idle = idle_thread_get(cpu);
1689 		if (IS_ERR(idle)) {
1690 			ret = PTR_ERR(idle);
1691 			goto out;
1692 		}
1693 
1694 		/*
1695 		 * Reset stale stack state from the last time this CPU was online.
1696 		 */
1697 		scs_task_reset(idle);
1698 		kasan_unpoison_task_stack(idle);
1699 	}
1700 
1701 	cpuhp_tasks_frozen = tasks_frozen;
1702 
1703 	cpuhp_set_state(cpu, st, target);
1704 	/*
1705 	 * If the current CPU state is in the range of the AP hotplug thread,
1706 	 * then we need to kick the thread once more.
1707 	 */
1708 	if (st->state > CPUHP_BRINGUP_CPU) {
1709 		ret = cpuhp_kick_ap_work(cpu);
1710 		/*
1711 		 * The AP side has done the error rollback already. Just
1712 		 * return the error code..
1713 		 */
1714 		if (ret)
1715 			goto out;
1716 	}
1717 
1718 	/*
1719 	 * Try to reach the target state. We max out on the BP at
1720 	 * CPUHP_BRINGUP_CPU. After that the AP hotplug thread is
1721 	 * responsible for bringing it up to the target state.
1722 	 */
1723 	target = min((int)target, CPUHP_BRINGUP_CPU);
1724 	ret = cpuhp_up_callbacks(cpu, st, target);
1725 out:
1726 	cpus_write_unlock();
1727 	arch_smt_update();
1728 	cpu_up_down_serialize_trainwrecks(tasks_frozen);
1729 	return ret;
1730 }
1731 
cpu_up(unsigned int cpu,enum cpuhp_state target)1732 static int cpu_up(unsigned int cpu, enum cpuhp_state target)
1733 {
1734 	int err = 0;
1735 
1736 	if (!cpu_possible(cpu)) {
1737 		pr_err("can't online cpu %d because it is not configured as may-hotadd at boot time\n",
1738 		       cpu);
1739 #if defined(CONFIG_IA64)
1740 		pr_err("please check additional_cpus= boot parameter\n");
1741 #endif
1742 		return -EINVAL;
1743 	}
1744 
1745 	err = try_online_node(cpu_to_node(cpu));
1746 	if (err)
1747 		return err;
1748 
1749 	cpu_maps_update_begin();
1750 
1751 	if (cpu_hotplug_disabled) {
1752 		err = -EBUSY;
1753 		goto out;
1754 	}
1755 	if (!cpu_bootable(cpu)) {
1756 		err = -EPERM;
1757 		goto out;
1758 	}
1759 
1760 	err = _cpu_up(cpu, 0, target);
1761 out:
1762 	cpu_maps_update_done();
1763 	return err;
1764 }
1765 
1766 /**
1767  * cpu_device_up - Bring up a cpu device
1768  * @dev: Pointer to the cpu device to online
1769  *
1770  * This function is meant to be used by device core cpu subsystem only.
1771  *
1772  * Other subsystems should use add_cpu() instead.
1773  *
1774  * Return: %0 on success or a negative errno code
1775  */
cpu_device_up(struct device * dev)1776 int cpu_device_up(struct device *dev)
1777 {
1778 	return cpu_up(dev->id, CPUHP_ONLINE);
1779 }
1780 
add_cpu(unsigned int cpu)1781 int add_cpu(unsigned int cpu)
1782 {
1783 	int ret;
1784 
1785 	lock_device_hotplug();
1786 	ret = device_online(get_cpu_device(cpu));
1787 	unlock_device_hotplug();
1788 
1789 	return ret;
1790 }
1791 EXPORT_SYMBOL_GPL(add_cpu);
1792 
1793 /**
1794  * bringup_hibernate_cpu - Bring up the CPU that we hibernated on
1795  * @sleep_cpu: The cpu we hibernated on and should be brought up.
1796  *
1797  * On some architectures like arm64, we can hibernate on any CPU, but on
1798  * wake up the CPU we hibernated on might be offline as a side effect of
1799  * using maxcpus= for example.
1800  *
1801  * Return: %0 on success or a negative errno code
1802  */
bringup_hibernate_cpu(unsigned int sleep_cpu)1803 int bringup_hibernate_cpu(unsigned int sleep_cpu)
1804 {
1805 	int ret;
1806 
1807 	if (!cpu_online(sleep_cpu)) {
1808 		pr_info("Hibernated on a CPU that is offline! Bringing CPU up.\n");
1809 		ret = cpu_up(sleep_cpu, CPUHP_ONLINE);
1810 		if (ret) {
1811 			pr_err("Failed to bring hibernate-CPU up!\n");
1812 			return ret;
1813 		}
1814 	}
1815 	return 0;
1816 }
1817 
cpuhp_bringup_mask(const struct cpumask * mask,unsigned int ncpus,enum cpuhp_state target)1818 static void __init cpuhp_bringup_mask(const struct cpumask *mask, unsigned int ncpus,
1819 				      enum cpuhp_state target)
1820 {
1821 	unsigned int cpu;
1822 
1823 	for_each_cpu(cpu, mask) {
1824 		struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1825 
1826 		if (cpu_up(cpu, target) && can_rollback_cpu(st)) {
1827 			/*
1828 			 * If this failed then cpu_up() might have only
1829 			 * rolled back to CPUHP_BP_KICK_AP for the final
1830 			 * online. Clean it up. NOOP if already rolled back.
1831 			 */
1832 			WARN_ON(cpuhp_invoke_callback_range(false, cpu, st, CPUHP_OFFLINE));
1833 		}
1834 
1835 		if (!--ncpus)
1836 			break;
1837 	}
1838 }
1839 
1840 #ifdef CONFIG_HOTPLUG_PARALLEL
1841 static bool __cpuhp_parallel_bringup __ro_after_init = true;
1842 
parallel_bringup_parse_param(char * arg)1843 static int __init parallel_bringup_parse_param(char *arg)
1844 {
1845 	return kstrtobool(arg, &__cpuhp_parallel_bringup);
1846 }
1847 early_param("cpuhp.parallel", parallel_bringup_parse_param);
1848 
cpuhp_smt_aware(void)1849 static inline bool cpuhp_smt_aware(void)
1850 {
1851 	return cpu_smt_max_threads > 1;
1852 }
1853 
cpuhp_get_primary_thread_mask(void)1854 static inline const struct cpumask *cpuhp_get_primary_thread_mask(void)
1855 {
1856 	return cpu_primary_thread_mask;
1857 }
1858 
1859 /*
1860  * On architectures which have enabled parallel bringup this invokes all BP
1861  * prepare states for each of the to be onlined APs first. The last state
1862  * sends the startup IPI to the APs. The APs proceed through the low level
1863  * bringup code in parallel and then wait for the control CPU to release
1864  * them one by one for the final onlining procedure.
1865  *
1866  * This avoids waiting for each AP to respond to the startup IPI in
1867  * CPUHP_BRINGUP_CPU.
1868  */
cpuhp_bringup_cpus_parallel(unsigned int ncpus)1869 static bool __init cpuhp_bringup_cpus_parallel(unsigned int ncpus)
1870 {
1871 	const struct cpumask *mask = cpu_present_mask;
1872 
1873 	if (__cpuhp_parallel_bringup)
1874 		__cpuhp_parallel_bringup = arch_cpuhp_init_parallel_bringup();
1875 	if (!__cpuhp_parallel_bringup)
1876 		return false;
1877 
1878 	if (cpuhp_smt_aware()) {
1879 		const struct cpumask *pmask = cpuhp_get_primary_thread_mask();
1880 		static struct cpumask tmp_mask __initdata;
1881 
1882 		/*
1883 		 * X86 requires to prevent that SMT siblings stopped while
1884 		 * the primary thread does a microcode update for various
1885 		 * reasons. Bring the primary threads up first.
1886 		 */
1887 		cpumask_and(&tmp_mask, mask, pmask);
1888 		cpuhp_bringup_mask(&tmp_mask, ncpus, CPUHP_BP_KICK_AP);
1889 		cpuhp_bringup_mask(&tmp_mask, ncpus, CPUHP_ONLINE);
1890 		/* Account for the online CPUs */
1891 		ncpus -= num_online_cpus();
1892 		if (!ncpus)
1893 			return true;
1894 		/* Create the mask for secondary CPUs */
1895 		cpumask_andnot(&tmp_mask, mask, pmask);
1896 		mask = &tmp_mask;
1897 	}
1898 
1899 	/* Bring the not-yet started CPUs up */
1900 	cpuhp_bringup_mask(mask, ncpus, CPUHP_BP_KICK_AP);
1901 	cpuhp_bringup_mask(mask, ncpus, CPUHP_ONLINE);
1902 	return true;
1903 }
1904 #else
cpuhp_bringup_cpus_parallel(unsigned int ncpus)1905 static inline bool cpuhp_bringup_cpus_parallel(unsigned int ncpus) { return false; }
1906 #endif /* CONFIG_HOTPLUG_PARALLEL */
1907 
bringup_nonboot_cpus(unsigned int setup_max_cpus)1908 void __init bringup_nonboot_cpus(unsigned int setup_max_cpus)
1909 {
1910 	/* Try parallel bringup optimization if enabled */
1911 	if (cpuhp_bringup_cpus_parallel(setup_max_cpus))
1912 		return;
1913 
1914 	/* Full per CPU serialized bringup */
1915 	cpuhp_bringup_mask(cpu_present_mask, setup_max_cpus, CPUHP_ONLINE);
1916 }
1917 
1918 #ifdef CONFIG_PM_SLEEP_SMP
1919 static cpumask_var_t frozen_cpus;
1920 
freeze_secondary_cpus(int primary)1921 int freeze_secondary_cpus(int primary)
1922 {
1923 	int cpu, error = 0;
1924 
1925 	cpu_maps_update_begin();
1926 	if (primary == -1) {
1927 		primary = cpumask_first(cpu_online_mask);
1928 		if (!housekeeping_cpu(primary, HK_TYPE_TIMER))
1929 			primary = housekeeping_any_cpu(HK_TYPE_TIMER);
1930 	} else {
1931 		if (!cpu_online(primary))
1932 			primary = cpumask_first(cpu_online_mask);
1933 	}
1934 
1935 	/*
1936 	 * We take down all of the non-boot CPUs in one shot to avoid races
1937 	 * with the userspace trying to use the CPU hotplug at the same time
1938 	 */
1939 	cpumask_clear(frozen_cpus);
1940 
1941 	pr_info("Disabling non-boot CPUs ...\n");
1942 	for_each_online_cpu(cpu) {
1943 		if (cpu == primary)
1944 			continue;
1945 
1946 		if (pm_wakeup_pending()) {
1947 			pr_info("Wakeup pending. Abort CPU freeze\n");
1948 			error = -EBUSY;
1949 			break;
1950 		}
1951 
1952 		trace_suspend_resume(TPS("CPU_OFF"), cpu, true);
1953 		error = _cpu_down(cpu, 1, CPUHP_OFFLINE);
1954 		trace_suspend_resume(TPS("CPU_OFF"), cpu, false);
1955 		if (!error)
1956 			cpumask_set_cpu(cpu, frozen_cpus);
1957 		else {
1958 			pr_err("Error taking CPU%d down: %d\n", cpu, error);
1959 			break;
1960 		}
1961 	}
1962 
1963 	if (!error)
1964 		BUG_ON(num_online_cpus() > 1);
1965 	else
1966 		pr_err("Non-boot CPUs are not disabled\n");
1967 
1968 	/*
1969 	 * Make sure the CPUs won't be enabled by someone else. We need to do
1970 	 * this even in case of failure as all freeze_secondary_cpus() users are
1971 	 * supposed to do thaw_secondary_cpus() on the failure path.
1972 	 */
1973 	cpu_hotplug_disabled++;
1974 
1975 	cpu_maps_update_done();
1976 	return error;
1977 }
1978 
arch_thaw_secondary_cpus_begin(void)1979 void __weak arch_thaw_secondary_cpus_begin(void)
1980 {
1981 }
1982 
arch_thaw_secondary_cpus_end(void)1983 void __weak arch_thaw_secondary_cpus_end(void)
1984 {
1985 }
1986 
thaw_secondary_cpus(void)1987 void thaw_secondary_cpus(void)
1988 {
1989 	int cpu, error;
1990 
1991 	/* Allow everyone to use the CPU hotplug again */
1992 	cpu_maps_update_begin();
1993 	__cpu_hotplug_enable();
1994 	if (cpumask_empty(frozen_cpus))
1995 		goto out;
1996 
1997 	pr_info("Enabling non-boot CPUs ...\n");
1998 
1999 	arch_thaw_secondary_cpus_begin();
2000 
2001 	for_each_cpu(cpu, frozen_cpus) {
2002 		trace_suspend_resume(TPS("CPU_ON"), cpu, true);
2003 		error = _cpu_up(cpu, 1, CPUHP_ONLINE);
2004 		trace_suspend_resume(TPS("CPU_ON"), cpu, false);
2005 		if (!error) {
2006 			pr_info("CPU%d is up\n", cpu);
2007 			continue;
2008 		}
2009 		pr_warn("Error taking CPU%d up: %d\n", cpu, error);
2010 	}
2011 
2012 	arch_thaw_secondary_cpus_end();
2013 
2014 	cpumask_clear(frozen_cpus);
2015 out:
2016 	cpu_maps_update_done();
2017 }
2018 
alloc_frozen_cpus(void)2019 static int __init alloc_frozen_cpus(void)
2020 {
2021 	if (!alloc_cpumask_var(&frozen_cpus, GFP_KERNEL|__GFP_ZERO))
2022 		return -ENOMEM;
2023 	return 0;
2024 }
2025 core_initcall(alloc_frozen_cpus);
2026 
2027 /*
2028  * When callbacks for CPU hotplug notifications are being executed, we must
2029  * ensure that the state of the system with respect to the tasks being frozen
2030  * or not, as reported by the notification, remains unchanged *throughout the
2031  * duration* of the execution of the callbacks.
2032  * Hence we need to prevent the freezer from racing with regular CPU hotplug.
2033  *
2034  * This synchronization is implemented by mutually excluding regular CPU
2035  * hotplug and Suspend/Hibernate call paths by hooking onto the Suspend/
2036  * Hibernate notifications.
2037  */
2038 static int
cpu_hotplug_pm_callback(struct notifier_block * nb,unsigned long action,void * ptr)2039 cpu_hotplug_pm_callback(struct notifier_block *nb,
2040 			unsigned long action, void *ptr)
2041 {
2042 	switch (action) {
2043 
2044 	case PM_SUSPEND_PREPARE:
2045 	case PM_HIBERNATION_PREPARE:
2046 		cpu_hotplug_disable();
2047 		break;
2048 
2049 	case PM_POST_SUSPEND:
2050 	case PM_POST_HIBERNATION:
2051 		cpu_hotplug_enable();
2052 		break;
2053 
2054 	default:
2055 		return NOTIFY_DONE;
2056 	}
2057 
2058 	return NOTIFY_OK;
2059 }
2060 
2061 
cpu_hotplug_pm_sync_init(void)2062 static int __init cpu_hotplug_pm_sync_init(void)
2063 {
2064 	/*
2065 	 * cpu_hotplug_pm_callback has higher priority than x86
2066 	 * bsp_pm_callback which depends on cpu_hotplug_pm_callback
2067 	 * to disable cpu hotplug to avoid cpu hotplug race.
2068 	 */
2069 	pm_notifier(cpu_hotplug_pm_callback, 0);
2070 	return 0;
2071 }
2072 core_initcall(cpu_hotplug_pm_sync_init);
2073 
2074 #endif /* CONFIG_PM_SLEEP_SMP */
2075 
2076 int __boot_cpu_id;
2077 
2078 #endif /* CONFIG_SMP */
2079 
2080 /* Boot processor state steps */
2081 static struct cpuhp_step cpuhp_hp_states[] = {
2082 	[CPUHP_OFFLINE] = {
2083 		.name			= "offline",
2084 		.startup.single		= NULL,
2085 		.teardown.single	= NULL,
2086 	},
2087 #ifdef CONFIG_SMP
2088 	[CPUHP_CREATE_THREADS]= {
2089 		.name			= "threads:prepare",
2090 		.startup.single		= smpboot_create_threads,
2091 		.teardown.single	= NULL,
2092 		.cant_stop		= true,
2093 	},
2094 	[CPUHP_PERF_PREPARE] = {
2095 		.name			= "perf:prepare",
2096 		.startup.single		= perf_event_init_cpu,
2097 		.teardown.single	= perf_event_exit_cpu,
2098 	},
2099 	[CPUHP_RANDOM_PREPARE] = {
2100 		.name			= "random:prepare",
2101 		.startup.single		= random_prepare_cpu,
2102 		.teardown.single	= NULL,
2103 	},
2104 	[CPUHP_WORKQUEUE_PREP] = {
2105 		.name			= "workqueue:prepare",
2106 		.startup.single		= workqueue_prepare_cpu,
2107 		.teardown.single	= NULL,
2108 	},
2109 	[CPUHP_HRTIMERS_PREPARE] = {
2110 		.name			= "hrtimers:prepare",
2111 		.startup.single		= hrtimers_prepare_cpu,
2112 		.teardown.single	= NULL,
2113 	},
2114 	[CPUHP_SMPCFD_PREPARE] = {
2115 		.name			= "smpcfd:prepare",
2116 		.startup.single		= smpcfd_prepare_cpu,
2117 		.teardown.single	= smpcfd_dead_cpu,
2118 	},
2119 	[CPUHP_RELAY_PREPARE] = {
2120 		.name			= "relay:prepare",
2121 		.startup.single		= relay_prepare_cpu,
2122 		.teardown.single	= NULL,
2123 	},
2124 	[CPUHP_SLAB_PREPARE] = {
2125 		.name			= "slab:prepare",
2126 		.startup.single		= slab_prepare_cpu,
2127 		.teardown.single	= slab_dead_cpu,
2128 	},
2129 	[CPUHP_RCUTREE_PREP] = {
2130 		.name			= "RCU/tree:prepare",
2131 		.startup.single		= rcutree_prepare_cpu,
2132 		.teardown.single	= rcutree_dead_cpu,
2133 	},
2134 	/*
2135 	 * On the tear-down path, timers_dead_cpu() must be invoked
2136 	 * before blk_mq_queue_reinit_notify() from notify_dead(),
2137 	 * otherwise a RCU stall occurs.
2138 	 */
2139 	[CPUHP_TIMERS_PREPARE] = {
2140 		.name			= "timers:prepare",
2141 		.startup.single		= timers_prepare_cpu,
2142 		.teardown.single	= timers_dead_cpu,
2143 	},
2144 
2145 #ifdef CONFIG_HOTPLUG_SPLIT_STARTUP
2146 	/*
2147 	 * Kicks the AP alive. AP will wait in cpuhp_ap_sync_alive() until
2148 	 * the next step will release it.
2149 	 */
2150 	[CPUHP_BP_KICK_AP] = {
2151 		.name			= "cpu:kick_ap",
2152 		.startup.single		= cpuhp_kick_ap_alive,
2153 	},
2154 
2155 	/*
2156 	 * Waits for the AP to reach cpuhp_ap_sync_alive() and then
2157 	 * releases it for the complete bringup.
2158 	 */
2159 	[CPUHP_BRINGUP_CPU] = {
2160 		.name			= "cpu:bringup",
2161 		.startup.single		= cpuhp_bringup_ap,
2162 		.teardown.single	= finish_cpu,
2163 		.cant_stop		= true,
2164 	},
2165 #else
2166 	/*
2167 	 * All-in-one CPU bringup state which includes the kick alive.
2168 	 */
2169 	[CPUHP_BRINGUP_CPU] = {
2170 		.name			= "cpu:bringup",
2171 		.startup.single		= bringup_cpu,
2172 		.teardown.single	= finish_cpu,
2173 		.cant_stop		= true,
2174 	},
2175 #endif
2176 	/* Final state before CPU kills itself */
2177 	[CPUHP_AP_IDLE_DEAD] = {
2178 		.name			= "idle:dead",
2179 	},
2180 	/*
2181 	 * Last state before CPU enters the idle loop to die. Transient state
2182 	 * for synchronization.
2183 	 */
2184 	[CPUHP_AP_OFFLINE] = {
2185 		.name			= "ap:offline",
2186 		.cant_stop		= true,
2187 	},
2188 	/* First state is scheduler control. Interrupts are disabled */
2189 	[CPUHP_AP_SCHED_STARTING] = {
2190 		.name			= "sched:starting",
2191 		.startup.single		= sched_cpu_starting,
2192 		.teardown.single	= sched_cpu_dying,
2193 	},
2194 	[CPUHP_AP_RCUTREE_DYING] = {
2195 		.name			= "RCU/tree:dying",
2196 		.startup.single		= NULL,
2197 		.teardown.single	= rcutree_dying_cpu,
2198 	},
2199 	[CPUHP_AP_SMPCFD_DYING] = {
2200 		.name			= "smpcfd:dying",
2201 		.startup.single		= NULL,
2202 		.teardown.single	= smpcfd_dying_cpu,
2203 	},
2204 	[CPUHP_AP_HRTIMERS_DYING] = {
2205 		.name			= "hrtimers:dying",
2206 		.startup.single		= NULL,
2207 		.teardown.single	= hrtimers_cpu_dying,
2208 	},
2209 
2210 	/* Entry state on starting. Interrupts enabled from here on. Transient
2211 	 * state for synchronsization */
2212 	[CPUHP_AP_ONLINE] = {
2213 		.name			= "ap:online",
2214 	},
2215 	/*
2216 	 * Handled on control processor until the plugged processor manages
2217 	 * this itself.
2218 	 */
2219 	[CPUHP_TEARDOWN_CPU] = {
2220 		.name			= "cpu:teardown",
2221 		.startup.single		= NULL,
2222 		.teardown.single	= takedown_cpu,
2223 		.cant_stop		= true,
2224 	},
2225 
2226 	[CPUHP_AP_SCHED_WAIT_EMPTY] = {
2227 		.name			= "sched:waitempty",
2228 		.startup.single		= NULL,
2229 		.teardown.single	= sched_cpu_wait_empty,
2230 	},
2231 
2232 	/* Handle smpboot threads park/unpark */
2233 	[CPUHP_AP_SMPBOOT_THREADS] = {
2234 		.name			= "smpboot/threads:online",
2235 		.startup.single		= smpboot_unpark_threads,
2236 		.teardown.single	= smpboot_park_threads,
2237 	},
2238 	[CPUHP_AP_IRQ_AFFINITY_ONLINE] = {
2239 		.name			= "irq/affinity:online",
2240 		.startup.single		= irq_affinity_online_cpu,
2241 		.teardown.single	= NULL,
2242 	},
2243 	[CPUHP_AP_PERF_ONLINE] = {
2244 		.name			= "perf:online",
2245 		.startup.single		= perf_event_init_cpu,
2246 		.teardown.single	= perf_event_exit_cpu,
2247 	},
2248 	[CPUHP_AP_WATCHDOG_ONLINE] = {
2249 		.name			= "lockup_detector:online",
2250 		.startup.single		= lockup_detector_online_cpu,
2251 		.teardown.single	= lockup_detector_offline_cpu,
2252 	},
2253 	[CPUHP_AP_WORKQUEUE_ONLINE] = {
2254 		.name			= "workqueue:online",
2255 		.startup.single		= workqueue_online_cpu,
2256 		.teardown.single	= workqueue_offline_cpu,
2257 	},
2258 	[CPUHP_AP_RANDOM_ONLINE] = {
2259 		.name			= "random:online",
2260 		.startup.single		= random_online_cpu,
2261 		.teardown.single	= NULL,
2262 	},
2263 	[CPUHP_AP_RCUTREE_ONLINE] = {
2264 		.name			= "RCU/tree:online",
2265 		.startup.single		= rcutree_online_cpu,
2266 		.teardown.single	= rcutree_offline_cpu,
2267 	},
2268 #endif
2269 	/*
2270 	 * The dynamically registered state space is here
2271 	 */
2272 
2273 #ifdef CONFIG_SMP
2274 	/* Last state is scheduler control setting the cpu active */
2275 	[CPUHP_AP_ACTIVE] = {
2276 		.name			= "sched:active",
2277 		.startup.single		= sched_cpu_activate,
2278 		.teardown.single	= sched_cpu_deactivate,
2279 	},
2280 #endif
2281 
2282 	/* CPU is fully up and running. */
2283 	[CPUHP_ONLINE] = {
2284 		.name			= "online",
2285 		.startup.single		= NULL,
2286 		.teardown.single	= NULL,
2287 	},
2288 };
2289 
2290 /* Sanity check for callbacks */
cpuhp_cb_check(enum cpuhp_state state)2291 static int cpuhp_cb_check(enum cpuhp_state state)
2292 {
2293 	if (state <= CPUHP_OFFLINE || state >= CPUHP_ONLINE)
2294 		return -EINVAL;
2295 	return 0;
2296 }
2297 
2298 /*
2299  * Returns a free for dynamic slot assignment of the Online state. The states
2300  * are protected by the cpuhp_slot_states mutex and an empty slot is identified
2301  * by having no name assigned.
2302  */
cpuhp_reserve_state(enum cpuhp_state state)2303 static int cpuhp_reserve_state(enum cpuhp_state state)
2304 {
2305 	enum cpuhp_state i, end;
2306 	struct cpuhp_step *step;
2307 
2308 	switch (state) {
2309 	case CPUHP_AP_ONLINE_DYN:
2310 		step = cpuhp_hp_states + CPUHP_AP_ONLINE_DYN;
2311 		end = CPUHP_AP_ONLINE_DYN_END;
2312 		break;
2313 	case CPUHP_BP_PREPARE_DYN:
2314 		step = cpuhp_hp_states + CPUHP_BP_PREPARE_DYN;
2315 		end = CPUHP_BP_PREPARE_DYN_END;
2316 		break;
2317 	default:
2318 		return -EINVAL;
2319 	}
2320 
2321 	for (i = state; i <= end; i++, step++) {
2322 		if (!step->name)
2323 			return i;
2324 	}
2325 	WARN(1, "No more dynamic states available for CPU hotplug\n");
2326 	return -ENOSPC;
2327 }
2328 
cpuhp_store_callbacks(enum cpuhp_state state,const char * name,int (* startup)(unsigned int cpu),int (* teardown)(unsigned int cpu),bool multi_instance)2329 static int cpuhp_store_callbacks(enum cpuhp_state state, const char *name,
2330 				 int (*startup)(unsigned int cpu),
2331 				 int (*teardown)(unsigned int cpu),
2332 				 bool multi_instance)
2333 {
2334 	/* (Un)Install the callbacks for further cpu hotplug operations */
2335 	struct cpuhp_step *sp;
2336 	int ret = 0;
2337 
2338 	/*
2339 	 * If name is NULL, then the state gets removed.
2340 	 *
2341 	 * CPUHP_AP_ONLINE_DYN and CPUHP_BP_PREPARE_DYN are handed out on
2342 	 * the first allocation from these dynamic ranges, so the removal
2343 	 * would trigger a new allocation and clear the wrong (already
2344 	 * empty) state, leaving the callbacks of the to be cleared state
2345 	 * dangling, which causes wreckage on the next hotplug operation.
2346 	 */
2347 	if (name && (state == CPUHP_AP_ONLINE_DYN ||
2348 		     state == CPUHP_BP_PREPARE_DYN)) {
2349 		ret = cpuhp_reserve_state(state);
2350 		if (ret < 0)
2351 			return ret;
2352 		state = ret;
2353 	}
2354 	sp = cpuhp_get_step(state);
2355 	if (name && sp->name)
2356 		return -EBUSY;
2357 
2358 	sp->startup.single = startup;
2359 	sp->teardown.single = teardown;
2360 	sp->name = name;
2361 	sp->multi_instance = multi_instance;
2362 	INIT_HLIST_HEAD(&sp->list);
2363 	return ret;
2364 }
2365 
cpuhp_get_teardown_cb(enum cpuhp_state state)2366 static void *cpuhp_get_teardown_cb(enum cpuhp_state state)
2367 {
2368 	return cpuhp_get_step(state)->teardown.single;
2369 }
2370 
2371 /*
2372  * Call the startup/teardown function for a step either on the AP or
2373  * on the current CPU.
2374  */
cpuhp_issue_call(int cpu,enum cpuhp_state state,bool bringup,struct hlist_node * node)2375 static int cpuhp_issue_call(int cpu, enum cpuhp_state state, bool bringup,
2376 			    struct hlist_node *node)
2377 {
2378 	struct cpuhp_step *sp = cpuhp_get_step(state);
2379 	int ret;
2380 
2381 	/*
2382 	 * If there's nothing to do, we done.
2383 	 * Relies on the union for multi_instance.
2384 	 */
2385 	if (cpuhp_step_empty(bringup, sp))
2386 		return 0;
2387 	/*
2388 	 * The non AP bound callbacks can fail on bringup. On teardown
2389 	 * e.g. module removal we crash for now.
2390 	 */
2391 #ifdef CONFIG_SMP
2392 	if (cpuhp_is_ap_state(state))
2393 		ret = cpuhp_invoke_ap_callback(cpu, state, bringup, node);
2394 	else
2395 		ret = cpuhp_invoke_callback(cpu, state, bringup, node, NULL);
2396 #else
2397 	ret = cpuhp_invoke_callback(cpu, state, bringup, node, NULL);
2398 #endif
2399 	BUG_ON(ret && !bringup);
2400 	return ret;
2401 }
2402 
2403 /*
2404  * Called from __cpuhp_setup_state on a recoverable failure.
2405  *
2406  * Note: The teardown callbacks for rollback are not allowed to fail!
2407  */
cpuhp_rollback_install(int failedcpu,enum cpuhp_state state,struct hlist_node * node)2408 static void cpuhp_rollback_install(int failedcpu, enum cpuhp_state state,
2409 				   struct hlist_node *node)
2410 {
2411 	int cpu;
2412 
2413 	/* Roll back the already executed steps on the other cpus */
2414 	for_each_present_cpu(cpu) {
2415 		struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
2416 		int cpustate = st->state;
2417 
2418 		if (cpu >= failedcpu)
2419 			break;
2420 
2421 		/* Did we invoke the startup call on that cpu ? */
2422 		if (cpustate >= state)
2423 			cpuhp_issue_call(cpu, state, false, node);
2424 	}
2425 }
2426 
__cpuhp_state_add_instance_cpuslocked(enum cpuhp_state state,struct hlist_node * node,bool invoke)2427 int __cpuhp_state_add_instance_cpuslocked(enum cpuhp_state state,
2428 					  struct hlist_node *node,
2429 					  bool invoke)
2430 {
2431 	struct cpuhp_step *sp;
2432 	int cpu;
2433 	int ret;
2434 
2435 	lockdep_assert_cpus_held();
2436 
2437 	sp = cpuhp_get_step(state);
2438 	if (sp->multi_instance == false)
2439 		return -EINVAL;
2440 
2441 	mutex_lock(&cpuhp_state_mutex);
2442 
2443 	if (!invoke || !sp->startup.multi)
2444 		goto add_node;
2445 
2446 	/*
2447 	 * Try to call the startup callback for each present cpu
2448 	 * depending on the hotplug state of the cpu.
2449 	 */
2450 	for_each_present_cpu(cpu) {
2451 		struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
2452 		int cpustate = st->state;
2453 
2454 		if (cpustate < state)
2455 			continue;
2456 
2457 		ret = cpuhp_issue_call(cpu, state, true, node);
2458 		if (ret) {
2459 			if (sp->teardown.multi)
2460 				cpuhp_rollback_install(cpu, state, node);
2461 			goto unlock;
2462 		}
2463 	}
2464 add_node:
2465 	ret = 0;
2466 	hlist_add_head(node, &sp->list);
2467 unlock:
2468 	mutex_unlock(&cpuhp_state_mutex);
2469 	return ret;
2470 }
2471 
__cpuhp_state_add_instance(enum cpuhp_state state,struct hlist_node * node,bool invoke)2472 int __cpuhp_state_add_instance(enum cpuhp_state state, struct hlist_node *node,
2473 			       bool invoke)
2474 {
2475 	int ret;
2476 
2477 	cpus_read_lock();
2478 	ret = __cpuhp_state_add_instance_cpuslocked(state, node, invoke);
2479 	cpus_read_unlock();
2480 	return ret;
2481 }
2482 EXPORT_SYMBOL_GPL(__cpuhp_state_add_instance);
2483 
2484 /**
2485  * __cpuhp_setup_state_cpuslocked - Setup the callbacks for an hotplug machine state
2486  * @state:		The state to setup
2487  * @name:		Name of the step
2488  * @invoke:		If true, the startup function is invoked for cpus where
2489  *			cpu state >= @state
2490  * @startup:		startup callback function
2491  * @teardown:		teardown callback function
2492  * @multi_instance:	State is set up for multiple instances which get
2493  *			added afterwards.
2494  *
2495  * The caller needs to hold cpus read locked while calling this function.
2496  * Return:
2497  *   On success:
2498  *      Positive state number if @state is CPUHP_AP_ONLINE_DYN;
2499  *      0 for all other states
2500  *   On failure: proper (negative) error code
2501  */
__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)2502 int __cpuhp_setup_state_cpuslocked(enum cpuhp_state state,
2503 				   const char *name, bool invoke,
2504 				   int (*startup)(unsigned int cpu),
2505 				   int (*teardown)(unsigned int cpu),
2506 				   bool multi_instance)
2507 {
2508 	int cpu, ret = 0;
2509 	bool dynstate;
2510 
2511 	lockdep_assert_cpus_held();
2512 
2513 	if (cpuhp_cb_check(state) || !name)
2514 		return -EINVAL;
2515 
2516 	mutex_lock(&cpuhp_state_mutex);
2517 
2518 	ret = cpuhp_store_callbacks(state, name, startup, teardown,
2519 				    multi_instance);
2520 
2521 	dynstate = state == CPUHP_AP_ONLINE_DYN;
2522 	if (ret > 0 && dynstate) {
2523 		state = ret;
2524 		ret = 0;
2525 	}
2526 
2527 	if (ret || !invoke || !startup)
2528 		goto out;
2529 
2530 	/*
2531 	 * Try to call the startup callback for each present cpu
2532 	 * depending on the hotplug state of the cpu.
2533 	 */
2534 	for_each_present_cpu(cpu) {
2535 		struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
2536 		int cpustate = st->state;
2537 
2538 		if (cpustate < state)
2539 			continue;
2540 
2541 		ret = cpuhp_issue_call(cpu, state, true, NULL);
2542 		if (ret) {
2543 			if (teardown)
2544 				cpuhp_rollback_install(cpu, state, NULL);
2545 			cpuhp_store_callbacks(state, NULL, NULL, NULL, false);
2546 			goto out;
2547 		}
2548 	}
2549 out:
2550 	mutex_unlock(&cpuhp_state_mutex);
2551 	/*
2552 	 * If the requested state is CPUHP_AP_ONLINE_DYN, return the
2553 	 * dynamically allocated state in case of success.
2554 	 */
2555 	if (!ret && dynstate)
2556 		return state;
2557 	return ret;
2558 }
2559 EXPORT_SYMBOL(__cpuhp_setup_state_cpuslocked);
2560 
__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)2561 int __cpuhp_setup_state(enum cpuhp_state state,
2562 			const char *name, bool invoke,
2563 			int (*startup)(unsigned int cpu),
2564 			int (*teardown)(unsigned int cpu),
2565 			bool multi_instance)
2566 {
2567 	int ret;
2568 
2569 	cpus_read_lock();
2570 	ret = __cpuhp_setup_state_cpuslocked(state, name, invoke, startup,
2571 					     teardown, multi_instance);
2572 	cpus_read_unlock();
2573 	return ret;
2574 }
2575 EXPORT_SYMBOL(__cpuhp_setup_state);
2576 
__cpuhp_state_remove_instance(enum cpuhp_state state,struct hlist_node * node,bool invoke)2577 int __cpuhp_state_remove_instance(enum cpuhp_state state,
2578 				  struct hlist_node *node, bool invoke)
2579 {
2580 	struct cpuhp_step *sp = cpuhp_get_step(state);
2581 	int cpu;
2582 
2583 	BUG_ON(cpuhp_cb_check(state));
2584 
2585 	if (!sp->multi_instance)
2586 		return -EINVAL;
2587 
2588 	cpus_read_lock();
2589 	mutex_lock(&cpuhp_state_mutex);
2590 
2591 	if (!invoke || !cpuhp_get_teardown_cb(state))
2592 		goto remove;
2593 	/*
2594 	 * Call the teardown callback for each present cpu depending
2595 	 * on the hotplug state of the cpu. This function is not
2596 	 * allowed to fail currently!
2597 	 */
2598 	for_each_present_cpu(cpu) {
2599 		struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
2600 		int cpustate = st->state;
2601 
2602 		if (cpustate >= state)
2603 			cpuhp_issue_call(cpu, state, false, node);
2604 	}
2605 
2606 remove:
2607 	hlist_del(node);
2608 	mutex_unlock(&cpuhp_state_mutex);
2609 	cpus_read_unlock();
2610 
2611 	return 0;
2612 }
2613 EXPORT_SYMBOL_GPL(__cpuhp_state_remove_instance);
2614 
2615 /**
2616  * __cpuhp_remove_state_cpuslocked - Remove the callbacks for an hotplug machine state
2617  * @state:	The state to remove
2618  * @invoke:	If true, the teardown function is invoked for cpus where
2619  *		cpu state >= @state
2620  *
2621  * The caller needs to hold cpus read locked while calling this function.
2622  * The teardown callback is currently not allowed to fail. Think
2623  * about module removal!
2624  */
__cpuhp_remove_state_cpuslocked(enum cpuhp_state state,bool invoke)2625 void __cpuhp_remove_state_cpuslocked(enum cpuhp_state state, bool invoke)
2626 {
2627 	struct cpuhp_step *sp = cpuhp_get_step(state);
2628 	int cpu;
2629 
2630 	BUG_ON(cpuhp_cb_check(state));
2631 
2632 	lockdep_assert_cpus_held();
2633 
2634 	mutex_lock(&cpuhp_state_mutex);
2635 	if (sp->multi_instance) {
2636 		WARN(!hlist_empty(&sp->list),
2637 		     "Error: Removing state %d which has instances left.\n",
2638 		     state);
2639 		goto remove;
2640 	}
2641 
2642 	if (!invoke || !cpuhp_get_teardown_cb(state))
2643 		goto remove;
2644 
2645 	/*
2646 	 * Call the teardown callback for each present cpu depending
2647 	 * on the hotplug state of the cpu. This function is not
2648 	 * allowed to fail currently!
2649 	 */
2650 	for_each_present_cpu(cpu) {
2651 		struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
2652 		int cpustate = st->state;
2653 
2654 		if (cpustate >= state)
2655 			cpuhp_issue_call(cpu, state, false, NULL);
2656 	}
2657 remove:
2658 	cpuhp_store_callbacks(state, NULL, NULL, NULL, false);
2659 	mutex_unlock(&cpuhp_state_mutex);
2660 }
2661 EXPORT_SYMBOL(__cpuhp_remove_state_cpuslocked);
2662 
__cpuhp_remove_state(enum cpuhp_state state,bool invoke)2663 void __cpuhp_remove_state(enum cpuhp_state state, bool invoke)
2664 {
2665 	cpus_read_lock();
2666 	__cpuhp_remove_state_cpuslocked(state, invoke);
2667 	cpus_read_unlock();
2668 }
2669 EXPORT_SYMBOL(__cpuhp_remove_state);
2670 
2671 #ifdef CONFIG_HOTPLUG_SMT
cpuhp_offline_cpu_device(unsigned int cpu)2672 static void cpuhp_offline_cpu_device(unsigned int cpu)
2673 {
2674 	struct device *dev = get_cpu_device(cpu);
2675 
2676 	dev->offline = true;
2677 	/* Tell user space about the state change */
2678 	kobject_uevent(&dev->kobj, KOBJ_OFFLINE);
2679 }
2680 
cpuhp_online_cpu_device(unsigned int cpu)2681 static void cpuhp_online_cpu_device(unsigned int cpu)
2682 {
2683 	struct device *dev = get_cpu_device(cpu);
2684 
2685 	dev->offline = false;
2686 	/* Tell user space about the state change */
2687 	kobject_uevent(&dev->kobj, KOBJ_ONLINE);
2688 }
2689 
cpuhp_smt_disable(enum cpuhp_smt_control ctrlval)2690 int cpuhp_smt_disable(enum cpuhp_smt_control ctrlval)
2691 {
2692 	int cpu, ret = 0;
2693 
2694 	cpu_maps_update_begin();
2695 	for_each_online_cpu(cpu) {
2696 		if (topology_is_primary_thread(cpu))
2697 			continue;
2698 		/*
2699 		 * Disable can be called with CPU_SMT_ENABLED when changing
2700 		 * from a higher to lower number of SMT threads per core.
2701 		 */
2702 		if (ctrlval == CPU_SMT_ENABLED && cpu_smt_thread_allowed(cpu))
2703 			continue;
2704 		ret = cpu_down_maps_locked(cpu, CPUHP_OFFLINE);
2705 		if (ret)
2706 			break;
2707 		/*
2708 		 * As this needs to hold the cpu maps lock it's impossible
2709 		 * to call device_offline() because that ends up calling
2710 		 * cpu_down() which takes cpu maps lock. cpu maps lock
2711 		 * needs to be held as this might race against in kernel
2712 		 * abusers of the hotplug machinery (thermal management).
2713 		 *
2714 		 * So nothing would update device:offline state. That would
2715 		 * leave the sysfs entry stale and prevent onlining after
2716 		 * smt control has been changed to 'off' again. This is
2717 		 * called under the sysfs hotplug lock, so it is properly
2718 		 * serialized against the regular offline usage.
2719 		 */
2720 		cpuhp_offline_cpu_device(cpu);
2721 	}
2722 	if (!ret)
2723 		cpu_smt_control = ctrlval;
2724 	cpu_maps_update_done();
2725 	return ret;
2726 }
2727 
cpuhp_smt_enable(void)2728 int cpuhp_smt_enable(void)
2729 {
2730 	int cpu, ret = 0;
2731 
2732 	cpu_maps_update_begin();
2733 	cpu_smt_control = CPU_SMT_ENABLED;
2734 	for_each_present_cpu(cpu) {
2735 		/* Skip online CPUs and CPUs on offline nodes */
2736 		if (cpu_online(cpu) || !node_online(cpu_to_node(cpu)))
2737 			continue;
2738 		if (!cpu_smt_thread_allowed(cpu))
2739 			continue;
2740 		ret = _cpu_up(cpu, 0, CPUHP_ONLINE);
2741 		if (ret)
2742 			break;
2743 		/* See comment in cpuhp_smt_disable() */
2744 		cpuhp_online_cpu_device(cpu);
2745 	}
2746 	cpu_maps_update_done();
2747 	return ret;
2748 }
2749 #endif
2750 
2751 #if defined(CONFIG_SYSFS) && defined(CONFIG_HOTPLUG_CPU)
state_show(struct device * dev,struct device_attribute * attr,char * buf)2752 static ssize_t state_show(struct device *dev,
2753 			  struct device_attribute *attr, char *buf)
2754 {
2755 	struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id);
2756 
2757 	return sprintf(buf, "%d\n", st->state);
2758 }
2759 static DEVICE_ATTR_RO(state);
2760 
target_store(struct device * dev,struct device_attribute * attr,const char * buf,size_t count)2761 static ssize_t target_store(struct device *dev, struct device_attribute *attr,
2762 			    const char *buf, size_t count)
2763 {
2764 	struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id);
2765 	struct cpuhp_step *sp;
2766 	int target, ret;
2767 
2768 	ret = kstrtoint(buf, 10, &target);
2769 	if (ret)
2770 		return ret;
2771 
2772 #ifdef CONFIG_CPU_HOTPLUG_STATE_CONTROL
2773 	if (target < CPUHP_OFFLINE || target > CPUHP_ONLINE)
2774 		return -EINVAL;
2775 #else
2776 	if (target != CPUHP_OFFLINE && target != CPUHP_ONLINE)
2777 		return -EINVAL;
2778 #endif
2779 
2780 	ret = lock_device_hotplug_sysfs();
2781 	if (ret)
2782 		return ret;
2783 
2784 	mutex_lock(&cpuhp_state_mutex);
2785 	sp = cpuhp_get_step(target);
2786 	ret = !sp->name || sp->cant_stop ? -EINVAL : 0;
2787 	mutex_unlock(&cpuhp_state_mutex);
2788 	if (ret)
2789 		goto out;
2790 
2791 	if (st->state < target)
2792 		ret = cpu_up(dev->id, target);
2793 	else if (st->state > target)
2794 		ret = cpu_down(dev->id, target);
2795 	else if (WARN_ON(st->target != target))
2796 		st->target = target;
2797 out:
2798 	unlock_device_hotplug();
2799 	return ret ? ret : count;
2800 }
2801 
target_show(struct device * dev,struct device_attribute * attr,char * buf)2802 static ssize_t target_show(struct device *dev,
2803 			   struct device_attribute *attr, char *buf)
2804 {
2805 	struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id);
2806 
2807 	return sprintf(buf, "%d\n", st->target);
2808 }
2809 static DEVICE_ATTR_RW(target);
2810 
fail_store(struct device * dev,struct device_attribute * attr,const char * buf,size_t count)2811 static ssize_t fail_store(struct device *dev, struct device_attribute *attr,
2812 			  const char *buf, size_t count)
2813 {
2814 	struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id);
2815 	struct cpuhp_step *sp;
2816 	int fail, ret;
2817 
2818 	ret = kstrtoint(buf, 10, &fail);
2819 	if (ret)
2820 		return ret;
2821 
2822 	if (fail == CPUHP_INVALID) {
2823 		st->fail = fail;
2824 		return count;
2825 	}
2826 
2827 	if (fail < CPUHP_OFFLINE || fail > CPUHP_ONLINE)
2828 		return -EINVAL;
2829 
2830 	/*
2831 	 * Cannot fail STARTING/DYING callbacks.
2832 	 */
2833 	if (cpuhp_is_atomic_state(fail))
2834 		return -EINVAL;
2835 
2836 	/*
2837 	 * DEAD callbacks cannot fail...
2838 	 * ... neither can CPUHP_BRINGUP_CPU during hotunplug. The latter
2839 	 * triggering STARTING callbacks, a failure in this state would
2840 	 * hinder rollback.
2841 	 */
2842 	if (fail <= CPUHP_BRINGUP_CPU && st->state > CPUHP_BRINGUP_CPU)
2843 		return -EINVAL;
2844 
2845 	/*
2846 	 * Cannot fail anything that doesn't have callbacks.
2847 	 */
2848 	mutex_lock(&cpuhp_state_mutex);
2849 	sp = cpuhp_get_step(fail);
2850 	if (!sp->startup.single && !sp->teardown.single)
2851 		ret = -EINVAL;
2852 	mutex_unlock(&cpuhp_state_mutex);
2853 	if (ret)
2854 		return ret;
2855 
2856 	st->fail = fail;
2857 
2858 	return count;
2859 }
2860 
fail_show(struct device * dev,struct device_attribute * attr,char * buf)2861 static ssize_t fail_show(struct device *dev,
2862 			 struct device_attribute *attr, char *buf)
2863 {
2864 	struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id);
2865 
2866 	return sprintf(buf, "%d\n", st->fail);
2867 }
2868 
2869 static DEVICE_ATTR_RW(fail);
2870 
2871 static struct attribute *cpuhp_cpu_attrs[] = {
2872 	&dev_attr_state.attr,
2873 	&dev_attr_target.attr,
2874 	&dev_attr_fail.attr,
2875 	NULL
2876 };
2877 
2878 static const struct attribute_group cpuhp_cpu_attr_group = {
2879 	.attrs = cpuhp_cpu_attrs,
2880 	.name = "hotplug",
2881 	NULL
2882 };
2883 
states_show(struct device * dev,struct device_attribute * attr,char * buf)2884 static ssize_t states_show(struct device *dev,
2885 				 struct device_attribute *attr, char *buf)
2886 {
2887 	ssize_t cur, res = 0;
2888 	int i;
2889 
2890 	mutex_lock(&cpuhp_state_mutex);
2891 	for (i = CPUHP_OFFLINE; i <= CPUHP_ONLINE; i++) {
2892 		struct cpuhp_step *sp = cpuhp_get_step(i);
2893 
2894 		if (sp->name) {
2895 			cur = sprintf(buf, "%3d: %s\n", i, sp->name);
2896 			buf += cur;
2897 			res += cur;
2898 		}
2899 	}
2900 	mutex_unlock(&cpuhp_state_mutex);
2901 	return res;
2902 }
2903 static DEVICE_ATTR_RO(states);
2904 
2905 static struct attribute *cpuhp_cpu_root_attrs[] = {
2906 	&dev_attr_states.attr,
2907 	NULL
2908 };
2909 
2910 static const struct attribute_group cpuhp_cpu_root_attr_group = {
2911 	.attrs = cpuhp_cpu_root_attrs,
2912 	.name = "hotplug",
2913 	NULL
2914 };
2915 
2916 #ifdef CONFIG_HOTPLUG_SMT
2917 
cpu_smt_num_threads_valid(unsigned int threads)2918 static bool cpu_smt_num_threads_valid(unsigned int threads)
2919 {
2920 	if (IS_ENABLED(CONFIG_SMT_NUM_THREADS_DYNAMIC))
2921 		return threads >= 1 && threads <= cpu_smt_max_threads;
2922 	return threads == 1 || threads == cpu_smt_max_threads;
2923 }
2924 
2925 static ssize_t
__store_smt_control(struct device * dev,struct device_attribute * attr,const char * buf,size_t count)2926 __store_smt_control(struct device *dev, struct device_attribute *attr,
2927 		    const char *buf, size_t count)
2928 {
2929 	int ctrlval, ret, num_threads, orig_threads;
2930 	bool force_off;
2931 
2932 	if (cpu_smt_control == CPU_SMT_FORCE_DISABLED)
2933 		return -EPERM;
2934 
2935 	if (cpu_smt_control == CPU_SMT_NOT_SUPPORTED)
2936 		return -ENODEV;
2937 
2938 	if (sysfs_streq(buf, "on")) {
2939 		ctrlval = CPU_SMT_ENABLED;
2940 		num_threads = cpu_smt_max_threads;
2941 	} else if (sysfs_streq(buf, "off")) {
2942 		ctrlval = CPU_SMT_DISABLED;
2943 		num_threads = 1;
2944 	} else if (sysfs_streq(buf, "forceoff")) {
2945 		ctrlval = CPU_SMT_FORCE_DISABLED;
2946 		num_threads = 1;
2947 	} else if (kstrtoint(buf, 10, &num_threads) == 0) {
2948 		if (num_threads == 1)
2949 			ctrlval = CPU_SMT_DISABLED;
2950 		else if (cpu_smt_num_threads_valid(num_threads))
2951 			ctrlval = CPU_SMT_ENABLED;
2952 		else
2953 			return -EINVAL;
2954 	} else {
2955 		return -EINVAL;
2956 	}
2957 
2958 	ret = lock_device_hotplug_sysfs();
2959 	if (ret)
2960 		return ret;
2961 
2962 	orig_threads = cpu_smt_num_threads;
2963 	cpu_smt_num_threads = num_threads;
2964 
2965 	force_off = ctrlval != cpu_smt_control && ctrlval == CPU_SMT_FORCE_DISABLED;
2966 
2967 	if (num_threads > orig_threads)
2968 		ret = cpuhp_smt_enable();
2969 	else if (num_threads < orig_threads || force_off)
2970 		ret = cpuhp_smt_disable(ctrlval);
2971 
2972 	unlock_device_hotplug();
2973 	return ret ? ret : count;
2974 }
2975 
2976 #else /* !CONFIG_HOTPLUG_SMT */
2977 static ssize_t
__store_smt_control(struct device * dev,struct device_attribute * attr,const char * buf,size_t count)2978 __store_smt_control(struct device *dev, struct device_attribute *attr,
2979 		    const char *buf, size_t count)
2980 {
2981 	return -ENODEV;
2982 }
2983 #endif /* CONFIG_HOTPLUG_SMT */
2984 
2985 static const char *smt_states[] = {
2986 	[CPU_SMT_ENABLED]		= "on",
2987 	[CPU_SMT_DISABLED]		= "off",
2988 	[CPU_SMT_FORCE_DISABLED]	= "forceoff",
2989 	[CPU_SMT_NOT_SUPPORTED]		= "notsupported",
2990 	[CPU_SMT_NOT_IMPLEMENTED]	= "notimplemented",
2991 };
2992 
control_show(struct device * dev,struct device_attribute * attr,char * buf)2993 static ssize_t control_show(struct device *dev,
2994 			    struct device_attribute *attr, char *buf)
2995 {
2996 	const char *state = smt_states[cpu_smt_control];
2997 
2998 #ifdef CONFIG_HOTPLUG_SMT
2999 	/*
3000 	 * If SMT is enabled but not all threads are enabled then show the
3001 	 * number of threads. If all threads are enabled show "on". Otherwise
3002 	 * show the state name.
3003 	 */
3004 	if (cpu_smt_control == CPU_SMT_ENABLED &&
3005 	    cpu_smt_num_threads != cpu_smt_max_threads)
3006 		return sysfs_emit(buf, "%d\n", cpu_smt_num_threads);
3007 #endif
3008 
3009 	return snprintf(buf, PAGE_SIZE - 2, "%s\n", state);
3010 }
3011 
control_store(struct device * dev,struct device_attribute * attr,const char * buf,size_t count)3012 static ssize_t control_store(struct device *dev, struct device_attribute *attr,
3013 			     const char *buf, size_t count)
3014 {
3015 	return __store_smt_control(dev, attr, buf, count);
3016 }
3017 static DEVICE_ATTR_RW(control);
3018 
active_show(struct device * dev,struct device_attribute * attr,char * buf)3019 static ssize_t active_show(struct device *dev,
3020 			   struct device_attribute *attr, char *buf)
3021 {
3022 	return snprintf(buf, PAGE_SIZE - 2, "%d\n", sched_smt_active());
3023 }
3024 static DEVICE_ATTR_RO(active);
3025 
3026 static struct attribute *cpuhp_smt_attrs[] = {
3027 	&dev_attr_control.attr,
3028 	&dev_attr_active.attr,
3029 	NULL
3030 };
3031 
3032 static const struct attribute_group cpuhp_smt_attr_group = {
3033 	.attrs = cpuhp_smt_attrs,
3034 	.name = "smt",
3035 	NULL
3036 };
3037 
cpu_smt_sysfs_init(void)3038 static int __init cpu_smt_sysfs_init(void)
3039 {
3040 	struct device *dev_root;
3041 	int ret = -ENODEV;
3042 
3043 	dev_root = bus_get_dev_root(&cpu_subsys);
3044 	if (dev_root) {
3045 		ret = sysfs_create_group(&dev_root->kobj, &cpuhp_smt_attr_group);
3046 		put_device(dev_root);
3047 	}
3048 	return ret;
3049 }
3050 
cpuhp_sysfs_init(void)3051 static int __init cpuhp_sysfs_init(void)
3052 {
3053 	struct device *dev_root;
3054 	int cpu, ret;
3055 
3056 	ret = cpu_smt_sysfs_init();
3057 	if (ret)
3058 		return ret;
3059 
3060 	dev_root = bus_get_dev_root(&cpu_subsys);
3061 	if (dev_root) {
3062 		ret = sysfs_create_group(&dev_root->kobj, &cpuhp_cpu_root_attr_group);
3063 		put_device(dev_root);
3064 		if (ret)
3065 			return ret;
3066 	}
3067 
3068 	for_each_possible_cpu(cpu) {
3069 		struct device *dev = get_cpu_device(cpu);
3070 
3071 		if (!dev)
3072 			continue;
3073 		ret = sysfs_create_group(&dev->kobj, &cpuhp_cpu_attr_group);
3074 		if (ret)
3075 			return ret;
3076 	}
3077 	return 0;
3078 }
3079 device_initcall(cpuhp_sysfs_init);
3080 #endif /* CONFIG_SYSFS && CONFIG_HOTPLUG_CPU */
3081 
3082 /*
3083  * cpu_bit_bitmap[] is a special, "compressed" data structure that
3084  * represents all NR_CPUS bits binary values of 1<<nr.
3085  *
3086  * It is used by cpumask_of() to get a constant address to a CPU
3087  * mask value that has a single bit set only.
3088  */
3089 
3090 /* cpu_bit_bitmap[0] is empty - so we can back into it */
3091 #define MASK_DECLARE_1(x)	[x+1][0] = (1UL << (x))
3092 #define MASK_DECLARE_2(x)	MASK_DECLARE_1(x), MASK_DECLARE_1(x+1)
3093 #define MASK_DECLARE_4(x)	MASK_DECLARE_2(x), MASK_DECLARE_2(x+2)
3094 #define MASK_DECLARE_8(x)	MASK_DECLARE_4(x), MASK_DECLARE_4(x+4)
3095 
3096 const unsigned long cpu_bit_bitmap[BITS_PER_LONG+1][BITS_TO_LONGS(NR_CPUS)] = {
3097 
3098 	MASK_DECLARE_8(0),	MASK_DECLARE_8(8),
3099 	MASK_DECLARE_8(16),	MASK_DECLARE_8(24),
3100 #if BITS_PER_LONG > 32
3101 	MASK_DECLARE_8(32),	MASK_DECLARE_8(40),
3102 	MASK_DECLARE_8(48),	MASK_DECLARE_8(56),
3103 #endif
3104 };
3105 EXPORT_SYMBOL_GPL(cpu_bit_bitmap);
3106 
3107 const DECLARE_BITMAP(cpu_all_bits, NR_CPUS) = CPU_BITS_ALL;
3108 EXPORT_SYMBOL(cpu_all_bits);
3109 
3110 #ifdef CONFIG_INIT_ALL_POSSIBLE
3111 struct cpumask __cpu_possible_mask __read_mostly
3112 	= {CPU_BITS_ALL};
3113 #else
3114 struct cpumask __cpu_possible_mask __read_mostly;
3115 #endif
3116 EXPORT_SYMBOL(__cpu_possible_mask);
3117 
3118 struct cpumask __cpu_online_mask __read_mostly;
3119 EXPORT_SYMBOL(__cpu_online_mask);
3120 
3121 struct cpumask __cpu_present_mask __read_mostly;
3122 EXPORT_SYMBOL(__cpu_present_mask);
3123 
3124 struct cpumask __cpu_active_mask __read_mostly;
3125 EXPORT_SYMBOL(__cpu_active_mask);
3126 
3127 struct cpumask __cpu_dying_mask __read_mostly;
3128 EXPORT_SYMBOL(__cpu_dying_mask);
3129 
3130 atomic_t __num_online_cpus __read_mostly;
3131 EXPORT_SYMBOL(__num_online_cpus);
3132 
init_cpu_present(const struct cpumask * src)3133 void init_cpu_present(const struct cpumask *src)
3134 {
3135 	cpumask_copy(&__cpu_present_mask, src);
3136 }
3137 
init_cpu_possible(const struct cpumask * src)3138 void init_cpu_possible(const struct cpumask *src)
3139 {
3140 	cpumask_copy(&__cpu_possible_mask, src);
3141 }
3142 
init_cpu_online(const struct cpumask * src)3143 void init_cpu_online(const struct cpumask *src)
3144 {
3145 	cpumask_copy(&__cpu_online_mask, src);
3146 }
3147 
set_cpu_online(unsigned int cpu,bool online)3148 void set_cpu_online(unsigned int cpu, bool online)
3149 {
3150 	/*
3151 	 * atomic_inc/dec() is required to handle the horrid abuse of this
3152 	 * function by the reboot and kexec code which invoke it from
3153 	 * IPI/NMI broadcasts when shutting down CPUs. Invocation from
3154 	 * regular CPU hotplug is properly serialized.
3155 	 *
3156 	 * Note, that the fact that __num_online_cpus is of type atomic_t
3157 	 * does not protect readers which are not serialized against
3158 	 * concurrent hotplug operations.
3159 	 */
3160 	if (online) {
3161 		if (!cpumask_test_and_set_cpu(cpu, &__cpu_online_mask))
3162 			atomic_inc(&__num_online_cpus);
3163 	} else {
3164 		if (cpumask_test_and_clear_cpu(cpu, &__cpu_online_mask))
3165 			atomic_dec(&__num_online_cpus);
3166 	}
3167 }
3168 
3169 /*
3170  * Activate the first processor.
3171  */
boot_cpu_init(void)3172 void __init boot_cpu_init(void)
3173 {
3174 	int cpu = smp_processor_id();
3175 
3176 	/* Mark the boot cpu "present", "online" etc for SMP and UP case */
3177 	set_cpu_online(cpu, true);
3178 	set_cpu_active(cpu, true);
3179 	set_cpu_present(cpu, true);
3180 	set_cpu_possible(cpu, true);
3181 
3182 #ifdef CONFIG_SMP
3183 	__boot_cpu_id = cpu;
3184 #endif
3185 }
3186 
3187 /*
3188  * Must be called _AFTER_ setting up the per_cpu areas
3189  */
boot_cpu_hotplug_init(void)3190 void __init boot_cpu_hotplug_init(void)
3191 {
3192 #ifdef CONFIG_SMP
3193 	cpumask_set_cpu(smp_processor_id(), &cpus_booted_once_mask);
3194 	atomic_set(this_cpu_ptr(&cpuhp_state.ap_sync_state), SYNC_STATE_ONLINE);
3195 #endif
3196 	this_cpu_write(cpuhp_state.state, CPUHP_ONLINE);
3197 	this_cpu_write(cpuhp_state.target, CPUHP_ONLINE);
3198 }
3199 
3200 /*
3201  * These are used for a global "mitigations=" cmdline option for toggling
3202  * optional CPU mitigations.
3203  */
3204 enum cpu_mitigations {
3205 	CPU_MITIGATIONS_OFF,
3206 	CPU_MITIGATIONS_AUTO,
3207 	CPU_MITIGATIONS_AUTO_NOSMT,
3208 };
3209 
3210 static enum cpu_mitigations cpu_mitigations __ro_after_init =
3211 	CPU_MITIGATIONS_AUTO;
3212 
mitigations_parse_cmdline(char * arg)3213 static int __init mitigations_parse_cmdline(char *arg)
3214 {
3215 	if (!strcmp(arg, "off"))
3216 		cpu_mitigations = CPU_MITIGATIONS_OFF;
3217 	else if (!strcmp(arg, "auto"))
3218 		cpu_mitigations = CPU_MITIGATIONS_AUTO;
3219 	else if (!strcmp(arg, "auto,nosmt"))
3220 		cpu_mitigations = CPU_MITIGATIONS_AUTO_NOSMT;
3221 	else
3222 		pr_crit("Unsupported mitigations=%s, system may still be vulnerable\n",
3223 			arg);
3224 
3225 	return 0;
3226 }
3227 early_param("mitigations", mitigations_parse_cmdline);
3228 
3229 /* mitigations=off */
cpu_mitigations_off(void)3230 bool cpu_mitigations_off(void)
3231 {
3232 	return cpu_mitigations == CPU_MITIGATIONS_OFF;
3233 }
3234 EXPORT_SYMBOL_GPL(cpu_mitigations_off);
3235 
3236 /* mitigations=auto,nosmt */
cpu_mitigations_auto_nosmt(void)3237 bool cpu_mitigations_auto_nosmt(void)
3238 {
3239 	return cpu_mitigations == CPU_MITIGATIONS_AUTO_NOSMT;
3240 }
3241 EXPORT_SYMBOL_GPL(cpu_mitigations_auto_nosmt);
3242