xref: /openbmc/linux/kernel/irq/irqdesc.c (revision f20c7d91)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (C) 1992, 1998-2006 Linus Torvalds, Ingo Molnar
4  * Copyright (C) 2005-2006, Thomas Gleixner, Russell King
5  *
6  * This file contains the interrupt descriptor management code. Detailed
7  * information is available in Documentation/core-api/genericirq.rst
8  *
9  */
10 #include <linux/irq.h>
11 #include <linux/slab.h>
12 #include <linux/export.h>
13 #include <linux/interrupt.h>
14 #include <linux/kernel_stat.h>
15 #include <linux/radix-tree.h>
16 #include <linux/bitmap.h>
17 #include <linux/irqdomain.h>
18 #include <linux/sysfs.h>
19 
20 #include "internals.h"
21 
22 /*
23  * lockdep: we want to handle all irq_desc locks as a single lock-class:
24  */
25 static struct lock_class_key irq_desc_lock_class;
26 
27 #if defined(CONFIG_SMP)
28 static int __init irq_affinity_setup(char *str)
29 {
30 	alloc_bootmem_cpumask_var(&irq_default_affinity);
31 	cpulist_parse(str, irq_default_affinity);
32 	/*
33 	 * Set at least the boot cpu. We don't want to end up with
34 	 * bugreports caused by random comandline masks
35 	 */
36 	cpumask_set_cpu(smp_processor_id(), irq_default_affinity);
37 	return 1;
38 }
39 __setup("irqaffinity=", irq_affinity_setup);
40 
41 static void __init init_irq_default_affinity(void)
42 {
43 	if (!cpumask_available(irq_default_affinity))
44 		zalloc_cpumask_var(&irq_default_affinity, GFP_NOWAIT);
45 	if (cpumask_empty(irq_default_affinity))
46 		cpumask_setall(irq_default_affinity);
47 }
48 #else
49 static void __init init_irq_default_affinity(void)
50 {
51 }
52 #endif
53 
54 #ifdef CONFIG_SMP
55 static int alloc_masks(struct irq_desc *desc, int node)
56 {
57 	if (!zalloc_cpumask_var_node(&desc->irq_common_data.affinity,
58 				     GFP_KERNEL, node))
59 		return -ENOMEM;
60 
61 #ifdef CONFIG_GENERIC_IRQ_EFFECTIVE_AFF_MASK
62 	if (!zalloc_cpumask_var_node(&desc->irq_common_data.effective_affinity,
63 				     GFP_KERNEL, node)) {
64 		free_cpumask_var(desc->irq_common_data.affinity);
65 		return -ENOMEM;
66 	}
67 #endif
68 
69 #ifdef CONFIG_GENERIC_PENDING_IRQ
70 	if (!zalloc_cpumask_var_node(&desc->pending_mask, GFP_KERNEL, node)) {
71 #ifdef CONFIG_GENERIC_IRQ_EFFECTIVE_AFF_MASK
72 		free_cpumask_var(desc->irq_common_data.effective_affinity);
73 #endif
74 		free_cpumask_var(desc->irq_common_data.affinity);
75 		return -ENOMEM;
76 	}
77 #endif
78 	return 0;
79 }
80 
81 static void desc_smp_init(struct irq_desc *desc, int node,
82 			  const struct cpumask *affinity)
83 {
84 	if (!affinity)
85 		affinity = irq_default_affinity;
86 	cpumask_copy(desc->irq_common_data.affinity, affinity);
87 
88 #ifdef CONFIG_GENERIC_PENDING_IRQ
89 	cpumask_clear(desc->pending_mask);
90 #endif
91 #ifdef CONFIG_NUMA
92 	desc->irq_common_data.node = node;
93 #endif
94 }
95 
96 #else
97 static inline int
98 alloc_masks(struct irq_desc *desc, int node) { return 0; }
99 static inline void
100 desc_smp_init(struct irq_desc *desc, int node, const struct cpumask *affinity) { }
101 #endif
102 
103 static void desc_set_defaults(unsigned int irq, struct irq_desc *desc, int node,
104 			      const struct cpumask *affinity, struct module *owner)
105 {
106 	int cpu;
107 
108 	desc->irq_common_data.handler_data = NULL;
109 	desc->irq_common_data.msi_desc = NULL;
110 
111 	desc->irq_data.common = &desc->irq_common_data;
112 	desc->irq_data.irq = irq;
113 	desc->irq_data.chip = &no_irq_chip;
114 	desc->irq_data.chip_data = NULL;
115 	irq_settings_clr_and_set(desc, ~0, _IRQ_DEFAULT_INIT_FLAGS);
116 	irqd_set(&desc->irq_data, IRQD_IRQ_DISABLED);
117 	irqd_set(&desc->irq_data, IRQD_IRQ_MASKED);
118 	desc->handle_irq = handle_bad_irq;
119 	desc->depth = 1;
120 	desc->irq_count = 0;
121 	desc->irqs_unhandled = 0;
122 	desc->tot_count = 0;
123 	desc->name = NULL;
124 	desc->owner = owner;
125 	for_each_possible_cpu(cpu)
126 		*per_cpu_ptr(desc->kstat_irqs, cpu) = 0;
127 	desc_smp_init(desc, node, affinity);
128 }
129 
130 int nr_irqs = NR_IRQS;
131 EXPORT_SYMBOL_GPL(nr_irqs);
132 
133 static DEFINE_MUTEX(sparse_irq_lock);
134 static DECLARE_BITMAP(allocated_irqs, IRQ_BITMAP_BITS);
135 
136 #ifdef CONFIG_SPARSE_IRQ
137 
138 static void irq_kobj_release(struct kobject *kobj);
139 
140 #ifdef CONFIG_SYSFS
141 static struct kobject *irq_kobj_base;
142 
143 #define IRQ_ATTR_RO(_name) \
144 static struct kobj_attribute _name##_attr = __ATTR_RO(_name)
145 
146 static ssize_t per_cpu_count_show(struct kobject *kobj,
147 				  struct kobj_attribute *attr, char *buf)
148 {
149 	struct irq_desc *desc = container_of(kobj, struct irq_desc, kobj);
150 	int cpu, irq = desc->irq_data.irq;
151 	ssize_t ret = 0;
152 	char *p = "";
153 
154 	for_each_possible_cpu(cpu) {
155 		unsigned int c = kstat_irqs_cpu(irq, cpu);
156 
157 		ret += scnprintf(buf + ret, PAGE_SIZE - ret, "%s%u", p, c);
158 		p = ",";
159 	}
160 
161 	ret += scnprintf(buf + ret, PAGE_SIZE - ret, "\n");
162 	return ret;
163 }
164 IRQ_ATTR_RO(per_cpu_count);
165 
166 static ssize_t chip_name_show(struct kobject *kobj,
167 			      struct kobj_attribute *attr, char *buf)
168 {
169 	struct irq_desc *desc = container_of(kobj, struct irq_desc, kobj);
170 	ssize_t ret = 0;
171 
172 	raw_spin_lock_irq(&desc->lock);
173 	if (desc->irq_data.chip && desc->irq_data.chip->name) {
174 		ret = scnprintf(buf, PAGE_SIZE, "%s\n",
175 				desc->irq_data.chip->name);
176 	}
177 	raw_spin_unlock_irq(&desc->lock);
178 
179 	return ret;
180 }
181 IRQ_ATTR_RO(chip_name);
182 
183 static ssize_t hwirq_show(struct kobject *kobj,
184 			  struct kobj_attribute *attr, char *buf)
185 {
186 	struct irq_desc *desc = container_of(kobj, struct irq_desc, kobj);
187 	ssize_t ret = 0;
188 
189 	raw_spin_lock_irq(&desc->lock);
190 	if (desc->irq_data.domain)
191 		ret = sprintf(buf, "%d\n", (int)desc->irq_data.hwirq);
192 	raw_spin_unlock_irq(&desc->lock);
193 
194 	return ret;
195 }
196 IRQ_ATTR_RO(hwirq);
197 
198 static ssize_t type_show(struct kobject *kobj,
199 			 struct kobj_attribute *attr, char *buf)
200 {
201 	struct irq_desc *desc = container_of(kobj, struct irq_desc, kobj);
202 	ssize_t ret = 0;
203 
204 	raw_spin_lock_irq(&desc->lock);
205 	ret = sprintf(buf, "%s\n",
206 		      irqd_is_level_type(&desc->irq_data) ? "level" : "edge");
207 	raw_spin_unlock_irq(&desc->lock);
208 
209 	return ret;
210 
211 }
212 IRQ_ATTR_RO(type);
213 
214 static ssize_t wakeup_show(struct kobject *kobj,
215 			   struct kobj_attribute *attr, char *buf)
216 {
217 	struct irq_desc *desc = container_of(kobj, struct irq_desc, kobj);
218 	ssize_t ret = 0;
219 
220 	raw_spin_lock_irq(&desc->lock);
221 	ret = sprintf(buf, "%s\n",
222 		      irqd_is_wakeup_set(&desc->irq_data) ? "enabled" : "disabled");
223 	raw_spin_unlock_irq(&desc->lock);
224 
225 	return ret;
226 
227 }
228 IRQ_ATTR_RO(wakeup);
229 
230 static ssize_t name_show(struct kobject *kobj,
231 			 struct kobj_attribute *attr, char *buf)
232 {
233 	struct irq_desc *desc = container_of(kobj, struct irq_desc, kobj);
234 	ssize_t ret = 0;
235 
236 	raw_spin_lock_irq(&desc->lock);
237 	if (desc->name)
238 		ret = scnprintf(buf, PAGE_SIZE, "%s\n", desc->name);
239 	raw_spin_unlock_irq(&desc->lock);
240 
241 	return ret;
242 }
243 IRQ_ATTR_RO(name);
244 
245 static ssize_t actions_show(struct kobject *kobj,
246 			    struct kobj_attribute *attr, char *buf)
247 {
248 	struct irq_desc *desc = container_of(kobj, struct irq_desc, kobj);
249 	struct irqaction *action;
250 	ssize_t ret = 0;
251 	char *p = "";
252 
253 	raw_spin_lock_irq(&desc->lock);
254 	for (action = desc->action; action != NULL; action = action->next) {
255 		ret += scnprintf(buf + ret, PAGE_SIZE - ret, "%s%s",
256 				 p, action->name);
257 		p = ",";
258 	}
259 	raw_spin_unlock_irq(&desc->lock);
260 
261 	if (ret)
262 		ret += scnprintf(buf + ret, PAGE_SIZE - ret, "\n");
263 
264 	return ret;
265 }
266 IRQ_ATTR_RO(actions);
267 
268 static struct attribute *irq_attrs[] = {
269 	&per_cpu_count_attr.attr,
270 	&chip_name_attr.attr,
271 	&hwirq_attr.attr,
272 	&type_attr.attr,
273 	&wakeup_attr.attr,
274 	&name_attr.attr,
275 	&actions_attr.attr,
276 	NULL
277 };
278 ATTRIBUTE_GROUPS(irq);
279 
280 static struct kobj_type irq_kobj_type = {
281 	.release	= irq_kobj_release,
282 	.sysfs_ops	= &kobj_sysfs_ops,
283 	.default_groups = irq_groups,
284 };
285 
286 static void irq_sysfs_add(int irq, struct irq_desc *desc)
287 {
288 	if (irq_kobj_base) {
289 		/*
290 		 * Continue even in case of failure as this is nothing
291 		 * crucial.
292 		 */
293 		if (kobject_add(&desc->kobj, irq_kobj_base, "%d", irq))
294 			pr_warn("Failed to add kobject for irq %d\n", irq);
295 	}
296 }
297 
298 static void irq_sysfs_del(struct irq_desc *desc)
299 {
300 	/*
301 	 * If irq_sysfs_init() has not yet been invoked (early boot), then
302 	 * irq_kobj_base is NULL and the descriptor was never added.
303 	 * kobject_del() complains about a object with no parent, so make
304 	 * it conditional.
305 	 */
306 	if (irq_kobj_base)
307 		kobject_del(&desc->kobj);
308 }
309 
310 static int __init irq_sysfs_init(void)
311 {
312 	struct irq_desc *desc;
313 	int irq;
314 
315 	/* Prevent concurrent irq alloc/free */
316 	irq_lock_sparse();
317 
318 	irq_kobj_base = kobject_create_and_add("irq", kernel_kobj);
319 	if (!irq_kobj_base) {
320 		irq_unlock_sparse();
321 		return -ENOMEM;
322 	}
323 
324 	/* Add the already allocated interrupts */
325 	for_each_irq_desc(irq, desc)
326 		irq_sysfs_add(irq, desc);
327 	irq_unlock_sparse();
328 
329 	return 0;
330 }
331 postcore_initcall(irq_sysfs_init);
332 
333 #else /* !CONFIG_SYSFS */
334 
335 static struct kobj_type irq_kobj_type = {
336 	.release	= irq_kobj_release,
337 };
338 
339 static void irq_sysfs_add(int irq, struct irq_desc *desc) {}
340 static void irq_sysfs_del(struct irq_desc *desc) {}
341 
342 #endif /* CONFIG_SYSFS */
343 
344 static RADIX_TREE(irq_desc_tree, GFP_KERNEL);
345 
346 static void irq_insert_desc(unsigned int irq, struct irq_desc *desc)
347 {
348 	radix_tree_insert(&irq_desc_tree, irq, desc);
349 }
350 
351 struct irq_desc *irq_to_desc(unsigned int irq)
352 {
353 	return radix_tree_lookup(&irq_desc_tree, irq);
354 }
355 EXPORT_SYMBOL(irq_to_desc);
356 
357 static void delete_irq_desc(unsigned int irq)
358 {
359 	radix_tree_delete(&irq_desc_tree, irq);
360 }
361 
362 #ifdef CONFIG_SMP
363 static void free_masks(struct irq_desc *desc)
364 {
365 #ifdef CONFIG_GENERIC_PENDING_IRQ
366 	free_cpumask_var(desc->pending_mask);
367 #endif
368 	free_cpumask_var(desc->irq_common_data.affinity);
369 #ifdef CONFIG_GENERIC_IRQ_EFFECTIVE_AFF_MASK
370 	free_cpumask_var(desc->irq_common_data.effective_affinity);
371 #endif
372 }
373 #else
374 static inline void free_masks(struct irq_desc *desc) { }
375 #endif
376 
377 void irq_lock_sparse(void)
378 {
379 	mutex_lock(&sparse_irq_lock);
380 }
381 
382 void irq_unlock_sparse(void)
383 {
384 	mutex_unlock(&sparse_irq_lock);
385 }
386 
387 static struct irq_desc *alloc_desc(int irq, int node, unsigned int flags,
388 				   const struct cpumask *affinity,
389 				   struct module *owner)
390 {
391 	struct irq_desc *desc;
392 
393 	desc = kzalloc_node(sizeof(*desc), GFP_KERNEL, node);
394 	if (!desc)
395 		return NULL;
396 	/* allocate based on nr_cpu_ids */
397 	desc->kstat_irqs = alloc_percpu(unsigned int);
398 	if (!desc->kstat_irqs)
399 		goto err_desc;
400 
401 	if (alloc_masks(desc, node))
402 		goto err_kstat;
403 
404 	raw_spin_lock_init(&desc->lock);
405 	lockdep_set_class(&desc->lock, &irq_desc_lock_class);
406 	mutex_init(&desc->request_mutex);
407 	init_rcu_head(&desc->rcu);
408 
409 	desc_set_defaults(irq, desc, node, affinity, owner);
410 	irqd_set(&desc->irq_data, flags);
411 	kobject_init(&desc->kobj, &irq_kobj_type);
412 
413 	return desc;
414 
415 err_kstat:
416 	free_percpu(desc->kstat_irqs);
417 err_desc:
418 	kfree(desc);
419 	return NULL;
420 }
421 
422 static void irq_kobj_release(struct kobject *kobj)
423 {
424 	struct irq_desc *desc = container_of(kobj, struct irq_desc, kobj);
425 
426 	free_masks(desc);
427 	free_percpu(desc->kstat_irqs);
428 	kfree(desc);
429 }
430 
431 static void delayed_free_desc(struct rcu_head *rhp)
432 {
433 	struct irq_desc *desc = container_of(rhp, struct irq_desc, rcu);
434 
435 	kobject_put(&desc->kobj);
436 }
437 
438 static void free_desc(unsigned int irq)
439 {
440 	struct irq_desc *desc = irq_to_desc(irq);
441 
442 	irq_remove_debugfs_entry(desc);
443 	unregister_irq_proc(irq, desc);
444 
445 	/*
446 	 * sparse_irq_lock protects also show_interrupts() and
447 	 * kstat_irq_usr(). Once we deleted the descriptor from the
448 	 * sparse tree we can free it. Access in proc will fail to
449 	 * lookup the descriptor.
450 	 *
451 	 * The sysfs entry must be serialized against a concurrent
452 	 * irq_sysfs_init() as well.
453 	 */
454 	irq_sysfs_del(desc);
455 	delete_irq_desc(irq);
456 
457 	/*
458 	 * We free the descriptor, masks and stat fields via RCU. That
459 	 * allows demultiplex interrupts to do rcu based management of
460 	 * the child interrupts.
461 	 * This also allows us to use rcu in kstat_irqs_usr().
462 	 */
463 	call_rcu(&desc->rcu, delayed_free_desc);
464 }
465 
466 static int alloc_descs(unsigned int start, unsigned int cnt, int node,
467 		       const struct irq_affinity_desc *affinity,
468 		       struct module *owner)
469 {
470 	struct irq_desc *desc;
471 	int i;
472 
473 	/* Validate affinity mask(s) */
474 	if (affinity) {
475 		for (i = 0; i < cnt; i++) {
476 			if (cpumask_empty(&affinity[i].mask))
477 				return -EINVAL;
478 		}
479 	}
480 
481 	for (i = 0; i < cnt; i++) {
482 		const struct cpumask *mask = NULL;
483 		unsigned int flags = 0;
484 
485 		if (affinity) {
486 			if (affinity->is_managed) {
487 				flags = IRQD_AFFINITY_MANAGED |
488 					IRQD_MANAGED_SHUTDOWN;
489 			}
490 			mask = &affinity->mask;
491 			node = cpu_to_node(cpumask_first(mask));
492 			affinity++;
493 		}
494 
495 		desc = alloc_desc(start + i, node, flags, mask, owner);
496 		if (!desc)
497 			goto err;
498 		irq_insert_desc(start + i, desc);
499 		irq_sysfs_add(start + i, desc);
500 		irq_add_debugfs_entry(start + i, desc);
501 	}
502 	bitmap_set(allocated_irqs, start, cnt);
503 	return start;
504 
505 err:
506 	for (i--; i >= 0; i--)
507 		free_desc(start + i);
508 	return -ENOMEM;
509 }
510 
511 static int irq_expand_nr_irqs(unsigned int nr)
512 {
513 	if (nr > IRQ_BITMAP_BITS)
514 		return -ENOMEM;
515 	nr_irqs = nr;
516 	return 0;
517 }
518 
519 int __init early_irq_init(void)
520 {
521 	int i, initcnt, node = first_online_node;
522 	struct irq_desc *desc;
523 
524 	init_irq_default_affinity();
525 
526 	/* Let arch update nr_irqs and return the nr of preallocated irqs */
527 	initcnt = arch_probe_nr_irqs();
528 	printk(KERN_INFO "NR_IRQS: %d, nr_irqs: %d, preallocated irqs: %d\n",
529 	       NR_IRQS, nr_irqs, initcnt);
530 
531 	if (WARN_ON(nr_irqs > IRQ_BITMAP_BITS))
532 		nr_irqs = IRQ_BITMAP_BITS;
533 
534 	if (WARN_ON(initcnt > IRQ_BITMAP_BITS))
535 		initcnt = IRQ_BITMAP_BITS;
536 
537 	if (initcnt > nr_irqs)
538 		nr_irqs = initcnt;
539 
540 	for (i = 0; i < initcnt; i++) {
541 		desc = alloc_desc(i, node, 0, NULL, NULL);
542 		set_bit(i, allocated_irqs);
543 		irq_insert_desc(i, desc);
544 	}
545 	return arch_early_irq_init();
546 }
547 
548 #else /* !CONFIG_SPARSE_IRQ */
549 
550 struct irq_desc irq_desc[NR_IRQS] __cacheline_aligned_in_smp = {
551 	[0 ... NR_IRQS-1] = {
552 		.handle_irq	= handle_bad_irq,
553 		.depth		= 1,
554 		.lock		= __RAW_SPIN_LOCK_UNLOCKED(irq_desc->lock),
555 	}
556 };
557 
558 int __init early_irq_init(void)
559 {
560 	int count, i, node = first_online_node;
561 	struct irq_desc *desc;
562 
563 	init_irq_default_affinity();
564 
565 	printk(KERN_INFO "NR_IRQS: %d\n", NR_IRQS);
566 
567 	desc = irq_desc;
568 	count = ARRAY_SIZE(irq_desc);
569 
570 	for (i = 0; i < count; i++) {
571 		desc[i].kstat_irqs = alloc_percpu(unsigned int);
572 		alloc_masks(&desc[i], node);
573 		raw_spin_lock_init(&desc[i].lock);
574 		lockdep_set_class(&desc[i].lock, &irq_desc_lock_class);
575 		mutex_init(&desc[i].request_mutex);
576 		desc_set_defaults(i, &desc[i], node, NULL, NULL);
577 	}
578 	return arch_early_irq_init();
579 }
580 
581 struct irq_desc *irq_to_desc(unsigned int irq)
582 {
583 	return (irq < NR_IRQS) ? irq_desc + irq : NULL;
584 }
585 EXPORT_SYMBOL(irq_to_desc);
586 
587 static void free_desc(unsigned int irq)
588 {
589 	struct irq_desc *desc = irq_to_desc(irq);
590 	unsigned long flags;
591 
592 	raw_spin_lock_irqsave(&desc->lock, flags);
593 	desc_set_defaults(irq, desc, irq_desc_get_node(desc), NULL, NULL);
594 	raw_spin_unlock_irqrestore(&desc->lock, flags);
595 }
596 
597 static inline int alloc_descs(unsigned int start, unsigned int cnt, int node,
598 			      const struct irq_affinity_desc *affinity,
599 			      struct module *owner)
600 {
601 	u32 i;
602 
603 	for (i = 0; i < cnt; i++) {
604 		struct irq_desc *desc = irq_to_desc(start + i);
605 
606 		desc->owner = owner;
607 	}
608 	bitmap_set(allocated_irqs, start, cnt);
609 	return start;
610 }
611 
612 static int irq_expand_nr_irqs(unsigned int nr)
613 {
614 	return -ENOMEM;
615 }
616 
617 void irq_mark_irq(unsigned int irq)
618 {
619 	mutex_lock(&sparse_irq_lock);
620 	bitmap_set(allocated_irqs, irq, 1);
621 	mutex_unlock(&sparse_irq_lock);
622 }
623 
624 #ifdef CONFIG_GENERIC_IRQ_LEGACY
625 void irq_init_desc(unsigned int irq)
626 {
627 	free_desc(irq);
628 }
629 #endif
630 
631 #endif /* !CONFIG_SPARSE_IRQ */
632 
633 /**
634  * generic_handle_irq - Invoke the handler for a particular irq
635  * @irq:	The irq number to handle
636  *
637  */
638 int generic_handle_irq(unsigned int irq)
639 {
640 	struct irq_desc *desc = irq_to_desc(irq);
641 	struct irq_data *data;
642 
643 	if (!desc)
644 		return -EINVAL;
645 
646 	data = irq_desc_get_irq_data(desc);
647 	if (WARN_ON_ONCE(!in_irq() && handle_enforce_irqctx(data)))
648 		return -EPERM;
649 
650 	generic_handle_irq_desc(desc);
651 	return 0;
652 }
653 EXPORT_SYMBOL_GPL(generic_handle_irq);
654 
655 #ifdef CONFIG_HANDLE_DOMAIN_IRQ
656 /**
657  * __handle_domain_irq - Invoke the handler for a HW irq belonging to a domain
658  * @domain:	The domain where to perform the lookup
659  * @hwirq:	The HW irq number to convert to a logical one
660  * @lookup:	Whether to perform the domain lookup or not
661  * @regs:	Register file coming from the low-level handling code
662  *
663  * Returns:	0 on success, or -EINVAL if conversion has failed
664  */
665 int __handle_domain_irq(struct irq_domain *domain, unsigned int hwirq,
666 			bool lookup, struct pt_regs *regs)
667 {
668 	struct pt_regs *old_regs = set_irq_regs(regs);
669 	unsigned int irq = hwirq;
670 	int ret = 0;
671 
672 	irq_enter();
673 
674 #ifdef CONFIG_IRQ_DOMAIN
675 	if (lookup)
676 		irq = irq_find_mapping(domain, hwirq);
677 #endif
678 
679 	/*
680 	 * Some hardware gives randomly wrong interrupts.  Rather
681 	 * than crashing, do something sensible.
682 	 */
683 	if (unlikely(!irq || irq >= nr_irqs)) {
684 		ack_bad_irq(irq);
685 		ret = -EINVAL;
686 	} else {
687 		generic_handle_irq(irq);
688 	}
689 
690 	irq_exit();
691 	set_irq_regs(old_regs);
692 	return ret;
693 }
694 
695 #ifdef CONFIG_IRQ_DOMAIN
696 /**
697  * handle_domain_nmi - Invoke the handler for a HW irq belonging to a domain
698  * @domain:	The domain where to perform the lookup
699  * @hwirq:	The HW irq number to convert to a logical one
700  * @regs:	Register file coming from the low-level handling code
701  *
702  *		This function must be called from an NMI context.
703  *
704  * Returns:	0 on success, or -EINVAL if conversion has failed
705  */
706 int handle_domain_nmi(struct irq_domain *domain, unsigned int hwirq,
707 		      struct pt_regs *regs)
708 {
709 	struct pt_regs *old_regs = set_irq_regs(regs);
710 	unsigned int irq;
711 	int ret = 0;
712 
713 	/*
714 	 * NMI context needs to be setup earlier in order to deal with tracing.
715 	 */
716 	WARN_ON(!in_nmi());
717 
718 	irq = irq_find_mapping(domain, hwirq);
719 
720 	/*
721 	 * ack_bad_irq is not NMI-safe, just report
722 	 * an invalid interrupt.
723 	 */
724 	if (likely(irq))
725 		generic_handle_irq(irq);
726 	else
727 		ret = -EINVAL;
728 
729 	set_irq_regs(old_regs);
730 	return ret;
731 }
732 #endif
733 #endif
734 
735 /* Dynamic interrupt handling */
736 
737 /**
738  * irq_free_descs - free irq descriptors
739  * @from:	Start of descriptor range
740  * @cnt:	Number of consecutive irqs to free
741  */
742 void irq_free_descs(unsigned int from, unsigned int cnt)
743 {
744 	int i;
745 
746 	if (from >= nr_irqs || (from + cnt) > nr_irqs)
747 		return;
748 
749 	mutex_lock(&sparse_irq_lock);
750 	for (i = 0; i < cnt; i++)
751 		free_desc(from + i);
752 
753 	bitmap_clear(allocated_irqs, from, cnt);
754 	mutex_unlock(&sparse_irq_lock);
755 }
756 EXPORT_SYMBOL_GPL(irq_free_descs);
757 
758 /**
759  * __irq_alloc_descs - allocate and initialize a range of irq descriptors
760  * @irq:	Allocate for specific irq number if irq >= 0
761  * @from:	Start the search from this irq number
762  * @cnt:	Number of consecutive irqs to allocate.
763  * @node:	Preferred node on which the irq descriptor should be allocated
764  * @owner:	Owning module (can be NULL)
765  * @affinity:	Optional pointer to an affinity mask array of size @cnt which
766  *		hints where the irq descriptors should be allocated and which
767  *		default affinities to use
768  *
769  * Returns the first irq number or error code
770  */
771 int __ref
772 __irq_alloc_descs(int irq, unsigned int from, unsigned int cnt, int node,
773 		  struct module *owner, const struct irq_affinity_desc *affinity)
774 {
775 	int start, ret;
776 
777 	if (!cnt)
778 		return -EINVAL;
779 
780 	if (irq >= 0) {
781 		if (from > irq)
782 			return -EINVAL;
783 		from = irq;
784 	} else {
785 		/*
786 		 * For interrupts which are freely allocated the
787 		 * architecture can force a lower bound to the @from
788 		 * argument. x86 uses this to exclude the GSI space.
789 		 */
790 		from = arch_dynirq_lower_bound(from);
791 	}
792 
793 	mutex_lock(&sparse_irq_lock);
794 
795 	start = bitmap_find_next_zero_area(allocated_irqs, IRQ_BITMAP_BITS,
796 					   from, cnt, 0);
797 	ret = -EEXIST;
798 	if (irq >=0 && start != irq)
799 		goto unlock;
800 
801 	if (start + cnt > nr_irqs) {
802 		ret = irq_expand_nr_irqs(start + cnt);
803 		if (ret)
804 			goto unlock;
805 	}
806 	ret = alloc_descs(start, cnt, node, affinity, owner);
807 unlock:
808 	mutex_unlock(&sparse_irq_lock);
809 	return ret;
810 }
811 EXPORT_SYMBOL_GPL(__irq_alloc_descs);
812 
813 #ifdef CONFIG_GENERIC_IRQ_LEGACY_ALLOC_HWIRQ
814 /**
815  * irq_alloc_hwirqs - Allocate an irq descriptor and initialize the hardware
816  * @cnt:	number of interrupts to allocate
817  * @node:	node on which to allocate
818  *
819  * Returns an interrupt number > 0 or 0, if the allocation fails.
820  */
821 unsigned int irq_alloc_hwirqs(int cnt, int node)
822 {
823 	int i, irq = __irq_alloc_descs(-1, 0, cnt, node, NULL, NULL);
824 
825 	if (irq < 0)
826 		return 0;
827 
828 	for (i = irq; cnt > 0; i++, cnt--) {
829 		if (arch_setup_hwirq(i, node))
830 			goto err;
831 		irq_clear_status_flags(i, _IRQ_NOREQUEST);
832 	}
833 	return irq;
834 
835 err:
836 	for (i--; i >= irq; i--) {
837 		irq_set_status_flags(i, _IRQ_NOREQUEST | _IRQ_NOPROBE);
838 		arch_teardown_hwirq(i);
839 	}
840 	irq_free_descs(irq, cnt);
841 	return 0;
842 }
843 EXPORT_SYMBOL_GPL(irq_alloc_hwirqs);
844 
845 /**
846  * irq_free_hwirqs - Free irq descriptor and cleanup the hardware
847  * @from:	Free from irq number
848  * @cnt:	number of interrupts to free
849  *
850  */
851 void irq_free_hwirqs(unsigned int from, int cnt)
852 {
853 	int i, j;
854 
855 	for (i = from, j = cnt; j > 0; i++, j--) {
856 		irq_set_status_flags(i, _IRQ_NOREQUEST | _IRQ_NOPROBE);
857 		arch_teardown_hwirq(i);
858 	}
859 	irq_free_descs(from, cnt);
860 }
861 EXPORT_SYMBOL_GPL(irq_free_hwirqs);
862 #endif
863 
864 /**
865  * irq_get_next_irq - get next allocated irq number
866  * @offset:	where to start the search
867  *
868  * Returns next irq number after offset or nr_irqs if none is found.
869  */
870 unsigned int irq_get_next_irq(unsigned int offset)
871 {
872 	return find_next_bit(allocated_irqs, nr_irqs, offset);
873 }
874 
875 struct irq_desc *
876 __irq_get_desc_lock(unsigned int irq, unsigned long *flags, bool bus,
877 		    unsigned int check)
878 {
879 	struct irq_desc *desc = irq_to_desc(irq);
880 
881 	if (desc) {
882 		if (check & _IRQ_DESC_CHECK) {
883 			if ((check & _IRQ_DESC_PERCPU) &&
884 			    !irq_settings_is_per_cpu_devid(desc))
885 				return NULL;
886 
887 			if (!(check & _IRQ_DESC_PERCPU) &&
888 			    irq_settings_is_per_cpu_devid(desc))
889 				return NULL;
890 		}
891 
892 		if (bus)
893 			chip_bus_lock(desc);
894 		raw_spin_lock_irqsave(&desc->lock, *flags);
895 	}
896 	return desc;
897 }
898 
899 void __irq_put_desc_unlock(struct irq_desc *desc, unsigned long flags, bool bus)
900 	__releases(&desc->lock)
901 {
902 	raw_spin_unlock_irqrestore(&desc->lock, flags);
903 	if (bus)
904 		chip_bus_sync_unlock(desc);
905 }
906 
907 int irq_set_percpu_devid_partition(unsigned int irq,
908 				   const struct cpumask *affinity)
909 {
910 	struct irq_desc *desc = irq_to_desc(irq);
911 
912 	if (!desc)
913 		return -EINVAL;
914 
915 	if (desc->percpu_enabled)
916 		return -EINVAL;
917 
918 	desc->percpu_enabled = kzalloc(sizeof(*desc->percpu_enabled), GFP_KERNEL);
919 
920 	if (!desc->percpu_enabled)
921 		return -ENOMEM;
922 
923 	if (affinity)
924 		desc->percpu_affinity = affinity;
925 	else
926 		desc->percpu_affinity = cpu_possible_mask;
927 
928 	irq_set_percpu_devid_flags(irq);
929 	return 0;
930 }
931 
932 int irq_set_percpu_devid(unsigned int irq)
933 {
934 	return irq_set_percpu_devid_partition(irq, NULL);
935 }
936 
937 int irq_get_percpu_devid_partition(unsigned int irq, struct cpumask *affinity)
938 {
939 	struct irq_desc *desc = irq_to_desc(irq);
940 
941 	if (!desc || !desc->percpu_enabled)
942 		return -EINVAL;
943 
944 	if (affinity)
945 		cpumask_copy(affinity, desc->percpu_affinity);
946 
947 	return 0;
948 }
949 EXPORT_SYMBOL_GPL(irq_get_percpu_devid_partition);
950 
951 void kstat_incr_irq_this_cpu(unsigned int irq)
952 {
953 	kstat_incr_irqs_this_cpu(irq_to_desc(irq));
954 }
955 
956 /**
957  * kstat_irqs_cpu - Get the statistics for an interrupt on a cpu
958  * @irq:	The interrupt number
959  * @cpu:	The cpu number
960  *
961  * Returns the sum of interrupt counts on @cpu since boot for
962  * @irq. The caller must ensure that the interrupt is not removed
963  * concurrently.
964  */
965 unsigned int kstat_irqs_cpu(unsigned int irq, int cpu)
966 {
967 	struct irq_desc *desc = irq_to_desc(irq);
968 
969 	return desc && desc->kstat_irqs ?
970 			*per_cpu_ptr(desc->kstat_irqs, cpu) : 0;
971 }
972 
973 static bool irq_is_nmi(struct irq_desc *desc)
974 {
975 	return desc->istate & IRQS_NMI;
976 }
977 
978 /**
979  * kstat_irqs - Get the statistics for an interrupt
980  * @irq:	The interrupt number
981  *
982  * Returns the sum of interrupt counts on all cpus since boot for
983  * @irq. The caller must ensure that the interrupt is not removed
984  * concurrently.
985  */
986 unsigned int kstat_irqs(unsigned int irq)
987 {
988 	struct irq_desc *desc = irq_to_desc(irq);
989 	unsigned int sum = 0;
990 	int cpu;
991 
992 	if (!desc || !desc->kstat_irqs)
993 		return 0;
994 	if (!irq_settings_is_per_cpu_devid(desc) &&
995 	    !irq_settings_is_per_cpu(desc) &&
996 	    !irq_is_nmi(desc))
997 	    return desc->tot_count;
998 
999 	for_each_possible_cpu(cpu)
1000 		sum += *per_cpu_ptr(desc->kstat_irqs, cpu);
1001 	return sum;
1002 }
1003 
1004 /**
1005  * kstat_irqs_usr - Get the statistics for an interrupt
1006  * @irq:	The interrupt number
1007  *
1008  * Returns the sum of interrupt counts on all cpus since boot for @irq.
1009  * Contrary to kstat_irqs() this can be called from any context.
1010  * It uses rcu since a concurrent removal of an interrupt descriptor is
1011  * observing an rcu grace period before delayed_free_desc()/irq_kobj_release().
1012  */
1013 unsigned int kstat_irqs_usr(unsigned int irq)
1014 {
1015 	unsigned int sum;
1016 
1017 	rcu_read_lock();
1018 	sum = kstat_irqs(irq);
1019 	rcu_read_unlock();
1020 	return sum;
1021 }
1022