Loading...
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 */
25static struct lock_class_key irq_desc_lock_class;
26
27#if defined(CONFIG_SMP)
28static 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 commandline masks
35 */
36 cpumask_set_cpu(smp_processor_id(), irq_default_affinity);
37 return 1;
38}
39__setup("irqaffinity=", irq_affinity_setup);
40
41static 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
49static void __init init_irq_default_affinity(void)
50{
51}
52#endif
53
54#ifdef CONFIG_SMP
55static 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
81static 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
97static inline int
98alloc_masks(struct irq_desc *desc, int node) { return 0; }
99static inline void
100desc_smp_init(struct irq_desc *desc, int node, const struct cpumask *affinity) { }
101#endif
102
103static 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
130int nr_irqs = NR_IRQS;
131EXPORT_SYMBOL_GPL(nr_irqs);
132
133static DEFINE_MUTEX(sparse_irq_lock);
134static DECLARE_BITMAP(allocated_irqs, IRQ_BITMAP_BITS);
135
136#ifdef CONFIG_SPARSE_IRQ
137
138static void irq_kobj_release(struct kobject *kobj);
139
140#ifdef CONFIG_SYSFS
141static struct kobject *irq_kobj_base;
142
143#define IRQ_ATTR_RO(_name) \
144static struct kobj_attribute _name##_attr = __ATTR_RO(_name)
145
146static 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 ssize_t ret = 0;
151 char *p = "";
152 int cpu;
153
154 for_each_possible_cpu(cpu) {
155 unsigned int c = irq_desc_kstat_cpu(desc, 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}
164IRQ_ATTR_RO(per_cpu_count);
165
166static 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}
181IRQ_ATTR_RO(chip_name);
182
183static 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, "%lu\n", desc->irq_data.hwirq);
192 raw_spin_unlock_irq(&desc->lock);
193
194 return ret;
195}
196IRQ_ATTR_RO(hwirq);
197
198static 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}
212IRQ_ATTR_RO(type);
213
214static 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}
228IRQ_ATTR_RO(wakeup);
229
230static 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}
243IRQ_ATTR_RO(name);
244
245static 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_each_action_of_desc(desc, action) {
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}
266IRQ_ATTR_RO(actions);
267
268static 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};
278ATTRIBUTE_GROUPS(irq);
279
280static struct kobj_type irq_kobj_type = {
281 .release = irq_kobj_release,
282 .sysfs_ops = &kobj_sysfs_ops,
283 .default_groups = irq_groups,
284};
285
286static 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 and failures in the late irq_sysfs_init()
292 * cannot be rolled back.
293 */
294 if (kobject_add(&desc->kobj, irq_kobj_base, "%d", irq))
295 pr_warn("Failed to add kobject for irq %d\n", irq);
296 else
297 desc->istate |= IRQS_SYSFS;
298 }
299}
300
301static void irq_sysfs_del(struct irq_desc *desc)
302{
303 /*
304 * Only invoke kobject_del() when kobject_add() was successfully
305 * invoked for the descriptor. This covers both early boot, where
306 * sysfs is not initialized yet, and the case of a failed
307 * kobject_add() invocation.
308 */
309 if (desc->istate & IRQS_SYSFS)
310 kobject_del(&desc->kobj);
311}
312
313static int __init irq_sysfs_init(void)
314{
315 struct irq_desc *desc;
316 int irq;
317
318 /* Prevent concurrent irq alloc/free */
319 irq_lock_sparse();
320
321 irq_kobj_base = kobject_create_and_add("irq", kernel_kobj);
322 if (!irq_kobj_base) {
323 irq_unlock_sparse();
324 return -ENOMEM;
325 }
326
327 /* Add the already allocated interrupts */
328 for_each_irq_desc(irq, desc)
329 irq_sysfs_add(irq, desc);
330 irq_unlock_sparse();
331
332 return 0;
333}
334postcore_initcall(irq_sysfs_init);
335
336#else /* !CONFIG_SYSFS */
337
338static struct kobj_type irq_kobj_type = {
339 .release = irq_kobj_release,
340};
341
342static void irq_sysfs_add(int irq, struct irq_desc *desc) {}
343static void irq_sysfs_del(struct irq_desc *desc) {}
344
345#endif /* CONFIG_SYSFS */
346
347static RADIX_TREE(irq_desc_tree, GFP_KERNEL);
348
349static void irq_insert_desc(unsigned int irq, struct irq_desc *desc)
350{
351 radix_tree_insert(&irq_desc_tree, irq, desc);
352}
353
354struct irq_desc *irq_to_desc(unsigned int irq)
355{
356 return radix_tree_lookup(&irq_desc_tree, irq);
357}
358#ifdef CONFIG_KVM_BOOK3S_64_HV_MODULE
359EXPORT_SYMBOL_GPL(irq_to_desc);
360#endif
361
362static void delete_irq_desc(unsigned int irq)
363{
364 radix_tree_delete(&irq_desc_tree, irq);
365}
366
367#ifdef CONFIG_SMP
368static void free_masks(struct irq_desc *desc)
369{
370#ifdef CONFIG_GENERIC_PENDING_IRQ
371 free_cpumask_var(desc->pending_mask);
372#endif
373 free_cpumask_var(desc->irq_common_data.affinity);
374#ifdef CONFIG_GENERIC_IRQ_EFFECTIVE_AFF_MASK
375 free_cpumask_var(desc->irq_common_data.effective_affinity);
376#endif
377}
378#else
379static inline void free_masks(struct irq_desc *desc) { }
380#endif
381
382void irq_lock_sparse(void)
383{
384 mutex_lock(&sparse_irq_lock);
385}
386
387void irq_unlock_sparse(void)
388{
389 mutex_unlock(&sparse_irq_lock);
390}
391
392static struct irq_desc *alloc_desc(int irq, int node, unsigned int flags,
393 const struct cpumask *affinity,
394 struct module *owner)
395{
396 struct irq_desc *desc;
397
398 desc = kzalloc_node(sizeof(*desc), GFP_KERNEL, node);
399 if (!desc)
400 return NULL;
401 /* allocate based on nr_cpu_ids */
402 desc->kstat_irqs = alloc_percpu(unsigned int);
403 if (!desc->kstat_irqs)
404 goto err_desc;
405
406 if (alloc_masks(desc, node))
407 goto err_kstat;
408
409 raw_spin_lock_init(&desc->lock);
410 lockdep_set_class(&desc->lock, &irq_desc_lock_class);
411 mutex_init(&desc->request_mutex);
412 init_rcu_head(&desc->rcu);
413 init_waitqueue_head(&desc->wait_for_threads);
414
415 desc_set_defaults(irq, desc, node, affinity, owner);
416 irqd_set(&desc->irq_data, flags);
417 kobject_init(&desc->kobj, &irq_kobj_type);
418
419 return desc;
420
421err_kstat:
422 free_percpu(desc->kstat_irqs);
423err_desc:
424 kfree(desc);
425 return NULL;
426}
427
428static void irq_kobj_release(struct kobject *kobj)
429{
430 struct irq_desc *desc = container_of(kobj, struct irq_desc, kobj);
431
432 free_masks(desc);
433 free_percpu(desc->kstat_irqs);
434 kfree(desc);
435}
436
437static void delayed_free_desc(struct rcu_head *rhp)
438{
439 struct irq_desc *desc = container_of(rhp, struct irq_desc, rcu);
440
441 kobject_put(&desc->kobj);
442}
443
444static void free_desc(unsigned int irq)
445{
446 struct irq_desc *desc = irq_to_desc(irq);
447
448 irq_remove_debugfs_entry(desc);
449 unregister_irq_proc(irq, desc);
450
451 /*
452 * sparse_irq_lock protects also show_interrupts() and
453 * kstat_irq_usr(). Once we deleted the descriptor from the
454 * sparse tree we can free it. Access in proc will fail to
455 * lookup the descriptor.
456 *
457 * The sysfs entry must be serialized against a concurrent
458 * irq_sysfs_init() as well.
459 */
460 irq_sysfs_del(desc);
461 delete_irq_desc(irq);
462
463 /*
464 * We free the descriptor, masks and stat fields via RCU. That
465 * allows demultiplex interrupts to do rcu based management of
466 * the child interrupts.
467 * This also allows us to use rcu in kstat_irqs_usr().
468 */
469 call_rcu(&desc->rcu, delayed_free_desc);
470}
471
472static int alloc_descs(unsigned int start, unsigned int cnt, int node,
473 const struct irq_affinity_desc *affinity,
474 struct module *owner)
475{
476 struct irq_desc *desc;
477 int i;
478
479 /* Validate affinity mask(s) */
480 if (affinity) {
481 for (i = 0; i < cnt; i++) {
482 if (cpumask_empty(&affinity[i].mask))
483 return -EINVAL;
484 }
485 }
486
487 for (i = 0; i < cnt; i++) {
488 const struct cpumask *mask = NULL;
489 unsigned int flags = 0;
490
491 if (affinity) {
492 if (affinity->is_managed) {
493 flags = IRQD_AFFINITY_MANAGED |
494 IRQD_MANAGED_SHUTDOWN;
495 }
496 mask = &affinity->mask;
497 node = cpu_to_node(cpumask_first(mask));
498 affinity++;
499 }
500
501 desc = alloc_desc(start + i, node, flags, mask, owner);
502 if (!desc)
503 goto err;
504 irq_insert_desc(start + i, desc);
505 irq_sysfs_add(start + i, desc);
506 irq_add_debugfs_entry(start + i, desc);
507 }
508 bitmap_set(allocated_irqs, start, cnt);
509 return start;
510
511err:
512 for (i--; i >= 0; i--)
513 free_desc(start + i);
514 return -ENOMEM;
515}
516
517static int irq_expand_nr_irqs(unsigned int nr)
518{
519 if (nr > IRQ_BITMAP_BITS)
520 return -ENOMEM;
521 nr_irqs = nr;
522 return 0;
523}
524
525int __init early_irq_init(void)
526{
527 int i, initcnt, node = first_online_node;
528 struct irq_desc *desc;
529
530 init_irq_default_affinity();
531
532 /* Let arch update nr_irqs and return the nr of preallocated irqs */
533 initcnt = arch_probe_nr_irqs();
534 printk(KERN_INFO "NR_IRQS: %d, nr_irqs: %d, preallocated irqs: %d\n",
535 NR_IRQS, nr_irqs, initcnt);
536
537 if (WARN_ON(nr_irqs > IRQ_BITMAP_BITS))
538 nr_irqs = IRQ_BITMAP_BITS;
539
540 if (WARN_ON(initcnt > IRQ_BITMAP_BITS))
541 initcnt = IRQ_BITMAP_BITS;
542
543 if (initcnt > nr_irqs)
544 nr_irqs = initcnt;
545
546 for (i = 0; i < initcnt; i++) {
547 desc = alloc_desc(i, node, 0, NULL, NULL);
548 set_bit(i, allocated_irqs);
549 irq_insert_desc(i, desc);
550 }
551 return arch_early_irq_init();
552}
553
554#else /* !CONFIG_SPARSE_IRQ */
555
556struct irq_desc irq_desc[NR_IRQS] __cacheline_aligned_in_smp = {
557 [0 ... NR_IRQS-1] = {
558 .handle_irq = handle_bad_irq,
559 .depth = 1,
560 .lock = __RAW_SPIN_LOCK_UNLOCKED(irq_desc->lock),
561 }
562};
563
564int __init early_irq_init(void)
565{
566 int count, i, node = first_online_node;
567 struct irq_desc *desc;
568
569 init_irq_default_affinity();
570
571 printk(KERN_INFO "NR_IRQS: %d\n", NR_IRQS);
572
573 desc = irq_desc;
574 count = ARRAY_SIZE(irq_desc);
575
576 for (i = 0; i < count; i++) {
577 desc[i].kstat_irqs = alloc_percpu(unsigned int);
578 alloc_masks(&desc[i], node);
579 raw_spin_lock_init(&desc[i].lock);
580 lockdep_set_class(&desc[i].lock, &irq_desc_lock_class);
581 mutex_init(&desc[i].request_mutex);
582 init_waitqueue_head(&desc[i].wait_for_threads);
583 desc_set_defaults(i, &desc[i], node, NULL, NULL);
584 }
585 return arch_early_irq_init();
586}
587
588struct irq_desc *irq_to_desc(unsigned int irq)
589{
590 return (irq < NR_IRQS) ? irq_desc + irq : NULL;
591}
592EXPORT_SYMBOL(irq_to_desc);
593
594static void free_desc(unsigned int irq)
595{
596 struct irq_desc *desc = irq_to_desc(irq);
597 unsigned long flags;
598
599 raw_spin_lock_irqsave(&desc->lock, flags);
600 desc_set_defaults(irq, desc, irq_desc_get_node(desc), NULL, NULL);
601 raw_spin_unlock_irqrestore(&desc->lock, flags);
602}
603
604static inline int alloc_descs(unsigned int start, unsigned int cnt, int node,
605 const struct irq_affinity_desc *affinity,
606 struct module *owner)
607{
608 u32 i;
609
610 for (i = 0; i < cnt; i++) {
611 struct irq_desc *desc = irq_to_desc(start + i);
612
613 desc->owner = owner;
614 }
615 bitmap_set(allocated_irqs, start, cnt);
616 return start;
617}
618
619static int irq_expand_nr_irqs(unsigned int nr)
620{
621 return -ENOMEM;
622}
623
624void irq_mark_irq(unsigned int irq)
625{
626 mutex_lock(&sparse_irq_lock);
627 bitmap_set(allocated_irqs, irq, 1);
628 mutex_unlock(&sparse_irq_lock);
629}
630
631#ifdef CONFIG_GENERIC_IRQ_LEGACY
632void irq_init_desc(unsigned int irq)
633{
634 free_desc(irq);
635}
636#endif
637
638#endif /* !CONFIG_SPARSE_IRQ */
639
640int handle_irq_desc(struct irq_desc *desc)
641{
642 struct irq_data *data;
643
644 if (!desc)
645 return -EINVAL;
646
647 data = irq_desc_get_irq_data(desc);
648 if (WARN_ON_ONCE(!in_hardirq() && handle_enforce_irqctx(data)))
649 return -EPERM;
650
651 generic_handle_irq_desc(desc);
652 return 0;
653}
654
655/**
656 * generic_handle_irq - Invoke the handler for a particular irq
657 * @irq: The irq number to handle
658 *
659 * Returns: 0 on success, or -EINVAL if conversion has failed
660 *
661 * This function must be called from an IRQ context with irq regs
662 * initialized.
663 */
664int generic_handle_irq(unsigned int irq)
665{
666 return handle_irq_desc(irq_to_desc(irq));
667}
668EXPORT_SYMBOL_GPL(generic_handle_irq);
669
670/**
671 * generic_handle_irq_safe - Invoke the handler for a particular irq from any
672 * context.
673 * @irq: The irq number to handle
674 *
675 * Returns: 0 on success, a negative value on error.
676 *
677 * This function can be called from any context (IRQ or process context). It
678 * will report an error if not invoked from IRQ context and the irq has been
679 * marked to enforce IRQ-context only.
680 */
681int generic_handle_irq_safe(unsigned int irq)
682{
683 unsigned long flags;
684 int ret;
685
686 local_irq_save(flags);
687 ret = handle_irq_desc(irq_to_desc(irq));
688 local_irq_restore(flags);
689 return ret;
690}
691EXPORT_SYMBOL_GPL(generic_handle_irq_safe);
692
693#ifdef CONFIG_IRQ_DOMAIN
694/**
695 * generic_handle_domain_irq - Invoke the handler for a HW irq belonging
696 * to a domain.
697 * @domain: The domain where to perform the lookup
698 * @hwirq: The HW irq number to convert to a logical one
699 *
700 * Returns: 0 on success, or -EINVAL if conversion has failed
701 *
702 * This function must be called from an IRQ context with irq regs
703 * initialized.
704 */
705int generic_handle_domain_irq(struct irq_domain *domain, unsigned int hwirq)
706{
707 return handle_irq_desc(irq_resolve_mapping(domain, hwirq));
708}
709EXPORT_SYMBOL_GPL(generic_handle_domain_irq);
710
711 /**
712 * generic_handle_irq_safe - Invoke the handler for a HW irq belonging
713 * to a domain from any context.
714 * @domain: The domain where to perform the lookup
715 * @hwirq: The HW irq number to convert to a logical one
716 *
717 * Returns: 0 on success, a negative value on error.
718 *
719 * This function can be called from any context (IRQ or process
720 * context). If the interrupt is marked as 'enforce IRQ-context only' then
721 * the function must be invoked from hard interrupt context.
722 */
723int generic_handle_domain_irq_safe(struct irq_domain *domain, unsigned int hwirq)
724{
725 unsigned long flags;
726 int ret;
727
728 local_irq_save(flags);
729 ret = handle_irq_desc(irq_resolve_mapping(domain, hwirq));
730 local_irq_restore(flags);
731 return ret;
732}
733EXPORT_SYMBOL_GPL(generic_handle_domain_irq_safe);
734
735/**
736 * generic_handle_domain_nmi - Invoke the handler for a HW nmi belonging
737 * to a domain.
738 * @domain: The domain where to perform the lookup
739 * @hwirq: The HW irq number to convert to a logical one
740 *
741 * Returns: 0 on success, or -EINVAL if conversion has failed
742 *
743 * This function must be called from an NMI context with irq regs
744 * initialized.
745 **/
746int generic_handle_domain_nmi(struct irq_domain *domain, unsigned int hwirq)
747{
748 WARN_ON_ONCE(!in_nmi());
749 return handle_irq_desc(irq_resolve_mapping(domain, hwirq));
750}
751#endif
752
753/* Dynamic interrupt handling */
754
755/**
756 * irq_free_descs - free irq descriptors
757 * @from: Start of descriptor range
758 * @cnt: Number of consecutive irqs to free
759 */
760void irq_free_descs(unsigned int from, unsigned int cnt)
761{
762 int i;
763
764 if (from >= nr_irqs || (from + cnt) > nr_irqs)
765 return;
766
767 mutex_lock(&sparse_irq_lock);
768 for (i = 0; i < cnt; i++)
769 free_desc(from + i);
770
771 bitmap_clear(allocated_irqs, from, cnt);
772 mutex_unlock(&sparse_irq_lock);
773}
774EXPORT_SYMBOL_GPL(irq_free_descs);
775
776/**
777 * __irq_alloc_descs - allocate and initialize a range of irq descriptors
778 * @irq: Allocate for specific irq number if irq >= 0
779 * @from: Start the search from this irq number
780 * @cnt: Number of consecutive irqs to allocate.
781 * @node: Preferred node on which the irq descriptor should be allocated
782 * @owner: Owning module (can be NULL)
783 * @affinity: Optional pointer to an affinity mask array of size @cnt which
784 * hints where the irq descriptors should be allocated and which
785 * default affinities to use
786 *
787 * Returns the first irq number or error code
788 */
789int __ref
790__irq_alloc_descs(int irq, unsigned int from, unsigned int cnt, int node,
791 struct module *owner, const struct irq_affinity_desc *affinity)
792{
793 int start, ret;
794
795 if (!cnt)
796 return -EINVAL;
797
798 if (irq >= 0) {
799 if (from > irq)
800 return -EINVAL;
801 from = irq;
802 } else {
803 /*
804 * For interrupts which are freely allocated the
805 * architecture can force a lower bound to the @from
806 * argument. x86 uses this to exclude the GSI space.
807 */
808 from = arch_dynirq_lower_bound(from);
809 }
810
811 mutex_lock(&sparse_irq_lock);
812
813 start = bitmap_find_next_zero_area(allocated_irqs, IRQ_BITMAP_BITS,
814 from, cnt, 0);
815 ret = -EEXIST;
816 if (irq >=0 && start != irq)
817 goto unlock;
818
819 if (start + cnt > nr_irqs) {
820 ret = irq_expand_nr_irqs(start + cnt);
821 if (ret)
822 goto unlock;
823 }
824 ret = alloc_descs(start, cnt, node, affinity, owner);
825unlock:
826 mutex_unlock(&sparse_irq_lock);
827 return ret;
828}
829EXPORT_SYMBOL_GPL(__irq_alloc_descs);
830
831/**
832 * irq_get_next_irq - get next allocated irq number
833 * @offset: where to start the search
834 *
835 * Returns next irq number after offset or nr_irqs if none is found.
836 */
837unsigned int irq_get_next_irq(unsigned int offset)
838{
839 return find_next_bit(allocated_irqs, nr_irqs, offset);
840}
841
842struct irq_desc *
843__irq_get_desc_lock(unsigned int irq, unsigned long *flags, bool bus,
844 unsigned int check)
845{
846 struct irq_desc *desc = irq_to_desc(irq);
847
848 if (desc) {
849 if (check & _IRQ_DESC_CHECK) {
850 if ((check & _IRQ_DESC_PERCPU) &&
851 !irq_settings_is_per_cpu_devid(desc))
852 return NULL;
853
854 if (!(check & _IRQ_DESC_PERCPU) &&
855 irq_settings_is_per_cpu_devid(desc))
856 return NULL;
857 }
858
859 if (bus)
860 chip_bus_lock(desc);
861 raw_spin_lock_irqsave(&desc->lock, *flags);
862 }
863 return desc;
864}
865
866void __irq_put_desc_unlock(struct irq_desc *desc, unsigned long flags, bool bus)
867 __releases(&desc->lock)
868{
869 raw_spin_unlock_irqrestore(&desc->lock, flags);
870 if (bus)
871 chip_bus_sync_unlock(desc);
872}
873
874int irq_set_percpu_devid_partition(unsigned int irq,
875 const struct cpumask *affinity)
876{
877 struct irq_desc *desc = irq_to_desc(irq);
878
879 if (!desc)
880 return -EINVAL;
881
882 if (desc->percpu_enabled)
883 return -EINVAL;
884
885 desc->percpu_enabled = kzalloc(sizeof(*desc->percpu_enabled), GFP_KERNEL);
886
887 if (!desc->percpu_enabled)
888 return -ENOMEM;
889
890 if (affinity)
891 desc->percpu_affinity = affinity;
892 else
893 desc->percpu_affinity = cpu_possible_mask;
894
895 irq_set_percpu_devid_flags(irq);
896 return 0;
897}
898
899int irq_set_percpu_devid(unsigned int irq)
900{
901 return irq_set_percpu_devid_partition(irq, NULL);
902}
903
904int irq_get_percpu_devid_partition(unsigned int irq, struct cpumask *affinity)
905{
906 struct irq_desc *desc = irq_to_desc(irq);
907
908 if (!desc || !desc->percpu_enabled)
909 return -EINVAL;
910
911 if (affinity)
912 cpumask_copy(affinity, desc->percpu_affinity);
913
914 return 0;
915}
916EXPORT_SYMBOL_GPL(irq_get_percpu_devid_partition);
917
918void kstat_incr_irq_this_cpu(unsigned int irq)
919{
920 kstat_incr_irqs_this_cpu(irq_to_desc(irq));
921}
922
923/**
924 * kstat_irqs_cpu - Get the statistics for an interrupt on a cpu
925 * @irq: The interrupt number
926 * @cpu: The cpu number
927 *
928 * Returns the sum of interrupt counts on @cpu since boot for
929 * @irq. The caller must ensure that the interrupt is not removed
930 * concurrently.
931 */
932unsigned int kstat_irqs_cpu(unsigned int irq, int cpu)
933{
934 struct irq_desc *desc = irq_to_desc(irq);
935
936 return desc && desc->kstat_irqs ?
937 *per_cpu_ptr(desc->kstat_irqs, cpu) : 0;
938}
939
940static bool irq_is_nmi(struct irq_desc *desc)
941{
942 return desc->istate & IRQS_NMI;
943}
944
945static unsigned int kstat_irqs(unsigned int irq)
946{
947 struct irq_desc *desc = irq_to_desc(irq);
948 unsigned int sum = 0;
949 int cpu;
950
951 if (!desc || !desc->kstat_irqs)
952 return 0;
953 if (!irq_settings_is_per_cpu_devid(desc) &&
954 !irq_settings_is_per_cpu(desc) &&
955 !irq_is_nmi(desc))
956 return data_race(desc->tot_count);
957
958 for_each_possible_cpu(cpu)
959 sum += data_race(*per_cpu_ptr(desc->kstat_irqs, cpu));
960 return sum;
961}
962
963/**
964 * kstat_irqs_usr - Get the statistics for an interrupt from thread context
965 * @irq: The interrupt number
966 *
967 * Returns the sum of interrupt counts on all cpus since boot for @irq.
968 *
969 * It uses rcu to protect the access since a concurrent removal of an
970 * interrupt descriptor is observing an rcu grace period before
971 * delayed_free_desc()/irq_kobj_release().
972 */
973unsigned int kstat_irqs_usr(unsigned int irq)
974{
975 unsigned int sum;
976
977 rcu_read_lock();
978 sum = kstat_irqs(irq);
979 rcu_read_unlock();
980 return sum;
981}
982
983#ifdef CONFIG_LOCKDEP
984void __irq_set_lockdep_class(unsigned int irq, struct lock_class_key *lock_class,
985 struct lock_class_key *request_class)
986{
987 struct irq_desc *desc = irq_to_desc(irq);
988
989 if (desc) {
990 lockdep_set_class(&desc->lock, lock_class);
991 lockdep_set_class(&desc->request_mutex, request_class);
992 }
993}
994EXPORT_SYMBOL_GPL(__irq_set_lockdep_class);
995#endif
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/maple_tree.h>
16#include <linux/irqdomain.h>
17#include <linux/sysfs.h>
18#include <linux/string_choices.h>
19
20#include "internals.h"
21
22/*
23 * lockdep: we want to handle all irq_desc locks as a single lock-class:
24 */
25static struct lock_class_key irq_desc_lock_class;
26
27#if defined(CONFIG_SMP)
28static 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 commandline masks
35 */
36 cpumask_set_cpu(smp_processor_id(), irq_default_affinity);
37 return 1;
38}
39__setup("irqaffinity=", irq_affinity_setup);
40
41static 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
49static void __init init_irq_default_affinity(void)
50{
51}
52#endif
53
54#ifdef CONFIG_SMP
55static 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
81static 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
96static void free_masks(struct irq_desc *desc)
97{
98#ifdef CONFIG_GENERIC_PENDING_IRQ
99 free_cpumask_var(desc->pending_mask);
100#endif
101 free_cpumask_var(desc->irq_common_data.affinity);
102#ifdef CONFIG_GENERIC_IRQ_EFFECTIVE_AFF_MASK
103 free_cpumask_var(desc->irq_common_data.effective_affinity);
104#endif
105}
106
107#else
108static inline int
109alloc_masks(struct irq_desc *desc, int node) { return 0; }
110static inline void
111desc_smp_init(struct irq_desc *desc, int node, const struct cpumask *affinity) { }
112static inline void free_masks(struct irq_desc *desc) { }
113#endif
114
115static void desc_set_defaults(unsigned int irq, struct irq_desc *desc, int node,
116 const struct cpumask *affinity, struct module *owner)
117{
118 int cpu;
119
120 desc->irq_common_data.handler_data = NULL;
121 desc->irq_common_data.msi_desc = NULL;
122
123 desc->irq_data.common = &desc->irq_common_data;
124 desc->irq_data.irq = irq;
125 desc->irq_data.chip = &no_irq_chip;
126 desc->irq_data.chip_data = NULL;
127 irq_settings_clr_and_set(desc, ~0, _IRQ_DEFAULT_INIT_FLAGS);
128 irqd_set(&desc->irq_data, IRQD_IRQ_DISABLED);
129 irqd_set(&desc->irq_data, IRQD_IRQ_MASKED);
130 desc->handle_irq = handle_bad_irq;
131 desc->depth = 1;
132 desc->irq_count = 0;
133 desc->irqs_unhandled = 0;
134 desc->tot_count = 0;
135 desc->name = NULL;
136 desc->owner = owner;
137 for_each_possible_cpu(cpu)
138 *per_cpu_ptr(desc->kstat_irqs, cpu) = (struct irqstat) { };
139 desc_smp_init(desc, node, affinity);
140}
141
142static unsigned int nr_irqs = NR_IRQS;
143
144/**
145 * irq_get_nr_irqs() - Number of interrupts supported by the system.
146 */
147unsigned int irq_get_nr_irqs(void)
148{
149 return nr_irqs;
150}
151EXPORT_SYMBOL_GPL(irq_get_nr_irqs);
152
153/**
154 * irq_set_nr_irqs() - Set the number of interrupts supported by the system.
155 * @nr: New number of interrupts.
156 *
157 * Return: @nr.
158 */
159unsigned int irq_set_nr_irqs(unsigned int nr)
160{
161 nr_irqs = nr;
162
163 return nr;
164}
165EXPORT_SYMBOL_GPL(irq_set_nr_irqs);
166
167static DEFINE_MUTEX(sparse_irq_lock);
168static struct maple_tree sparse_irqs = MTREE_INIT_EXT(sparse_irqs,
169 MT_FLAGS_ALLOC_RANGE |
170 MT_FLAGS_LOCK_EXTERN |
171 MT_FLAGS_USE_RCU,
172 sparse_irq_lock);
173
174static int irq_find_free_area(unsigned int from, unsigned int cnt)
175{
176 MA_STATE(mas, &sparse_irqs, 0, 0);
177
178 if (mas_empty_area(&mas, from, MAX_SPARSE_IRQS, cnt))
179 return -ENOSPC;
180 return mas.index;
181}
182
183static unsigned int irq_find_at_or_after(unsigned int offset)
184{
185 unsigned long index = offset;
186 struct irq_desc *desc;
187
188 guard(rcu)();
189 desc = mt_find(&sparse_irqs, &index, nr_irqs);
190
191 return desc ? irq_desc_get_irq(desc) : nr_irqs;
192}
193
194static void irq_insert_desc(unsigned int irq, struct irq_desc *desc)
195{
196 MA_STATE(mas, &sparse_irqs, irq, irq);
197 WARN_ON(mas_store_gfp(&mas, desc, GFP_KERNEL) != 0);
198}
199
200static void delete_irq_desc(unsigned int irq)
201{
202 MA_STATE(mas, &sparse_irqs, irq, irq);
203 mas_erase(&mas);
204}
205
206#ifdef CONFIG_SPARSE_IRQ
207static const struct kobj_type irq_kobj_type;
208#endif
209
210static int init_desc(struct irq_desc *desc, int irq, int node,
211 unsigned int flags,
212 const struct cpumask *affinity,
213 struct module *owner)
214{
215 desc->kstat_irqs = alloc_percpu(struct irqstat);
216 if (!desc->kstat_irqs)
217 return -ENOMEM;
218
219 if (alloc_masks(desc, node)) {
220 free_percpu(desc->kstat_irqs);
221 return -ENOMEM;
222 }
223
224 raw_spin_lock_init(&desc->lock);
225 lockdep_set_class(&desc->lock, &irq_desc_lock_class);
226 mutex_init(&desc->request_mutex);
227 init_waitqueue_head(&desc->wait_for_threads);
228 desc_set_defaults(irq, desc, node, affinity, owner);
229 irqd_set(&desc->irq_data, flags);
230 irq_resend_init(desc);
231#ifdef CONFIG_SPARSE_IRQ
232 kobject_init(&desc->kobj, &irq_kobj_type);
233 init_rcu_head(&desc->rcu);
234#endif
235
236 return 0;
237}
238
239#ifdef CONFIG_SPARSE_IRQ
240
241static void irq_kobj_release(struct kobject *kobj);
242
243#ifdef CONFIG_SYSFS
244static struct kobject *irq_kobj_base;
245
246#define IRQ_ATTR_RO(_name) \
247static struct kobj_attribute _name##_attr = __ATTR_RO(_name)
248
249static ssize_t per_cpu_count_show(struct kobject *kobj,
250 struct kobj_attribute *attr, char *buf)
251{
252 struct irq_desc *desc = container_of(kobj, struct irq_desc, kobj);
253 ssize_t ret = 0;
254 char *p = "";
255 int cpu;
256
257 for_each_possible_cpu(cpu) {
258 unsigned int c = irq_desc_kstat_cpu(desc, cpu);
259
260 ret += scnprintf(buf + ret, PAGE_SIZE - ret, "%s%u", p, c);
261 p = ",";
262 }
263
264 ret += scnprintf(buf + ret, PAGE_SIZE - ret, "\n");
265 return ret;
266}
267IRQ_ATTR_RO(per_cpu_count);
268
269static ssize_t chip_name_show(struct kobject *kobj,
270 struct kobj_attribute *attr, char *buf)
271{
272 struct irq_desc *desc = container_of(kobj, struct irq_desc, kobj);
273 ssize_t ret = 0;
274
275 raw_spin_lock_irq(&desc->lock);
276 if (desc->irq_data.chip && desc->irq_data.chip->name) {
277 ret = scnprintf(buf, PAGE_SIZE, "%s\n",
278 desc->irq_data.chip->name);
279 }
280 raw_spin_unlock_irq(&desc->lock);
281
282 return ret;
283}
284IRQ_ATTR_RO(chip_name);
285
286static ssize_t hwirq_show(struct kobject *kobj,
287 struct kobj_attribute *attr, char *buf)
288{
289 struct irq_desc *desc = container_of(kobj, struct irq_desc, kobj);
290 ssize_t ret = 0;
291
292 raw_spin_lock_irq(&desc->lock);
293 if (desc->irq_data.domain)
294 ret = sprintf(buf, "%lu\n", desc->irq_data.hwirq);
295 raw_spin_unlock_irq(&desc->lock);
296
297 return ret;
298}
299IRQ_ATTR_RO(hwirq);
300
301static ssize_t type_show(struct kobject *kobj,
302 struct kobj_attribute *attr, char *buf)
303{
304 struct irq_desc *desc = container_of(kobj, struct irq_desc, kobj);
305 ssize_t ret = 0;
306
307 raw_spin_lock_irq(&desc->lock);
308 ret = sprintf(buf, "%s\n",
309 irqd_is_level_type(&desc->irq_data) ? "level" : "edge");
310 raw_spin_unlock_irq(&desc->lock);
311
312 return ret;
313
314}
315IRQ_ATTR_RO(type);
316
317static ssize_t wakeup_show(struct kobject *kobj,
318 struct kobj_attribute *attr, char *buf)
319{
320 struct irq_desc *desc = container_of(kobj, struct irq_desc, kobj);
321 ssize_t ret = 0;
322
323 raw_spin_lock_irq(&desc->lock);
324 ret = sprintf(buf, "%s\n", str_enabled_disabled(irqd_is_wakeup_set(&desc->irq_data)));
325 raw_spin_unlock_irq(&desc->lock);
326
327 return ret;
328
329}
330IRQ_ATTR_RO(wakeup);
331
332static ssize_t name_show(struct kobject *kobj,
333 struct kobj_attribute *attr, char *buf)
334{
335 struct irq_desc *desc = container_of(kobj, struct irq_desc, kobj);
336 ssize_t ret = 0;
337
338 raw_spin_lock_irq(&desc->lock);
339 if (desc->name)
340 ret = scnprintf(buf, PAGE_SIZE, "%s\n", desc->name);
341 raw_spin_unlock_irq(&desc->lock);
342
343 return ret;
344}
345IRQ_ATTR_RO(name);
346
347static ssize_t actions_show(struct kobject *kobj,
348 struct kobj_attribute *attr, char *buf)
349{
350 struct irq_desc *desc = container_of(kobj, struct irq_desc, kobj);
351 struct irqaction *action;
352 ssize_t ret = 0;
353 char *p = "";
354
355 raw_spin_lock_irq(&desc->lock);
356 for_each_action_of_desc(desc, action) {
357 ret += scnprintf(buf + ret, PAGE_SIZE - ret, "%s%s",
358 p, action->name);
359 p = ",";
360 }
361 raw_spin_unlock_irq(&desc->lock);
362
363 if (ret)
364 ret += scnprintf(buf + ret, PAGE_SIZE - ret, "\n");
365
366 return ret;
367}
368IRQ_ATTR_RO(actions);
369
370static struct attribute *irq_attrs[] = {
371 &per_cpu_count_attr.attr,
372 &chip_name_attr.attr,
373 &hwirq_attr.attr,
374 &type_attr.attr,
375 &wakeup_attr.attr,
376 &name_attr.attr,
377 &actions_attr.attr,
378 NULL
379};
380ATTRIBUTE_GROUPS(irq);
381
382static const struct kobj_type irq_kobj_type = {
383 .release = irq_kobj_release,
384 .sysfs_ops = &kobj_sysfs_ops,
385 .default_groups = irq_groups,
386};
387
388static void irq_sysfs_add(int irq, struct irq_desc *desc)
389{
390 if (irq_kobj_base) {
391 /*
392 * Continue even in case of failure as this is nothing
393 * crucial and failures in the late irq_sysfs_init()
394 * cannot be rolled back.
395 */
396 if (kobject_add(&desc->kobj, irq_kobj_base, "%d", irq))
397 pr_warn("Failed to add kobject for irq %d\n", irq);
398 else
399 desc->istate |= IRQS_SYSFS;
400 }
401}
402
403static void irq_sysfs_del(struct irq_desc *desc)
404{
405 /*
406 * Only invoke kobject_del() when kobject_add() was successfully
407 * invoked for the descriptor. This covers both early boot, where
408 * sysfs is not initialized yet, and the case of a failed
409 * kobject_add() invocation.
410 */
411 if (desc->istate & IRQS_SYSFS)
412 kobject_del(&desc->kobj);
413}
414
415static int __init irq_sysfs_init(void)
416{
417 struct irq_desc *desc;
418 int irq;
419
420 /* Prevent concurrent irq alloc/free */
421 irq_lock_sparse();
422
423 irq_kobj_base = kobject_create_and_add("irq", kernel_kobj);
424 if (!irq_kobj_base) {
425 irq_unlock_sparse();
426 return -ENOMEM;
427 }
428
429 /* Add the already allocated interrupts */
430 for_each_irq_desc(irq, desc)
431 irq_sysfs_add(irq, desc);
432 irq_unlock_sparse();
433
434 return 0;
435}
436postcore_initcall(irq_sysfs_init);
437
438#else /* !CONFIG_SYSFS */
439
440static const struct kobj_type irq_kobj_type = {
441 .release = irq_kobj_release,
442};
443
444static void irq_sysfs_add(int irq, struct irq_desc *desc) {}
445static void irq_sysfs_del(struct irq_desc *desc) {}
446
447#endif /* CONFIG_SYSFS */
448
449struct irq_desc *irq_to_desc(unsigned int irq)
450{
451 return mtree_load(&sparse_irqs, irq);
452}
453#ifdef CONFIG_KVM_BOOK3S_64_HV_MODULE
454EXPORT_SYMBOL_GPL(irq_to_desc);
455#endif
456
457void irq_lock_sparse(void)
458{
459 mutex_lock(&sparse_irq_lock);
460}
461
462void irq_unlock_sparse(void)
463{
464 mutex_unlock(&sparse_irq_lock);
465}
466
467static struct irq_desc *alloc_desc(int irq, int node, unsigned int flags,
468 const struct cpumask *affinity,
469 struct module *owner)
470{
471 struct irq_desc *desc;
472 int ret;
473
474 desc = kzalloc_node(sizeof(*desc), GFP_KERNEL, node);
475 if (!desc)
476 return NULL;
477
478 ret = init_desc(desc, irq, node, flags, affinity, owner);
479 if (unlikely(ret)) {
480 kfree(desc);
481 return NULL;
482 }
483
484 return desc;
485}
486
487static void irq_kobj_release(struct kobject *kobj)
488{
489 struct irq_desc *desc = container_of(kobj, struct irq_desc, kobj);
490
491 free_masks(desc);
492 free_percpu(desc->kstat_irqs);
493 kfree(desc);
494}
495
496static void delayed_free_desc(struct rcu_head *rhp)
497{
498 struct irq_desc *desc = container_of(rhp, struct irq_desc, rcu);
499
500 kobject_put(&desc->kobj);
501}
502
503static void free_desc(unsigned int irq)
504{
505 struct irq_desc *desc = irq_to_desc(irq);
506
507 irq_remove_debugfs_entry(desc);
508 unregister_irq_proc(irq, desc);
509
510 /*
511 * sparse_irq_lock protects also show_interrupts() and
512 * kstat_irq_usr(). Once we deleted the descriptor from the
513 * sparse tree we can free it. Access in proc will fail to
514 * lookup the descriptor.
515 *
516 * The sysfs entry must be serialized against a concurrent
517 * irq_sysfs_init() as well.
518 */
519 irq_sysfs_del(desc);
520 delete_irq_desc(irq);
521
522 /*
523 * We free the descriptor, masks and stat fields via RCU. That
524 * allows demultiplex interrupts to do rcu based management of
525 * the child interrupts.
526 * This also allows us to use rcu in kstat_irqs_usr().
527 */
528 call_rcu(&desc->rcu, delayed_free_desc);
529}
530
531static int alloc_descs(unsigned int start, unsigned int cnt, int node,
532 const struct irq_affinity_desc *affinity,
533 struct module *owner)
534{
535 struct irq_desc *desc;
536 int i;
537
538 /* Validate affinity mask(s) */
539 if (affinity) {
540 for (i = 0; i < cnt; i++) {
541 if (cpumask_empty(&affinity[i].mask))
542 return -EINVAL;
543 }
544 }
545
546 for (i = 0; i < cnt; i++) {
547 const struct cpumask *mask = NULL;
548 unsigned int flags = 0;
549
550 if (affinity) {
551 if (affinity->is_managed) {
552 flags = IRQD_AFFINITY_MANAGED |
553 IRQD_MANAGED_SHUTDOWN;
554 }
555 flags |= IRQD_AFFINITY_SET;
556 mask = &affinity->mask;
557 node = cpu_to_node(cpumask_first(mask));
558 affinity++;
559 }
560
561 desc = alloc_desc(start + i, node, flags, mask, owner);
562 if (!desc)
563 goto err;
564 irq_insert_desc(start + i, desc);
565 irq_sysfs_add(start + i, desc);
566 irq_add_debugfs_entry(start + i, desc);
567 }
568 return start;
569
570err:
571 for (i--; i >= 0; i--)
572 free_desc(start + i);
573 return -ENOMEM;
574}
575
576static int irq_expand_nr_irqs(unsigned int nr)
577{
578 if (nr > MAX_SPARSE_IRQS)
579 return -ENOMEM;
580 nr_irqs = nr;
581 return 0;
582}
583
584int __init early_irq_init(void)
585{
586 int i, initcnt, node = first_online_node;
587 struct irq_desc *desc;
588
589 init_irq_default_affinity();
590
591 /* Let arch update nr_irqs and return the nr of preallocated irqs */
592 initcnt = arch_probe_nr_irqs();
593 printk(KERN_INFO "NR_IRQS: %d, nr_irqs: %d, preallocated irqs: %d\n",
594 NR_IRQS, nr_irqs, initcnt);
595
596 if (WARN_ON(nr_irqs > MAX_SPARSE_IRQS))
597 nr_irqs = MAX_SPARSE_IRQS;
598
599 if (WARN_ON(initcnt > MAX_SPARSE_IRQS))
600 initcnt = MAX_SPARSE_IRQS;
601
602 if (initcnt > nr_irqs)
603 nr_irqs = initcnt;
604
605 for (i = 0; i < initcnt; i++) {
606 desc = alloc_desc(i, node, 0, NULL, NULL);
607 irq_insert_desc(i, desc);
608 }
609 return arch_early_irq_init();
610}
611
612#else /* !CONFIG_SPARSE_IRQ */
613
614struct irq_desc irq_desc[NR_IRQS] __cacheline_aligned_in_smp = {
615 [0 ... NR_IRQS-1] = {
616 .handle_irq = handle_bad_irq,
617 .depth = 1,
618 .lock = __RAW_SPIN_LOCK_UNLOCKED(irq_desc->lock),
619 }
620};
621
622int __init early_irq_init(void)
623{
624 int count, i, node = first_online_node;
625 int ret;
626
627 init_irq_default_affinity();
628
629 printk(KERN_INFO "NR_IRQS: %d\n", NR_IRQS);
630
631 count = ARRAY_SIZE(irq_desc);
632
633 for (i = 0; i < count; i++) {
634 ret = init_desc(irq_desc + i, i, node, 0, NULL, NULL);
635 if (unlikely(ret))
636 goto __free_desc_res;
637 }
638
639 return arch_early_irq_init();
640
641__free_desc_res:
642 while (--i >= 0) {
643 free_masks(irq_desc + i);
644 free_percpu(irq_desc[i].kstat_irqs);
645 }
646
647 return ret;
648}
649
650struct irq_desc *irq_to_desc(unsigned int irq)
651{
652 return (irq < NR_IRQS) ? irq_desc + irq : NULL;
653}
654EXPORT_SYMBOL(irq_to_desc);
655
656static void free_desc(unsigned int irq)
657{
658 struct irq_desc *desc = irq_to_desc(irq);
659 unsigned long flags;
660
661 raw_spin_lock_irqsave(&desc->lock, flags);
662 desc_set_defaults(irq, desc, irq_desc_get_node(desc), NULL, NULL);
663 raw_spin_unlock_irqrestore(&desc->lock, flags);
664 delete_irq_desc(irq);
665}
666
667static inline int alloc_descs(unsigned int start, unsigned int cnt, int node,
668 const struct irq_affinity_desc *affinity,
669 struct module *owner)
670{
671 u32 i;
672
673 for (i = 0; i < cnt; i++) {
674 struct irq_desc *desc = irq_to_desc(start + i);
675
676 desc->owner = owner;
677 irq_insert_desc(start + i, desc);
678 }
679 return start;
680}
681
682static int irq_expand_nr_irqs(unsigned int nr)
683{
684 return -ENOMEM;
685}
686
687void irq_mark_irq(unsigned int irq)
688{
689 mutex_lock(&sparse_irq_lock);
690 irq_insert_desc(irq, irq_desc + irq);
691 mutex_unlock(&sparse_irq_lock);
692}
693
694#ifdef CONFIG_GENERIC_IRQ_LEGACY
695void irq_init_desc(unsigned int irq)
696{
697 free_desc(irq);
698}
699#endif
700
701#endif /* !CONFIG_SPARSE_IRQ */
702
703int handle_irq_desc(struct irq_desc *desc)
704{
705 struct irq_data *data;
706
707 if (!desc)
708 return -EINVAL;
709
710 data = irq_desc_get_irq_data(desc);
711 if (WARN_ON_ONCE(!in_hardirq() && handle_enforce_irqctx(data)))
712 return -EPERM;
713
714 generic_handle_irq_desc(desc);
715 return 0;
716}
717
718/**
719 * generic_handle_irq - Invoke the handler for a particular irq
720 * @irq: The irq number to handle
721 *
722 * Returns: 0 on success, or -EINVAL if conversion has failed
723 *
724 * This function must be called from an IRQ context with irq regs
725 * initialized.
726 */
727int generic_handle_irq(unsigned int irq)
728{
729 return handle_irq_desc(irq_to_desc(irq));
730}
731EXPORT_SYMBOL_GPL(generic_handle_irq);
732
733/**
734 * generic_handle_irq_safe - Invoke the handler for a particular irq from any
735 * context.
736 * @irq: The irq number to handle
737 *
738 * Returns: 0 on success, a negative value on error.
739 *
740 * This function can be called from any context (IRQ or process context). It
741 * will report an error if not invoked from IRQ context and the irq has been
742 * marked to enforce IRQ-context only.
743 */
744int generic_handle_irq_safe(unsigned int irq)
745{
746 unsigned long flags;
747 int ret;
748
749 local_irq_save(flags);
750 ret = handle_irq_desc(irq_to_desc(irq));
751 local_irq_restore(flags);
752 return ret;
753}
754EXPORT_SYMBOL_GPL(generic_handle_irq_safe);
755
756#ifdef CONFIG_IRQ_DOMAIN
757/**
758 * generic_handle_domain_irq - Invoke the handler for a HW irq belonging
759 * to a domain.
760 * @domain: The domain where to perform the lookup
761 * @hwirq: The HW irq number to convert to a logical one
762 *
763 * Returns: 0 on success, or -EINVAL if conversion has failed
764 *
765 * This function must be called from an IRQ context with irq regs
766 * initialized.
767 */
768int generic_handle_domain_irq(struct irq_domain *domain, unsigned int hwirq)
769{
770 return handle_irq_desc(irq_resolve_mapping(domain, hwirq));
771}
772EXPORT_SYMBOL_GPL(generic_handle_domain_irq);
773
774 /**
775 * generic_handle_irq_safe - Invoke the handler for a HW irq belonging
776 * to a domain from any context.
777 * @domain: The domain where to perform the lookup
778 * @hwirq: The HW irq number to convert to a logical one
779 *
780 * Returns: 0 on success, a negative value on error.
781 *
782 * This function can be called from any context (IRQ or process
783 * context). If the interrupt is marked as 'enforce IRQ-context only' then
784 * the function must be invoked from hard interrupt context.
785 */
786int generic_handle_domain_irq_safe(struct irq_domain *domain, unsigned int hwirq)
787{
788 unsigned long flags;
789 int ret;
790
791 local_irq_save(flags);
792 ret = handle_irq_desc(irq_resolve_mapping(domain, hwirq));
793 local_irq_restore(flags);
794 return ret;
795}
796EXPORT_SYMBOL_GPL(generic_handle_domain_irq_safe);
797
798/**
799 * generic_handle_domain_nmi - Invoke the handler for a HW nmi belonging
800 * to a domain.
801 * @domain: The domain where to perform the lookup
802 * @hwirq: The HW irq number to convert to a logical one
803 *
804 * Returns: 0 on success, or -EINVAL if conversion has failed
805 *
806 * This function must be called from an NMI context with irq regs
807 * initialized.
808 **/
809int generic_handle_domain_nmi(struct irq_domain *domain, unsigned int hwirq)
810{
811 WARN_ON_ONCE(!in_nmi());
812 return handle_irq_desc(irq_resolve_mapping(domain, hwirq));
813}
814#endif
815
816/* Dynamic interrupt handling */
817
818/**
819 * irq_free_descs - free irq descriptors
820 * @from: Start of descriptor range
821 * @cnt: Number of consecutive irqs to free
822 */
823void irq_free_descs(unsigned int from, unsigned int cnt)
824{
825 int i;
826
827 if (from >= nr_irqs || (from + cnt) > nr_irqs)
828 return;
829
830 mutex_lock(&sparse_irq_lock);
831 for (i = 0; i < cnt; i++)
832 free_desc(from + i);
833
834 mutex_unlock(&sparse_irq_lock);
835}
836EXPORT_SYMBOL_GPL(irq_free_descs);
837
838/**
839 * __irq_alloc_descs - allocate and initialize a range of irq descriptors
840 * @irq: Allocate for specific irq number if irq >= 0
841 * @from: Start the search from this irq number
842 * @cnt: Number of consecutive irqs to allocate.
843 * @node: Preferred node on which the irq descriptor should be allocated
844 * @owner: Owning module (can be NULL)
845 * @affinity: Optional pointer to an affinity mask array of size @cnt which
846 * hints where the irq descriptors should be allocated and which
847 * default affinities to use
848 *
849 * Returns the first irq number or error code
850 */
851int __ref
852__irq_alloc_descs(int irq, unsigned int from, unsigned int cnt, int node,
853 struct module *owner, const struct irq_affinity_desc *affinity)
854{
855 int start, ret;
856
857 if (!cnt)
858 return -EINVAL;
859
860 if (irq >= 0) {
861 if (from > irq)
862 return -EINVAL;
863 from = irq;
864 } else {
865 /*
866 * For interrupts which are freely allocated the
867 * architecture can force a lower bound to the @from
868 * argument. x86 uses this to exclude the GSI space.
869 */
870 from = arch_dynirq_lower_bound(from);
871 }
872
873 mutex_lock(&sparse_irq_lock);
874
875 start = irq_find_free_area(from, cnt);
876 ret = -EEXIST;
877 if (irq >=0 && start != irq)
878 goto unlock;
879
880 if (start + cnt > nr_irqs) {
881 ret = irq_expand_nr_irqs(start + cnt);
882 if (ret)
883 goto unlock;
884 }
885 ret = alloc_descs(start, cnt, node, affinity, owner);
886unlock:
887 mutex_unlock(&sparse_irq_lock);
888 return ret;
889}
890EXPORT_SYMBOL_GPL(__irq_alloc_descs);
891
892/**
893 * irq_get_next_irq - get next allocated irq number
894 * @offset: where to start the search
895 *
896 * Returns next irq number after offset or nr_irqs if none is found.
897 */
898unsigned int irq_get_next_irq(unsigned int offset)
899{
900 return irq_find_at_or_after(offset);
901}
902
903struct irq_desc *
904__irq_get_desc_lock(unsigned int irq, unsigned long *flags, bool bus,
905 unsigned int check)
906{
907 struct irq_desc *desc = irq_to_desc(irq);
908
909 if (desc) {
910 if (check & _IRQ_DESC_CHECK) {
911 if ((check & _IRQ_DESC_PERCPU) &&
912 !irq_settings_is_per_cpu_devid(desc))
913 return NULL;
914
915 if (!(check & _IRQ_DESC_PERCPU) &&
916 irq_settings_is_per_cpu_devid(desc))
917 return NULL;
918 }
919
920 if (bus)
921 chip_bus_lock(desc);
922 raw_spin_lock_irqsave(&desc->lock, *flags);
923 }
924 return desc;
925}
926
927void __irq_put_desc_unlock(struct irq_desc *desc, unsigned long flags, bool bus)
928 __releases(&desc->lock)
929{
930 raw_spin_unlock_irqrestore(&desc->lock, flags);
931 if (bus)
932 chip_bus_sync_unlock(desc);
933}
934
935int irq_set_percpu_devid_partition(unsigned int irq,
936 const struct cpumask *affinity)
937{
938 struct irq_desc *desc = irq_to_desc(irq);
939
940 if (!desc || desc->percpu_enabled)
941 return -EINVAL;
942
943 desc->percpu_enabled = kzalloc(sizeof(*desc->percpu_enabled), GFP_KERNEL);
944
945 if (!desc->percpu_enabled)
946 return -ENOMEM;
947
948 desc->percpu_affinity = affinity ? : cpu_possible_mask;
949
950 irq_set_percpu_devid_flags(irq);
951 return 0;
952}
953
954int irq_set_percpu_devid(unsigned int irq)
955{
956 return irq_set_percpu_devid_partition(irq, NULL);
957}
958
959int irq_get_percpu_devid_partition(unsigned int irq, struct cpumask *affinity)
960{
961 struct irq_desc *desc = irq_to_desc(irq);
962
963 if (!desc || !desc->percpu_enabled)
964 return -EINVAL;
965
966 if (affinity)
967 cpumask_copy(affinity, desc->percpu_affinity);
968
969 return 0;
970}
971EXPORT_SYMBOL_GPL(irq_get_percpu_devid_partition);
972
973void kstat_incr_irq_this_cpu(unsigned int irq)
974{
975 kstat_incr_irqs_this_cpu(irq_to_desc(irq));
976}
977
978/**
979 * kstat_irqs_cpu - Get the statistics for an interrupt on a cpu
980 * @irq: The interrupt number
981 * @cpu: The cpu number
982 *
983 * Returns the sum of interrupt counts on @cpu since boot for
984 * @irq. The caller must ensure that the interrupt is not removed
985 * concurrently.
986 */
987unsigned int kstat_irqs_cpu(unsigned int irq, int cpu)
988{
989 struct irq_desc *desc = irq_to_desc(irq);
990
991 return desc && desc->kstat_irqs ? per_cpu(desc->kstat_irqs->cnt, cpu) : 0;
992}
993
994unsigned int kstat_irqs_desc(struct irq_desc *desc, const struct cpumask *cpumask)
995{
996 unsigned int sum = 0;
997 int cpu;
998
999 if (!irq_settings_is_per_cpu_devid(desc) &&
1000 !irq_settings_is_per_cpu(desc) &&
1001 !irq_is_nmi(desc))
1002 return data_race(desc->tot_count);
1003
1004 for_each_cpu(cpu, cpumask)
1005 sum += data_race(per_cpu(desc->kstat_irqs->cnt, cpu));
1006 return sum;
1007}
1008
1009static unsigned int kstat_irqs(unsigned int irq)
1010{
1011 struct irq_desc *desc = irq_to_desc(irq);
1012
1013 if (!desc || !desc->kstat_irqs)
1014 return 0;
1015 return kstat_irqs_desc(desc, cpu_possible_mask);
1016}
1017
1018#ifdef CONFIG_GENERIC_IRQ_STAT_SNAPSHOT
1019
1020void kstat_snapshot_irqs(void)
1021{
1022 struct irq_desc *desc;
1023 unsigned int irq;
1024
1025 for_each_irq_desc(irq, desc) {
1026 if (!desc->kstat_irqs)
1027 continue;
1028 this_cpu_write(desc->kstat_irqs->ref, this_cpu_read(desc->kstat_irqs->cnt));
1029 }
1030}
1031
1032unsigned int kstat_get_irq_since_snapshot(unsigned int irq)
1033{
1034 struct irq_desc *desc = irq_to_desc(irq);
1035
1036 if (!desc || !desc->kstat_irqs)
1037 return 0;
1038 return this_cpu_read(desc->kstat_irqs->cnt) - this_cpu_read(desc->kstat_irqs->ref);
1039}
1040
1041#endif
1042
1043/**
1044 * kstat_irqs_usr - Get the statistics for an interrupt from thread context
1045 * @irq: The interrupt number
1046 *
1047 * Returns the sum of interrupt counts on all cpus since boot for @irq.
1048 *
1049 * It uses rcu to protect the access since a concurrent removal of an
1050 * interrupt descriptor is observing an rcu grace period before
1051 * delayed_free_desc()/irq_kobj_release().
1052 */
1053unsigned int kstat_irqs_usr(unsigned int irq)
1054{
1055 unsigned int sum;
1056
1057 rcu_read_lock();
1058 sum = kstat_irqs(irq);
1059 rcu_read_unlock();
1060 return sum;
1061}
1062
1063#ifdef CONFIG_LOCKDEP
1064void __irq_set_lockdep_class(unsigned int irq, struct lock_class_key *lock_class,
1065 struct lock_class_key *request_class)
1066{
1067 struct irq_desc *desc = irq_to_desc(irq);
1068
1069 if (desc) {
1070 lockdep_set_class(&desc->lock, lock_class);
1071 lockdep_set_class(&desc->request_mutex, request_class);
1072 }
1073}
1074EXPORT_SYMBOL_GPL(__irq_set_lockdep_class);
1075#endif