1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * Xen event channels
4 *
5 * Xen models interrupts with abstract event channels.  Because each
6 * domain gets 1024 event channels, but NR_IRQ is not that large, we
7 * must dynamically map irqs<->event channels.  The event channels
8 * interface with the rest of the kernel by defining a xen interrupt
9 * chip.  When an event is received, it is mapped to an irq and sent
10 * through the normal interrupt processing path.
11 *
12 * There are four kinds of events which can be mapped to an event
13 * channel:
14 *
15 * 1. Inter-domain notifications.  This includes all the virtual
16 *    device events, since they're driven by front-ends in another domain
17 *    (typically dom0).
18 * 2. VIRQs, typically used for timers.  These are per-cpu events.
19 * 3. IPIs.
20 * 4. PIRQs - Hardware interrupts.
21 *
22 * Jeremy Fitzhardinge <jeremy@xensource.com>, XenSource Inc, 2007
23 */
24
25#define pr_fmt(fmt) "xen:" KBUILD_MODNAME ": " fmt
26
27#include <linux/linkage.h>
28#include <linux/interrupt.h>
29#include <linux/irq.h>
30#include <linux/moduleparam.h>
31#include <linux/string.h>
32#include <linux/memblock.h>
33#include <linux/slab.h>
34#include <linux/irqnr.h>
35#include <linux/pci.h>
36#include <linux/rcupdate.h>
37#include <linux/spinlock.h>
38#include <linux/cpuhotplug.h>
39#include <linux/atomic.h>
40#include <linux/ktime.h>
41
42#ifdef CONFIG_X86
43#include <asm/desc.h>
44#include <asm/ptrace.h>
45#include <asm/idtentry.h>
46#include <asm/irq.h>
47#include <asm/io_apic.h>
48#include <asm/i8259.h>
49#include <asm/xen/pci.h>
50#endif
51#include <asm/sync_bitops.h>
52#include <asm/xen/hypercall.h>
53#include <asm/xen/hypervisor.h>
54#include <xen/page.h>
55
56#include <xen/xen.h>
57#include <xen/hvm.h>
58#include <xen/xen-ops.h>
59#include <xen/events.h>
60#include <xen/interface/xen.h>
61#include <xen/interface/event_channel.h>
62#include <xen/interface/hvm/hvm_op.h>
63#include <xen/interface/hvm/params.h>
64#include <xen/interface/physdev.h>
65#include <xen/interface/sched.h>
66#include <xen/interface/vcpu.h>
67#include <asm/hw_irq.h>
68
69#include "events_internal.h"
70
71#undef MODULE_PARAM_PREFIX
72#define MODULE_PARAM_PREFIX "xen."
73
74/* Interrupt types. */
75enum xen_irq_type {
76	IRQT_UNBOUND = 0,
77	IRQT_PIRQ,
78	IRQT_VIRQ,
79	IRQT_IPI,
80	IRQT_EVTCHN
81};
82
83/*
84 * Packed IRQ information:
85 * type - enum xen_irq_type
86 * event channel - irq->event channel mapping
87 * cpu - cpu this event channel is bound to
88 * index - type-specific information:
89 *    PIRQ - vector, with MSB being "needs EIO", or physical IRQ of the HVM
90 *           guest, or GSI (real passthrough IRQ) of the device.
91 *    VIRQ - virq number
92 *    IPI - IPI vector
93 *    EVTCHN -
94 */
95struct irq_info {
96	struct list_head list;
97	struct list_head eoi_list;
98	struct rcu_work rwork;
99	short refcnt;
100	short spurious_cnt;
101	short type;             /* type */
102	u8 mask_reason;         /* Why is event channel masked */
103#define EVT_MASK_REASON_EXPLICIT	0x01
104#define EVT_MASK_REASON_TEMPORARY	0x02
105#define EVT_MASK_REASON_EOI_PENDING	0x04
106	u8 is_active;		/* Is event just being handled? */
107	unsigned irq;
108	evtchn_port_t evtchn;   /* event channel */
109	unsigned short cpu;     /* cpu bound */
110	unsigned short eoi_cpu; /* EOI must happen on this cpu-1 */
111	unsigned int irq_epoch; /* If eoi_cpu valid: irq_epoch of event */
112	u64 eoi_time;           /* Time in jiffies when to EOI. */
113	raw_spinlock_t lock;
114
115	union {
116		unsigned short virq;
117		enum ipi_vector ipi;
118		struct {
119			unsigned short pirq;
120			unsigned short gsi;
121			unsigned char vector;
122			unsigned char flags;
123			uint16_t domid;
124		} pirq;
125	} u;
126};
127
128#define PIRQ_NEEDS_EOI	(1 << 0)
129#define PIRQ_SHAREABLE	(1 << 1)
130#define PIRQ_MSI_GROUP	(1 << 2)
131
132static uint __read_mostly event_loop_timeout = 2;
133module_param(event_loop_timeout, uint, 0644);
134
135static uint __read_mostly event_eoi_delay = 10;
136module_param(event_eoi_delay, uint, 0644);
137
138const struct evtchn_ops *evtchn_ops;
139
140/*
141 * This lock protects updates to the following mapping and reference-count
142 * arrays. The lock does not need to be acquired to read the mapping tables.
143 */
144static DEFINE_MUTEX(irq_mapping_update_lock);
145
146/*
147 * Lock hierarchy:
148 *
149 * irq_mapping_update_lock
150 *   IRQ-desc lock
151 *     percpu eoi_list_lock
152 *       irq_info->lock
153 */
154
155static LIST_HEAD(xen_irq_list_head);
156
157/* IRQ <-> VIRQ mapping. */
158static DEFINE_PER_CPU(int [NR_VIRQS], virq_to_irq) = {[0 ... NR_VIRQS-1] = -1};
159
160/* IRQ <-> IPI mapping */
161static DEFINE_PER_CPU(int [XEN_NR_IPIS], ipi_to_irq) = {[0 ... XEN_NR_IPIS-1] = -1};
162
163static int **evtchn_to_irq;
164#ifdef CONFIG_X86
165static unsigned long *pirq_eoi_map;
166#endif
167static bool (*pirq_needs_eoi)(unsigned irq);
168
169#define EVTCHN_ROW(e)  (e / (PAGE_SIZE/sizeof(**evtchn_to_irq)))
170#define EVTCHN_COL(e)  (e % (PAGE_SIZE/sizeof(**evtchn_to_irq)))
171#define EVTCHN_PER_ROW (PAGE_SIZE / sizeof(**evtchn_to_irq))
172
173/* Xen will never allocate port zero for any purpose. */
174#define VALID_EVTCHN(chn)	((chn) != 0)
175
176static struct irq_info *legacy_info_ptrs[NR_IRQS_LEGACY];
177
178static struct irq_chip xen_dynamic_chip;
179static struct irq_chip xen_lateeoi_chip;
180static struct irq_chip xen_percpu_chip;
181static struct irq_chip xen_pirq_chip;
182static void enable_dynirq(struct irq_data *data);
183static void disable_dynirq(struct irq_data *data);
184
185static DEFINE_PER_CPU(unsigned int, irq_epoch);
186
187static void clear_evtchn_to_irq_row(int *evtchn_row)
188{
189	unsigned col;
190
191	for (col = 0; col < EVTCHN_PER_ROW; col++)
192		WRITE_ONCE(evtchn_row[col], -1);
193}
194
195static void clear_evtchn_to_irq_all(void)
196{
197	unsigned row;
198
199	for (row = 0; row < EVTCHN_ROW(xen_evtchn_max_channels()); row++) {
200		if (evtchn_to_irq[row] == NULL)
201			continue;
202		clear_evtchn_to_irq_row(evtchn_to_irq[row]);
203	}
204}
205
206static int set_evtchn_to_irq(evtchn_port_t evtchn, unsigned int irq)
207{
208	unsigned row;
209	unsigned col;
210	int *evtchn_row;
211
212	if (evtchn >= xen_evtchn_max_channels())
213		return -EINVAL;
214
215	row = EVTCHN_ROW(evtchn);
216	col = EVTCHN_COL(evtchn);
217
218	if (evtchn_to_irq[row] == NULL) {
219		/* Unallocated irq entries return -1 anyway */
220		if (irq == -1)
221			return 0;
222
223		evtchn_row = (int *) __get_free_pages(GFP_KERNEL, 0);
224		if (evtchn_row == NULL)
225			return -ENOMEM;
226
227		clear_evtchn_to_irq_row(evtchn_row);
228
229		/*
230		 * We've prepared an empty row for the mapping. If a different
231		 * thread was faster inserting it, we can drop ours.
232		 */
233		if (cmpxchg(&evtchn_to_irq[row], NULL, evtchn_row) != NULL)
234			free_page((unsigned long) evtchn_row);
235	}
236
237	WRITE_ONCE(evtchn_to_irq[row][col], irq);
238	return 0;
239}
240
241int get_evtchn_to_irq(evtchn_port_t evtchn)
242{
243	if (evtchn >= xen_evtchn_max_channels())
244		return -1;
245	if (evtchn_to_irq[EVTCHN_ROW(evtchn)] == NULL)
246		return -1;
247	return READ_ONCE(evtchn_to_irq[EVTCHN_ROW(evtchn)][EVTCHN_COL(evtchn)]);
248}
249
250/* Get info for IRQ */
251static struct irq_info *info_for_irq(unsigned irq)
252{
253	if (irq < nr_legacy_irqs())
254		return legacy_info_ptrs[irq];
255	else
256		return irq_get_chip_data(irq);
257}
258
259static void set_info_for_irq(unsigned int irq, struct irq_info *info)
260{
261	if (irq < nr_legacy_irqs())
262		legacy_info_ptrs[irq] = info;
263	else
264		irq_set_chip_data(irq, info);
265}
266
267static void delayed_free_irq(struct work_struct *work)
268{
269	struct irq_info *info = container_of(to_rcu_work(work), struct irq_info,
270					     rwork);
271	unsigned int irq = info->irq;
272
273	/* Remove the info pointer only now, with no potential users left. */
274	set_info_for_irq(irq, NULL);
275
276	kfree(info);
277
278	/* Legacy IRQ descriptors are managed by the arch. */
279	if (irq >= nr_legacy_irqs())
280		irq_free_desc(irq);
281}
282
283/* Constructors for packed IRQ information. */
284static int xen_irq_info_common_setup(struct irq_info *info,
285				     unsigned irq,
286				     enum xen_irq_type type,
287				     evtchn_port_t evtchn,
288				     unsigned short cpu)
289{
290	int ret;
291
292	BUG_ON(info->type != IRQT_UNBOUND && info->type != type);
293
294	info->type = type;
295	info->irq = irq;
296	info->evtchn = evtchn;
297	info->cpu = cpu;
298	info->mask_reason = EVT_MASK_REASON_EXPLICIT;
299	raw_spin_lock_init(&info->lock);
300
301	ret = set_evtchn_to_irq(evtchn, irq);
302	if (ret < 0)
303		return ret;
304
305	irq_clear_status_flags(irq, IRQ_NOREQUEST|IRQ_NOAUTOEN);
306
307	return xen_evtchn_port_setup(evtchn);
308}
309
310static int xen_irq_info_evtchn_setup(unsigned irq,
311				     evtchn_port_t evtchn)
312{
313	struct irq_info *info = info_for_irq(irq);
314
315	return xen_irq_info_common_setup(info, irq, IRQT_EVTCHN, evtchn, 0);
316}
317
318static int xen_irq_info_ipi_setup(unsigned cpu,
319				  unsigned irq,
320				  evtchn_port_t evtchn,
321				  enum ipi_vector ipi)
322{
323	struct irq_info *info = info_for_irq(irq);
324
325	info->u.ipi = ipi;
326
327	per_cpu(ipi_to_irq, cpu)[ipi] = irq;
328
329	return xen_irq_info_common_setup(info, irq, IRQT_IPI, evtchn, 0);
330}
331
332static int xen_irq_info_virq_setup(unsigned cpu,
333				   unsigned irq,
334				   evtchn_port_t evtchn,
335				   unsigned virq)
336{
337	struct irq_info *info = info_for_irq(irq);
338
339	info->u.virq = virq;
340
341	per_cpu(virq_to_irq, cpu)[virq] = irq;
342
343	return xen_irq_info_common_setup(info, irq, IRQT_VIRQ, evtchn, 0);
344}
345
346static int xen_irq_info_pirq_setup(unsigned irq,
347				   evtchn_port_t evtchn,
348				   unsigned pirq,
349				   unsigned gsi,
350				   uint16_t domid,
351				   unsigned char flags)
352{
353	struct irq_info *info = info_for_irq(irq);
354
355	info->u.pirq.pirq = pirq;
356	info->u.pirq.gsi = gsi;
357	info->u.pirq.domid = domid;
358	info->u.pirq.flags = flags;
359
360	return xen_irq_info_common_setup(info, irq, IRQT_PIRQ, evtchn, 0);
361}
362
363static void xen_irq_info_cleanup(struct irq_info *info)
364{
365	set_evtchn_to_irq(info->evtchn, -1);
366	xen_evtchn_port_remove(info->evtchn, info->cpu);
367	info->evtchn = 0;
368}
369
370/*
371 * Accessors for packed IRQ information.
372 */
373evtchn_port_t evtchn_from_irq(unsigned irq)
374{
375	const struct irq_info *info = NULL;
376
377	if (likely(irq < nr_irqs))
378		info = info_for_irq(irq);
379	if (!info)
380		return 0;
381
382	return info->evtchn;
383}
384
385unsigned int irq_from_evtchn(evtchn_port_t evtchn)
386{
387	return get_evtchn_to_irq(evtchn);
388}
389EXPORT_SYMBOL_GPL(irq_from_evtchn);
390
391int irq_from_virq(unsigned int cpu, unsigned int virq)
392{
393	return per_cpu(virq_to_irq, cpu)[virq];
394}
395
396static enum ipi_vector ipi_from_irq(unsigned irq)
397{
398	struct irq_info *info = info_for_irq(irq);
399
400	BUG_ON(info == NULL);
401	BUG_ON(info->type != IRQT_IPI);
402
403	return info->u.ipi;
404}
405
406static unsigned virq_from_irq(unsigned irq)
407{
408	struct irq_info *info = info_for_irq(irq);
409
410	BUG_ON(info == NULL);
411	BUG_ON(info->type != IRQT_VIRQ);
412
413	return info->u.virq;
414}
415
416static unsigned pirq_from_irq(unsigned irq)
417{
418	struct irq_info *info = info_for_irq(irq);
419
420	BUG_ON(info == NULL);
421	BUG_ON(info->type != IRQT_PIRQ);
422
423	return info->u.pirq.pirq;
424}
425
426static enum xen_irq_type type_from_irq(unsigned irq)
427{
428	return info_for_irq(irq)->type;
429}
430
431static unsigned cpu_from_irq(unsigned irq)
432{
433	return info_for_irq(irq)->cpu;
434}
435
436unsigned int cpu_from_evtchn(evtchn_port_t evtchn)
437{
438	int irq = get_evtchn_to_irq(evtchn);
439	unsigned ret = 0;
440
441	if (irq != -1)
442		ret = cpu_from_irq(irq);
443
444	return ret;
445}
446
447static void do_mask(struct irq_info *info, u8 reason)
448{
449	unsigned long flags;
450
451	raw_spin_lock_irqsave(&info->lock, flags);
452
453	if (!info->mask_reason)
454		mask_evtchn(info->evtchn);
455
456	info->mask_reason |= reason;
457
458	raw_spin_unlock_irqrestore(&info->lock, flags);
459}
460
461static void do_unmask(struct irq_info *info, u8 reason)
462{
463	unsigned long flags;
464
465	raw_spin_lock_irqsave(&info->lock, flags);
466
467	info->mask_reason &= ~reason;
468
469	if (!info->mask_reason)
470		unmask_evtchn(info->evtchn);
471
472	raw_spin_unlock_irqrestore(&info->lock, flags);
473}
474
475#ifdef CONFIG_X86
476static bool pirq_check_eoi_map(unsigned irq)
477{
478	return test_bit(pirq_from_irq(irq), pirq_eoi_map);
479}
480#endif
481
482static bool pirq_needs_eoi_flag(unsigned irq)
483{
484	struct irq_info *info = info_for_irq(irq);
485	BUG_ON(info->type != IRQT_PIRQ);
486
487	return info->u.pirq.flags & PIRQ_NEEDS_EOI;
488}
489
490static void bind_evtchn_to_cpu(evtchn_port_t evtchn, unsigned int cpu)
491{
492	int irq = get_evtchn_to_irq(evtchn);
493	struct irq_info *info = info_for_irq(irq);
494
495	BUG_ON(irq == -1);
496#ifdef CONFIG_SMP
497	cpumask_copy(irq_get_affinity_mask(irq), cpumask_of(cpu));
498#endif
499	xen_evtchn_port_bind_to_cpu(evtchn, cpu, info->cpu);
500
501	info->cpu = cpu;
502}
503
504/**
505 * notify_remote_via_irq - send event to remote end of event channel via irq
506 * @irq: irq of event channel to send event to
507 *
508 * Unlike notify_remote_via_evtchn(), this is safe to use across
509 * save/restore. Notifications on a broken connection are silently
510 * dropped.
511 */
512void notify_remote_via_irq(int irq)
513{
514	evtchn_port_t evtchn = evtchn_from_irq(irq);
515
516	if (VALID_EVTCHN(evtchn))
517		notify_remote_via_evtchn(evtchn);
518}
519EXPORT_SYMBOL_GPL(notify_remote_via_irq);
520
521struct lateeoi_work {
522	struct delayed_work delayed;
523	spinlock_t eoi_list_lock;
524	struct list_head eoi_list;
525};
526
527static DEFINE_PER_CPU(struct lateeoi_work, lateeoi);
528
529static void lateeoi_list_del(struct irq_info *info)
530{
531	struct lateeoi_work *eoi = &per_cpu(lateeoi, info->eoi_cpu);
532	unsigned long flags;
533
534	spin_lock_irqsave(&eoi->eoi_list_lock, flags);
535	list_del_init(&info->eoi_list);
536	spin_unlock_irqrestore(&eoi->eoi_list_lock, flags);
537}
538
539static void lateeoi_list_add(struct irq_info *info)
540{
541	struct lateeoi_work *eoi = &per_cpu(lateeoi, info->eoi_cpu);
542	struct irq_info *elem;
543	u64 now = get_jiffies_64();
544	unsigned long delay;
545	unsigned long flags;
546
547	if (now < info->eoi_time)
548		delay = info->eoi_time - now;
549	else
550		delay = 1;
551
552	spin_lock_irqsave(&eoi->eoi_list_lock, flags);
553
554	elem = list_first_entry_or_null(&eoi->eoi_list, struct irq_info,
555					eoi_list);
556	if (!elem || info->eoi_time < elem->eoi_time) {
557		list_add(&info->eoi_list, &eoi->eoi_list);
558		mod_delayed_work_on(info->eoi_cpu, system_wq,
559				    &eoi->delayed, delay);
560	} else {
561		list_for_each_entry_reverse(elem, &eoi->eoi_list, eoi_list) {
562			if (elem->eoi_time <= info->eoi_time)
563				break;
564		}
565		list_add(&info->eoi_list, &elem->eoi_list);
566	}
567
568	spin_unlock_irqrestore(&eoi->eoi_list_lock, flags);
569}
570
571static void xen_irq_lateeoi_locked(struct irq_info *info, bool spurious)
572{
573	evtchn_port_t evtchn;
574	unsigned int cpu;
575	unsigned int delay = 0;
576
577	evtchn = info->evtchn;
578	if (!VALID_EVTCHN(evtchn) || !list_empty(&info->eoi_list))
579		return;
580
581	if (spurious) {
582		if ((1 << info->spurious_cnt) < (HZ << 2))
583			info->spurious_cnt++;
584		if (info->spurious_cnt > 1) {
585			delay = 1 << (info->spurious_cnt - 2);
586			if (delay > HZ)
587				delay = HZ;
588			if (!info->eoi_time)
589				info->eoi_cpu = smp_processor_id();
590			info->eoi_time = get_jiffies_64() + delay;
591		}
592	} else {
593		info->spurious_cnt = 0;
594	}
595
596	cpu = info->eoi_cpu;
597	if (info->eoi_time &&
598	    (info->irq_epoch == per_cpu(irq_epoch, cpu) || delay)) {
599		lateeoi_list_add(info);
600		return;
601	}
602
603	info->eoi_time = 0;
604
605	/* is_active hasn't been reset yet, do it now. */
606	smp_store_release(&info->is_active, 0);
607	do_unmask(info, EVT_MASK_REASON_EOI_PENDING);
608}
609
610static void xen_irq_lateeoi_worker(struct work_struct *work)
611{
612	struct lateeoi_work *eoi;
613	struct irq_info *info;
614	u64 now = get_jiffies_64();
615	unsigned long flags;
616
617	eoi = container_of(to_delayed_work(work), struct lateeoi_work, delayed);
618
619	rcu_read_lock();
620
621	while (true) {
622		spin_lock_irqsave(&eoi->eoi_list_lock, flags);
623
624		info = list_first_entry_or_null(&eoi->eoi_list, struct irq_info,
625						eoi_list);
626
627		if (info == NULL)
628			break;
629
630		if (now < info->eoi_time) {
631			mod_delayed_work_on(info->eoi_cpu, system_wq,
632					    &eoi->delayed,
633					    info->eoi_time - now);
634			break;
635		}
636
637		list_del_init(&info->eoi_list);
638
639		spin_unlock_irqrestore(&eoi->eoi_list_lock, flags);
640
641		info->eoi_time = 0;
642
643		xen_irq_lateeoi_locked(info, false);
644	}
645
646	spin_unlock_irqrestore(&eoi->eoi_list_lock, flags);
647
648	rcu_read_unlock();
649}
650
651static void xen_cpu_init_eoi(unsigned int cpu)
652{
653	struct lateeoi_work *eoi = &per_cpu(lateeoi, cpu);
654
655	INIT_DELAYED_WORK(&eoi->delayed, xen_irq_lateeoi_worker);
656	spin_lock_init(&eoi->eoi_list_lock);
657	INIT_LIST_HEAD(&eoi->eoi_list);
658}
659
660void xen_irq_lateeoi(unsigned int irq, unsigned int eoi_flags)
661{
662	struct irq_info *info;
663
664	rcu_read_lock();
665
666	info = info_for_irq(irq);
667
668	if (info)
669		xen_irq_lateeoi_locked(info, eoi_flags & XEN_EOI_FLAG_SPURIOUS);
670
671	rcu_read_unlock();
672}
673EXPORT_SYMBOL_GPL(xen_irq_lateeoi);
674
675static void xen_irq_init(unsigned irq)
676{
677	struct irq_info *info;
678
679#ifdef CONFIG_SMP
680	/* By default all event channels notify CPU#0. */
681	cpumask_copy(irq_get_affinity_mask(irq), cpumask_of(0));
682#endif
683
684	info = kzalloc(sizeof(*info), GFP_KERNEL);
685	if (info == NULL)
686		panic("Unable to allocate metadata for IRQ%d\n", irq);
687
688	info->type = IRQT_UNBOUND;
689	info->refcnt = -1;
690	INIT_RCU_WORK(&info->rwork, delayed_free_irq);
691
692	set_info_for_irq(irq, info);
693
694	INIT_LIST_HEAD(&info->eoi_list);
695	list_add_tail(&info->list, &xen_irq_list_head);
696}
697
698static int __must_check xen_allocate_irqs_dynamic(int nvec)
699{
700	int i, irq = irq_alloc_descs(-1, 0, nvec, -1);
701
702	if (irq >= 0) {
703		for (i = 0; i < nvec; i++)
704			xen_irq_init(irq + i);
705	}
706
707	return irq;
708}
709
710static inline int __must_check xen_allocate_irq_dynamic(void)
711{
712
713	return xen_allocate_irqs_dynamic(1);
714}
715
716static int __must_check xen_allocate_irq_gsi(unsigned gsi)
717{
718	int irq;
719
720	/*
721	 * A PV guest has no concept of a GSI (since it has no ACPI
722	 * nor access to/knowledge of the physical APICs). Therefore
723	 * all IRQs are dynamically allocated from the entire IRQ
724	 * space.
725	 */
726	if (xen_pv_domain() && !xen_initial_domain())
727		return xen_allocate_irq_dynamic();
728
729	/* Legacy IRQ descriptors are already allocated by the arch. */
730	if (gsi < nr_legacy_irqs())
731		irq = gsi;
732	else
733		irq = irq_alloc_desc_at(gsi, -1);
734
735	xen_irq_init(irq);
736
737	return irq;
738}
739
740static void xen_free_irq(unsigned irq)
741{
742	struct irq_info *info = info_for_irq(irq);
743
744	if (WARN_ON(!info))
745		return;
746
747	if (!list_empty(&info->eoi_list))
748		lateeoi_list_del(info);
749
750	list_del(&info->list);
751
752	WARN_ON(info->refcnt > 0);
753
754	queue_rcu_work(system_wq, &info->rwork);
755}
756
757static void xen_evtchn_close(evtchn_port_t port)
758{
759	struct evtchn_close close;
760
761	close.port = port;
762	if (HYPERVISOR_event_channel_op(EVTCHNOP_close, &close) != 0)
763		BUG();
764}
765
766static void event_handler_exit(struct irq_info *info)
767{
768	smp_store_release(&info->is_active, 0);
769	clear_evtchn(info->evtchn);
770}
771
772static void pirq_query_unmask(int irq)
773{
774	struct physdev_irq_status_query irq_status;
775	struct irq_info *info = info_for_irq(irq);
776
777	BUG_ON(info->type != IRQT_PIRQ);
778
779	irq_status.irq = pirq_from_irq(irq);
780	if (HYPERVISOR_physdev_op(PHYSDEVOP_irq_status_query, &irq_status))
781		irq_status.flags = 0;
782
783	info->u.pirq.flags &= ~PIRQ_NEEDS_EOI;
784	if (irq_status.flags & XENIRQSTAT_needs_eoi)
785		info->u.pirq.flags |= PIRQ_NEEDS_EOI;
786}
787
788static void eoi_pirq(struct irq_data *data)
789{
790	struct irq_info *info = info_for_irq(data->irq);
791	evtchn_port_t evtchn = info ? info->evtchn : 0;
792	struct physdev_eoi eoi = { .irq = pirq_from_irq(data->irq) };
793	int rc = 0;
794
795	if (!VALID_EVTCHN(evtchn))
796		return;
797
798	if (unlikely(irqd_is_setaffinity_pending(data)) &&
799	    likely(!irqd_irq_disabled(data))) {
800		do_mask(info, EVT_MASK_REASON_TEMPORARY);
801
802		event_handler_exit(info);
803
804		irq_move_masked_irq(data);
805
806		do_unmask(info, EVT_MASK_REASON_TEMPORARY);
807	} else
808		event_handler_exit(info);
809
810	if (pirq_needs_eoi(data->irq)) {
811		rc = HYPERVISOR_physdev_op(PHYSDEVOP_eoi, &eoi);
812		WARN_ON(rc);
813	}
814}
815
816static void mask_ack_pirq(struct irq_data *data)
817{
818	disable_dynirq(data);
819	eoi_pirq(data);
820}
821
822static unsigned int __startup_pirq(unsigned int irq)
823{
824	struct evtchn_bind_pirq bind_pirq;
825	struct irq_info *info = info_for_irq(irq);
826	evtchn_port_t evtchn = evtchn_from_irq(irq);
827	int rc;
828
829	BUG_ON(info->type != IRQT_PIRQ);
830
831	if (VALID_EVTCHN(evtchn))
832		goto out;
833
834	bind_pirq.pirq = pirq_from_irq(irq);
835	/* NB. We are happy to share unless we are probing. */
836	bind_pirq.flags = info->u.pirq.flags & PIRQ_SHAREABLE ?
837					BIND_PIRQ__WILL_SHARE : 0;
838	rc = HYPERVISOR_event_channel_op(EVTCHNOP_bind_pirq, &bind_pirq);
839	if (rc != 0) {
840		pr_warn("Failed to obtain physical IRQ %d\n", irq);
841		return 0;
842	}
843	evtchn = bind_pirq.port;
844
845	pirq_query_unmask(irq);
846
847	rc = set_evtchn_to_irq(evtchn, irq);
848	if (rc)
849		goto err;
850
851	info->evtchn = evtchn;
852	bind_evtchn_to_cpu(evtchn, 0);
853
854	rc = xen_evtchn_port_setup(evtchn);
855	if (rc)
856		goto err;
857
858out:
859	do_unmask(info, EVT_MASK_REASON_EXPLICIT);
860
861	eoi_pirq(irq_get_irq_data(irq));
862
863	return 0;
864
865err:
866	pr_err("irq%d: Failed to set port to irq mapping (%d)\n", irq, rc);
867	xen_evtchn_close(evtchn);
868	return 0;
869}
870
871static unsigned int startup_pirq(struct irq_data *data)
872{
873	return __startup_pirq(data->irq);
874}
875
876static void shutdown_pirq(struct irq_data *data)
877{
878	unsigned int irq = data->irq;
879	struct irq_info *info = info_for_irq(irq);
880	evtchn_port_t evtchn = evtchn_from_irq(irq);
881
882	BUG_ON(info->type != IRQT_PIRQ);
883
884	if (!VALID_EVTCHN(evtchn))
885		return;
886
887	do_mask(info, EVT_MASK_REASON_EXPLICIT);
888	xen_evtchn_close(evtchn);
889	xen_irq_info_cleanup(info);
890}
891
892static void enable_pirq(struct irq_data *data)
893{
894	enable_dynirq(data);
895}
896
897static void disable_pirq(struct irq_data *data)
898{
899	disable_dynirq(data);
900}
901
902int xen_irq_from_gsi(unsigned gsi)
903{
904	struct irq_info *info;
905
906	list_for_each_entry(info, &xen_irq_list_head, list) {
907		if (info->type != IRQT_PIRQ)
908			continue;
909
910		if (info->u.pirq.gsi == gsi)
911			return info->irq;
912	}
913
914	return -1;
915}
916EXPORT_SYMBOL_GPL(xen_irq_from_gsi);
917
918static void __unbind_from_irq(unsigned int irq)
919{
920	evtchn_port_t evtchn = evtchn_from_irq(irq);
921	struct irq_info *info = info_for_irq(irq);
922
923	if (info->refcnt > 0) {
924		info->refcnt--;
925		if (info->refcnt != 0)
926			return;
927	}
928
929	if (VALID_EVTCHN(evtchn)) {
930		unsigned int cpu = cpu_from_irq(irq);
931
932		xen_evtchn_close(evtchn);
933
934		switch (type_from_irq(irq)) {
935		case IRQT_VIRQ:
936			per_cpu(virq_to_irq, cpu)[virq_from_irq(irq)] = -1;
937			break;
938		case IRQT_IPI:
939			per_cpu(ipi_to_irq, cpu)[ipi_from_irq(irq)] = -1;
940			break;
941		default:
942			break;
943		}
944
945		xen_irq_info_cleanup(info);
946	}
947
948	xen_free_irq(irq);
949}
950
951/*
952 * Do not make any assumptions regarding the relationship between the
953 * IRQ number returned here and the Xen pirq argument.
954 *
955 * Note: We don't assign an event channel until the irq actually started
956 * up.  Return an existing irq if we've already got one for the gsi.
957 *
958 * Shareable implies level triggered, not shareable implies edge
959 * triggered here.
960 */
961int xen_bind_pirq_gsi_to_irq(unsigned gsi,
962			     unsigned pirq, int shareable, char *name)
963{
964	int irq = -1;
965	struct physdev_irq irq_op;
966	int ret;
967
968	mutex_lock(&irq_mapping_update_lock);
969
970	irq = xen_irq_from_gsi(gsi);
971	if (irq != -1) {
972		pr_info("%s: returning irq %d for gsi %u\n",
973			__func__, irq, gsi);
974		goto out;
975	}
976
977	irq = xen_allocate_irq_gsi(gsi);
978	if (irq < 0)
979		goto out;
980
981	irq_op.irq = irq;
982	irq_op.vector = 0;
983
984	/* Only the privileged domain can do this. For non-priv, the pcifront
985	 * driver provides a PCI bus that does the call to do exactly
986	 * this in the priv domain. */
987	if (xen_initial_domain() &&
988	    HYPERVISOR_physdev_op(PHYSDEVOP_alloc_irq_vector, &irq_op)) {
989		xen_free_irq(irq);
990		irq = -ENOSPC;
991		goto out;
992	}
993
994	ret = xen_irq_info_pirq_setup(irq, 0, pirq, gsi, DOMID_SELF,
995			       shareable ? PIRQ_SHAREABLE : 0);
996	if (ret < 0) {
997		__unbind_from_irq(irq);
998		irq = ret;
999		goto out;
1000	}
1001
1002	pirq_query_unmask(irq);
1003	/* We try to use the handler with the appropriate semantic for the
1004	 * type of interrupt: if the interrupt is an edge triggered
1005	 * interrupt we use handle_edge_irq.
1006	 *
1007	 * On the other hand if the interrupt is level triggered we use
1008	 * handle_fasteoi_irq like the native code does for this kind of
1009	 * interrupts.
1010	 *
1011	 * Depending on the Xen version, pirq_needs_eoi might return true
1012	 * not only for level triggered interrupts but for edge triggered
1013	 * interrupts too. In any case Xen always honors the eoi mechanism,
1014	 * not injecting any more pirqs of the same kind if the first one
1015	 * hasn't received an eoi yet. Therefore using the fasteoi handler
1016	 * is the right choice either way.
1017	 */
1018	if (shareable)
1019		irq_set_chip_and_handler_name(irq, &xen_pirq_chip,
1020				handle_fasteoi_irq, name);
1021	else
1022		irq_set_chip_and_handler_name(irq, &xen_pirq_chip,
1023				handle_edge_irq, name);
1024
1025out:
1026	mutex_unlock(&irq_mapping_update_lock);
1027
1028	return irq;
1029}
1030
1031#ifdef CONFIG_PCI_MSI
1032int xen_allocate_pirq_msi(struct pci_dev *dev, struct msi_desc *msidesc)
1033{
1034	int rc;
1035	struct physdev_get_free_pirq op_get_free_pirq;
1036
1037	op_get_free_pirq.type = MAP_PIRQ_TYPE_MSI;
1038	rc = HYPERVISOR_physdev_op(PHYSDEVOP_get_free_pirq, &op_get_free_pirq);
1039
1040	WARN_ONCE(rc == -ENOSYS,
1041		  "hypervisor does not support the PHYSDEVOP_get_free_pirq interface\n");
1042
1043	return rc ? -1 : op_get_free_pirq.pirq;
1044}
1045
1046int xen_bind_pirq_msi_to_irq(struct pci_dev *dev, struct msi_desc *msidesc,
1047			     int pirq, int nvec, const char *name, domid_t domid)
1048{
1049	int i, irq, ret;
1050
1051	mutex_lock(&irq_mapping_update_lock);
1052
1053	irq = xen_allocate_irqs_dynamic(nvec);
1054	if (irq < 0)
1055		goto out;
1056
1057	for (i = 0; i < nvec; i++) {
1058		irq_set_chip_and_handler_name(irq + i, &xen_pirq_chip, handle_edge_irq, name);
1059
1060		ret = xen_irq_info_pirq_setup(irq + i, 0, pirq + i, 0, domid,
1061					      i == 0 ? 0 : PIRQ_MSI_GROUP);
1062		if (ret < 0)
1063			goto error_irq;
1064	}
1065
1066	ret = irq_set_msi_desc(irq, msidesc);
1067	if (ret < 0)
1068		goto error_irq;
1069out:
1070	mutex_unlock(&irq_mapping_update_lock);
1071	return irq;
1072error_irq:
1073	while (nvec--)
1074		__unbind_from_irq(irq + nvec);
1075	mutex_unlock(&irq_mapping_update_lock);
1076	return ret;
1077}
1078#endif
1079
1080int xen_destroy_irq(int irq)
1081{
1082	struct physdev_unmap_pirq unmap_irq;
1083	struct irq_info *info = info_for_irq(irq);
1084	int rc = -ENOENT;
1085
1086	mutex_lock(&irq_mapping_update_lock);
1087
1088	/*
1089	 * If trying to remove a vector in a MSI group different
1090	 * than the first one skip the PIRQ unmap unless this vector
1091	 * is the first one in the group.
1092	 */
1093	if (xen_initial_domain() && !(info->u.pirq.flags & PIRQ_MSI_GROUP)) {
1094		unmap_irq.pirq = info->u.pirq.pirq;
1095		unmap_irq.domid = info->u.pirq.domid;
1096		rc = HYPERVISOR_physdev_op(PHYSDEVOP_unmap_pirq, &unmap_irq);
1097		/* If another domain quits without making the pci_disable_msix
1098		 * call, the Xen hypervisor takes care of freeing the PIRQs
1099		 * (free_domain_pirqs).
1100		 */
1101		if ((rc == -ESRCH && info->u.pirq.domid != DOMID_SELF))
1102			pr_info("domain %d does not have %d anymore\n",
1103				info->u.pirq.domid, info->u.pirq.pirq);
1104		else if (rc) {
1105			pr_warn("unmap irq failed %d\n", rc);
1106			goto out;
1107		}
1108	}
1109
1110	xen_free_irq(irq);
1111
1112out:
1113	mutex_unlock(&irq_mapping_update_lock);
1114	return rc;
1115}
1116
1117int xen_irq_from_pirq(unsigned pirq)
1118{
1119	int irq;
1120
1121	struct irq_info *info;
1122
1123	mutex_lock(&irq_mapping_update_lock);
1124
1125	list_for_each_entry(info, &xen_irq_list_head, list) {
1126		if (info->type != IRQT_PIRQ)
1127			continue;
1128		irq = info->irq;
1129		if (info->u.pirq.pirq == pirq)
1130			goto out;
1131	}
1132	irq = -1;
1133out:
1134	mutex_unlock(&irq_mapping_update_lock);
1135
1136	return irq;
1137}
1138
1139
1140int xen_pirq_from_irq(unsigned irq)
1141{
1142	return pirq_from_irq(irq);
1143}
1144EXPORT_SYMBOL_GPL(xen_pirq_from_irq);
1145
1146static int bind_evtchn_to_irq_chip(evtchn_port_t evtchn, struct irq_chip *chip)
1147{
1148	int irq;
1149	int ret;
1150
1151	if (evtchn >= xen_evtchn_max_channels())
1152		return -ENOMEM;
1153
1154	mutex_lock(&irq_mapping_update_lock);
1155
1156	irq = get_evtchn_to_irq(evtchn);
1157
1158	if (irq == -1) {
1159		irq = xen_allocate_irq_dynamic();
1160		if (irq < 0)
1161			goto out;
1162
1163		irq_set_chip_and_handler_name(irq, chip,
1164					      handle_edge_irq, "event");
1165
1166		ret = xen_irq_info_evtchn_setup(irq, evtchn);
1167		if (ret < 0) {
1168			__unbind_from_irq(irq);
1169			irq = ret;
1170			goto out;
1171		}
1172		/* New interdomain events are bound to VCPU 0. */
1173		bind_evtchn_to_cpu(evtchn, 0);
1174	} else {
1175		struct irq_info *info = info_for_irq(irq);
1176		WARN_ON(info == NULL || info->type != IRQT_EVTCHN);
1177	}
1178
1179out:
1180	mutex_unlock(&irq_mapping_update_lock);
1181
1182	return irq;
1183}
1184
1185int bind_evtchn_to_irq(evtchn_port_t evtchn)
1186{
1187	return bind_evtchn_to_irq_chip(evtchn, &xen_dynamic_chip);
1188}
1189EXPORT_SYMBOL_GPL(bind_evtchn_to_irq);
1190
1191int bind_evtchn_to_irq_lateeoi(evtchn_port_t evtchn)
1192{
1193	return bind_evtchn_to_irq_chip(evtchn, &xen_lateeoi_chip);
1194}
1195EXPORT_SYMBOL_GPL(bind_evtchn_to_irq_lateeoi);
1196
1197static int bind_ipi_to_irq(unsigned int ipi, unsigned int cpu)
1198{
1199	struct evtchn_bind_ipi bind_ipi;
1200	evtchn_port_t evtchn;
1201	int ret, irq;
1202
1203	mutex_lock(&irq_mapping_update_lock);
1204
1205	irq = per_cpu(ipi_to_irq, cpu)[ipi];
1206
1207	if (irq == -1) {
1208		irq = xen_allocate_irq_dynamic();
1209		if (irq < 0)
1210			goto out;
1211
1212		irq_set_chip_and_handler_name(irq, &xen_percpu_chip,
1213					      handle_percpu_irq, "ipi");
1214
1215		bind_ipi.vcpu = xen_vcpu_nr(cpu);
1216		if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_ipi,
1217						&bind_ipi) != 0)
1218			BUG();
1219		evtchn = bind_ipi.port;
1220
1221		ret = xen_irq_info_ipi_setup(cpu, irq, evtchn, ipi);
1222		if (ret < 0) {
1223			__unbind_from_irq(irq);
1224			irq = ret;
1225			goto out;
1226		}
1227		bind_evtchn_to_cpu(evtchn, cpu);
1228	} else {
1229		struct irq_info *info = info_for_irq(irq);
1230		WARN_ON(info == NULL || info->type != IRQT_IPI);
1231	}
1232
1233 out:
1234	mutex_unlock(&irq_mapping_update_lock);
1235	return irq;
1236}
1237
1238static int bind_interdomain_evtchn_to_irq_chip(unsigned int remote_domain,
1239					       evtchn_port_t remote_port,
1240					       struct irq_chip *chip)
1241{
1242	struct evtchn_bind_interdomain bind_interdomain;
1243	int err;
1244
1245	bind_interdomain.remote_dom  = remote_domain;
1246	bind_interdomain.remote_port = remote_port;
1247
1248	err = HYPERVISOR_event_channel_op(EVTCHNOP_bind_interdomain,
1249					  &bind_interdomain);
1250
1251	return err ? : bind_evtchn_to_irq_chip(bind_interdomain.local_port,
1252					       chip);
1253}
1254
1255int bind_interdomain_evtchn_to_irq_lateeoi(unsigned int remote_domain,
1256					   evtchn_port_t remote_port)
1257{
1258	return bind_interdomain_evtchn_to_irq_chip(remote_domain, remote_port,
1259						   &xen_lateeoi_chip);
1260}
1261EXPORT_SYMBOL_GPL(bind_interdomain_evtchn_to_irq_lateeoi);
1262
1263static int find_virq(unsigned int virq, unsigned int cpu, evtchn_port_t *evtchn)
1264{
1265	struct evtchn_status status;
1266	evtchn_port_t port;
1267	int rc = -ENOENT;
1268
1269	memset(&status, 0, sizeof(status));
1270	for (port = 0; port < xen_evtchn_max_channels(); port++) {
1271		status.dom = DOMID_SELF;
1272		status.port = port;
1273		rc = HYPERVISOR_event_channel_op(EVTCHNOP_status, &status);
1274		if (rc < 0)
1275			continue;
1276		if (status.status != EVTCHNSTAT_virq)
1277			continue;
1278		if (status.u.virq == virq && status.vcpu == xen_vcpu_nr(cpu)) {
1279			*evtchn = port;
1280			break;
1281		}
1282	}
1283	return rc;
1284}
1285
1286/**
1287 * xen_evtchn_nr_channels - number of usable event channel ports
1288 *
1289 * This may be less than the maximum supported by the current
1290 * hypervisor ABI. Use xen_evtchn_max_channels() for the maximum
1291 * supported.
1292 */
1293unsigned xen_evtchn_nr_channels(void)
1294{
1295        return evtchn_ops->nr_channels();
1296}
1297EXPORT_SYMBOL_GPL(xen_evtchn_nr_channels);
1298
1299int bind_virq_to_irq(unsigned int virq, unsigned int cpu, bool percpu)
1300{
1301	struct evtchn_bind_virq bind_virq;
1302	evtchn_port_t evtchn = 0;
1303	int irq, ret;
1304
1305	mutex_lock(&irq_mapping_update_lock);
1306
1307	irq = per_cpu(virq_to_irq, cpu)[virq];
1308
1309	if (irq == -1) {
1310		irq = xen_allocate_irq_dynamic();
1311		if (irq < 0)
1312			goto out;
1313
1314		if (percpu)
1315			irq_set_chip_and_handler_name(irq, &xen_percpu_chip,
1316						      handle_percpu_irq, "virq");
1317		else
1318			irq_set_chip_and_handler_name(irq, &xen_dynamic_chip,
1319						      handle_edge_irq, "virq");
1320
1321		bind_virq.virq = virq;
1322		bind_virq.vcpu = xen_vcpu_nr(cpu);
1323		ret = HYPERVISOR_event_channel_op(EVTCHNOP_bind_virq,
1324						&bind_virq);
1325		if (ret == 0)
1326			evtchn = bind_virq.port;
1327		else {
1328			if (ret == -EEXIST)
1329				ret = find_virq(virq, cpu, &evtchn);
1330			BUG_ON(ret < 0);
1331		}
1332
1333		ret = xen_irq_info_virq_setup(cpu, irq, evtchn, virq);
1334		if (ret < 0) {
1335			__unbind_from_irq(irq);
1336			irq = ret;
1337			goto out;
1338		}
1339
1340		bind_evtchn_to_cpu(evtchn, cpu);
1341	} else {
1342		struct irq_info *info = info_for_irq(irq);
1343		WARN_ON(info == NULL || info->type != IRQT_VIRQ);
1344	}
1345
1346out:
1347	mutex_unlock(&irq_mapping_update_lock);
1348
1349	return irq;
1350}
1351
1352static void unbind_from_irq(unsigned int irq)
1353{
1354	mutex_lock(&irq_mapping_update_lock);
1355	__unbind_from_irq(irq);
1356	mutex_unlock(&irq_mapping_update_lock);
1357}
1358
1359static int bind_evtchn_to_irqhandler_chip(evtchn_port_t evtchn,
1360					  irq_handler_t handler,
1361					  unsigned long irqflags,
1362					  const char *devname, void *dev_id,
1363					  struct irq_chip *chip)
1364{
1365	int irq, retval;
1366
1367	irq = bind_evtchn_to_irq_chip(evtchn, chip);
1368	if (irq < 0)
1369		return irq;
1370	retval = request_irq(irq, handler, irqflags, devname, dev_id);
1371	if (retval != 0) {
1372		unbind_from_irq(irq);
1373		return retval;
1374	}
1375
1376	return irq;
1377}
1378
1379int bind_evtchn_to_irqhandler(evtchn_port_t evtchn,
1380			      irq_handler_t handler,
1381			      unsigned long irqflags,
1382			      const char *devname, void *dev_id)
1383{
1384	return bind_evtchn_to_irqhandler_chip(evtchn, handler, irqflags,
1385					      devname, dev_id,
1386					      &xen_dynamic_chip);
1387}
1388EXPORT_SYMBOL_GPL(bind_evtchn_to_irqhandler);
1389
1390int bind_evtchn_to_irqhandler_lateeoi(evtchn_port_t evtchn,
1391				      irq_handler_t handler,
1392				      unsigned long irqflags,
1393				      const char *devname, void *dev_id)
1394{
1395	return bind_evtchn_to_irqhandler_chip(evtchn, handler, irqflags,
1396					      devname, dev_id,
1397					      &xen_lateeoi_chip);
1398}
1399EXPORT_SYMBOL_GPL(bind_evtchn_to_irqhandler_lateeoi);
1400
1401static int bind_interdomain_evtchn_to_irqhandler_chip(
1402		unsigned int remote_domain, evtchn_port_t remote_port,
1403		irq_handler_t handler, unsigned long irqflags,
1404		const char *devname, void *dev_id, struct irq_chip *chip)
1405{
1406	int irq, retval;
1407
1408	irq = bind_interdomain_evtchn_to_irq_chip(remote_domain, remote_port,
1409						  chip);
1410	if (irq < 0)
1411		return irq;
1412
1413	retval = request_irq(irq, handler, irqflags, devname, dev_id);
1414	if (retval != 0) {
1415		unbind_from_irq(irq);
1416		return retval;
1417	}
1418
1419	return irq;
1420}
1421
1422int bind_interdomain_evtchn_to_irqhandler_lateeoi(unsigned int remote_domain,
1423						  evtchn_port_t remote_port,
1424						  irq_handler_t handler,
1425						  unsigned long irqflags,
1426						  const char *devname,
1427						  void *dev_id)
1428{
1429	return bind_interdomain_evtchn_to_irqhandler_chip(remote_domain,
1430				remote_port, handler, irqflags, devname,
1431				dev_id, &xen_lateeoi_chip);
1432}
1433EXPORT_SYMBOL_GPL(bind_interdomain_evtchn_to_irqhandler_lateeoi);
1434
1435int bind_virq_to_irqhandler(unsigned int virq, unsigned int cpu,
1436			    irq_handler_t handler,
1437			    unsigned long irqflags, const char *devname, void *dev_id)
1438{
1439	int irq, retval;
1440
1441	irq = bind_virq_to_irq(virq, cpu, irqflags & IRQF_PERCPU);
1442	if (irq < 0)
1443		return irq;
1444	retval = request_irq(irq, handler, irqflags, devname, dev_id);
1445	if (retval != 0) {
1446		unbind_from_irq(irq);
1447		return retval;
1448	}
1449
1450	return irq;
1451}
1452EXPORT_SYMBOL_GPL(bind_virq_to_irqhandler);
1453
1454int bind_ipi_to_irqhandler(enum ipi_vector ipi,
1455			   unsigned int cpu,
1456			   irq_handler_t handler,
1457			   unsigned long irqflags,
1458			   const char *devname,
1459			   void *dev_id)
1460{
1461	int irq, retval;
1462
1463	irq = bind_ipi_to_irq(ipi, cpu);
1464	if (irq < 0)
1465		return irq;
1466
1467	irqflags |= IRQF_NO_SUSPEND | IRQF_FORCE_RESUME | IRQF_EARLY_RESUME;
1468	retval = request_irq(irq, handler, irqflags, devname, dev_id);
1469	if (retval != 0) {
1470		unbind_from_irq(irq);
1471		return retval;
1472	}
1473
1474	return irq;
1475}
1476
1477void unbind_from_irqhandler(unsigned int irq, void *dev_id)
1478{
1479	struct irq_info *info = info_for_irq(irq);
1480
1481	if (WARN_ON(!info))
1482		return;
1483	free_irq(irq, dev_id);
1484	unbind_from_irq(irq);
1485}
1486EXPORT_SYMBOL_GPL(unbind_from_irqhandler);
1487
1488/**
1489 * xen_set_irq_priority() - set an event channel priority.
1490 * @irq:irq bound to an event channel.
1491 * @priority: priority between XEN_IRQ_PRIORITY_MAX and XEN_IRQ_PRIORITY_MIN.
1492 */
1493int xen_set_irq_priority(unsigned irq, unsigned priority)
1494{
1495	struct evtchn_set_priority set_priority;
1496
1497	set_priority.port = evtchn_from_irq(irq);
1498	set_priority.priority = priority;
1499
1500	return HYPERVISOR_event_channel_op(EVTCHNOP_set_priority,
1501					   &set_priority);
1502}
1503EXPORT_SYMBOL_GPL(xen_set_irq_priority);
1504
1505int evtchn_make_refcounted(evtchn_port_t evtchn)
1506{
1507	int irq = get_evtchn_to_irq(evtchn);
1508	struct irq_info *info;
1509
1510	if (irq == -1)
1511		return -ENOENT;
1512
1513	info = info_for_irq(irq);
1514
1515	if (!info)
1516		return -ENOENT;
1517
1518	WARN_ON(info->refcnt != -1);
1519
1520	info->refcnt = 1;
1521
1522	return 0;
1523}
1524EXPORT_SYMBOL_GPL(evtchn_make_refcounted);
1525
1526int evtchn_get(evtchn_port_t evtchn)
1527{
1528	int irq;
1529	struct irq_info *info;
1530	int err = -ENOENT;
1531
1532	if (evtchn >= xen_evtchn_max_channels())
1533		return -EINVAL;
1534
1535	mutex_lock(&irq_mapping_update_lock);
1536
1537	irq = get_evtchn_to_irq(evtchn);
1538	if (irq == -1)
1539		goto done;
1540
1541	info = info_for_irq(irq);
1542
1543	if (!info)
1544		goto done;
1545
1546	err = -EINVAL;
1547	if (info->refcnt <= 0 || info->refcnt == SHRT_MAX)
1548		goto done;
1549
1550	info->refcnt++;
1551	err = 0;
1552 done:
1553	mutex_unlock(&irq_mapping_update_lock);
1554
1555	return err;
1556}
1557EXPORT_SYMBOL_GPL(evtchn_get);
1558
1559void evtchn_put(evtchn_port_t evtchn)
1560{
1561	int irq = get_evtchn_to_irq(evtchn);
1562	if (WARN_ON(irq == -1))
1563		return;
1564	unbind_from_irq(irq);
1565}
1566EXPORT_SYMBOL_GPL(evtchn_put);
1567
1568void xen_send_IPI_one(unsigned int cpu, enum ipi_vector vector)
1569{
1570	int irq;
1571
1572#ifdef CONFIG_X86
1573	if (unlikely(vector == XEN_NMI_VECTOR)) {
1574		int rc =  HYPERVISOR_vcpu_op(VCPUOP_send_nmi, xen_vcpu_nr(cpu),
1575					     NULL);
1576		if (rc < 0)
1577			printk(KERN_WARNING "Sending nmi to CPU%d failed (rc:%d)\n", cpu, rc);
1578		return;
1579	}
1580#endif
1581	irq = per_cpu(ipi_to_irq, cpu)[vector];
1582	BUG_ON(irq < 0);
1583	notify_remote_via_irq(irq);
1584}
1585
1586struct evtchn_loop_ctrl {
1587	ktime_t timeout;
1588	unsigned count;
1589	bool defer_eoi;
1590};
1591
1592void handle_irq_for_port(evtchn_port_t port, struct evtchn_loop_ctrl *ctrl)
1593{
1594	int irq;
1595	struct irq_info *info;
1596
1597	irq = get_evtchn_to_irq(port);
1598	if (irq == -1)
1599		return;
1600
1601	/*
1602	 * Check for timeout every 256 events.
1603	 * We are setting the timeout value only after the first 256
1604	 * events in order to not hurt the common case of few loop
1605	 * iterations. The 256 is basically an arbitrary value.
1606	 *
1607	 * In case we are hitting the timeout we need to defer all further
1608	 * EOIs in order to ensure to leave the event handling loop rather
1609	 * sooner than later.
1610	 */
1611	if (!ctrl->defer_eoi && !(++ctrl->count & 0xff)) {
1612		ktime_t kt = ktime_get();
1613
1614		if (!ctrl->timeout) {
1615			kt = ktime_add_ms(kt,
1616					  jiffies_to_msecs(event_loop_timeout));
1617			ctrl->timeout = kt;
1618		} else if (kt > ctrl->timeout) {
1619			ctrl->defer_eoi = true;
1620		}
1621	}
1622
1623	info = info_for_irq(irq);
1624	if (xchg_acquire(&info->is_active, 1))
1625		return;
1626
1627	if (ctrl->defer_eoi) {
1628		info->eoi_cpu = smp_processor_id();
1629		info->irq_epoch = __this_cpu_read(irq_epoch);
1630		info->eoi_time = get_jiffies_64() + event_eoi_delay;
1631	}
1632
1633	generic_handle_irq(irq);
1634}
1635
1636static void __xen_evtchn_do_upcall(void)
1637{
1638	struct vcpu_info *vcpu_info = __this_cpu_read(xen_vcpu);
1639	int cpu = smp_processor_id();
1640	struct evtchn_loop_ctrl ctrl = { 0 };
1641
1642	/*
1643	 * When closing an event channel the associated IRQ must not be freed
1644	 * until all cpus have left the event handling loop. This is ensured
1645	 * by taking the rcu_read_lock() while handling events, as freeing of
1646	 * the IRQ is handled via queue_rcu_work() _after_ closing the event
1647	 * channel.
1648	 */
1649	rcu_read_lock();
1650
1651	do {
1652		vcpu_info->evtchn_upcall_pending = 0;
1653
1654		xen_evtchn_handle_events(cpu, &ctrl);
1655
1656		BUG_ON(!irqs_disabled());
1657
1658		virt_rmb(); /* Hypervisor can set upcall pending. */
1659
1660	} while (vcpu_info->evtchn_upcall_pending);
1661
1662	rcu_read_unlock();
1663
1664	/*
1665	 * Increment irq_epoch only now to defer EOIs only for
1666	 * xen_irq_lateeoi() invocations occurring from inside the loop
1667	 * above.
1668	 */
1669	__this_cpu_inc(irq_epoch);
1670}
1671
1672void xen_evtchn_do_upcall(struct pt_regs *regs)
1673{
1674	struct pt_regs *old_regs = set_irq_regs(regs);
1675
1676	irq_enter();
1677
1678	__xen_evtchn_do_upcall();
1679
1680	irq_exit();
1681	set_irq_regs(old_regs);
1682}
1683
1684void xen_hvm_evtchn_do_upcall(void)
1685{
1686	__xen_evtchn_do_upcall();
1687}
1688EXPORT_SYMBOL_GPL(xen_hvm_evtchn_do_upcall);
1689
1690/* Rebind a new event channel to an existing irq. */
1691void rebind_evtchn_irq(evtchn_port_t evtchn, int irq)
1692{
1693	struct irq_info *info = info_for_irq(irq);
1694
1695	if (WARN_ON(!info))
1696		return;
1697
1698	/* Make sure the irq is masked, since the new event channel
1699	   will also be masked. */
1700	disable_irq(irq);
1701
1702	mutex_lock(&irq_mapping_update_lock);
1703
1704	/* After resume the irq<->evtchn mappings are all cleared out */
1705	BUG_ON(get_evtchn_to_irq(evtchn) != -1);
1706	/* Expect irq to have been bound before,
1707	   so there should be a proper type */
1708	BUG_ON(info->type == IRQT_UNBOUND);
1709
1710	(void)xen_irq_info_evtchn_setup(irq, evtchn);
1711
1712	mutex_unlock(&irq_mapping_update_lock);
1713
1714        bind_evtchn_to_cpu(evtchn, info->cpu);
1715	/* This will be deferred until interrupt is processed */
1716	irq_set_affinity(irq, cpumask_of(info->cpu));
1717
1718	/* Unmask the event channel. */
1719	enable_irq(irq);
1720}
1721
1722/* Rebind an evtchn so that it gets delivered to a specific cpu */
1723static int xen_rebind_evtchn_to_cpu(struct irq_info *info, unsigned int tcpu)
1724{
1725	struct evtchn_bind_vcpu bind_vcpu;
1726	evtchn_port_t evtchn = info ? info->evtchn : 0;
1727
1728	if (!VALID_EVTCHN(evtchn))
1729		return -1;
1730
1731	if (!xen_support_evtchn_rebind())
1732		return -1;
1733
1734	/* Send future instances of this interrupt to other vcpu. */
1735	bind_vcpu.port = evtchn;
1736	bind_vcpu.vcpu = xen_vcpu_nr(tcpu);
1737
1738	/*
1739	 * Mask the event while changing the VCPU binding to prevent
1740	 * it being delivered on an unexpected VCPU.
1741	 */
1742	do_mask(info, EVT_MASK_REASON_TEMPORARY);
1743
1744	/*
1745	 * If this fails, it usually just indicates that we're dealing with a
1746	 * virq or IPI channel, which don't actually need to be rebound. Ignore
1747	 * it, but don't do the xenlinux-level rebind in that case.
1748	 */
1749	if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_vcpu, &bind_vcpu) >= 0)
1750		bind_evtchn_to_cpu(evtchn, tcpu);
1751
1752	do_unmask(info, EVT_MASK_REASON_TEMPORARY);
1753
1754	return 0;
1755}
1756
1757static int set_affinity_irq(struct irq_data *data, const struct cpumask *dest,
1758			    bool force)
1759{
1760	unsigned tcpu = cpumask_first_and(dest, cpu_online_mask);
1761	int ret = xen_rebind_evtchn_to_cpu(info_for_irq(data->irq), tcpu);
1762
1763	if (!ret)
1764		irq_data_update_effective_affinity(data, cpumask_of(tcpu));
1765
1766	return ret;
1767}
1768
1769/* To be called with desc->lock held. */
1770int xen_set_affinity_evtchn(struct irq_desc *desc, unsigned int tcpu)
1771{
1772	struct irq_data *d = irq_desc_get_irq_data(desc);
1773
1774	return set_affinity_irq(d, cpumask_of(tcpu), false);
1775}
1776EXPORT_SYMBOL_GPL(xen_set_affinity_evtchn);
1777
1778static void enable_dynirq(struct irq_data *data)
1779{
1780	struct irq_info *info = info_for_irq(data->irq);
1781	evtchn_port_t evtchn = info ? info->evtchn : 0;
1782
1783	if (VALID_EVTCHN(evtchn))
1784		do_unmask(info, EVT_MASK_REASON_EXPLICIT);
1785}
1786
1787static void disable_dynirq(struct irq_data *data)
1788{
1789	struct irq_info *info = info_for_irq(data->irq);
1790	evtchn_port_t evtchn = info ? info->evtchn : 0;
1791
1792	if (VALID_EVTCHN(evtchn))
1793		do_mask(info, EVT_MASK_REASON_EXPLICIT);
1794}
1795
1796static void ack_dynirq(struct irq_data *data)
1797{
1798	struct irq_info *info = info_for_irq(data->irq);
1799	evtchn_port_t evtchn = info ? info->evtchn : 0;
1800
1801	if (!VALID_EVTCHN(evtchn))
1802		return;
1803
1804	if (unlikely(irqd_is_setaffinity_pending(data)) &&
1805	    likely(!irqd_irq_disabled(data))) {
1806		do_mask(info, EVT_MASK_REASON_TEMPORARY);
1807
1808		event_handler_exit(info);
1809
1810		irq_move_masked_irq(data);
1811
1812		do_unmask(info, EVT_MASK_REASON_TEMPORARY);
1813	} else
1814		event_handler_exit(info);
1815}
1816
1817static void mask_ack_dynirq(struct irq_data *data)
1818{
1819	disable_dynirq(data);
1820	ack_dynirq(data);
1821}
1822
1823static void lateeoi_ack_dynirq(struct irq_data *data)
1824{
1825	struct irq_info *info = info_for_irq(data->irq);
1826	evtchn_port_t evtchn = info ? info->evtchn : 0;
1827
1828	if (!VALID_EVTCHN(evtchn))
1829		return;
1830
1831	do_mask(info, EVT_MASK_REASON_EOI_PENDING);
1832
1833	if (unlikely(irqd_is_setaffinity_pending(data)) &&
1834	    likely(!irqd_irq_disabled(data))) {
1835		do_mask(info, EVT_MASK_REASON_TEMPORARY);
1836
1837		clear_evtchn(evtchn);
1838
1839		irq_move_masked_irq(data);
1840
1841		do_unmask(info, EVT_MASK_REASON_TEMPORARY);
1842	} else
1843		clear_evtchn(evtchn);
1844}
1845
1846static void lateeoi_mask_ack_dynirq(struct irq_data *data)
1847{
1848	struct irq_info *info = info_for_irq(data->irq);
1849	evtchn_port_t evtchn = info ? info->evtchn : 0;
1850
1851	if (VALID_EVTCHN(evtchn)) {
1852		do_mask(info, EVT_MASK_REASON_EXPLICIT);
1853		ack_dynirq(data);
1854	}
1855}
1856
1857static int retrigger_dynirq(struct irq_data *data)
1858{
1859	struct irq_info *info = info_for_irq(data->irq);
1860	evtchn_port_t evtchn = info ? info->evtchn : 0;
1861
1862	if (!VALID_EVTCHN(evtchn))
1863		return 0;
1864
1865	do_mask(info, EVT_MASK_REASON_TEMPORARY);
1866	set_evtchn(evtchn);
1867	do_unmask(info, EVT_MASK_REASON_TEMPORARY);
1868
1869	return 1;
1870}
1871
1872static void restore_pirqs(void)
1873{
1874	int pirq, rc, irq, gsi;
1875	struct physdev_map_pirq map_irq;
1876	struct irq_info *info;
1877
1878	list_for_each_entry(info, &xen_irq_list_head, list) {
1879		if (info->type != IRQT_PIRQ)
1880			continue;
1881
1882		pirq = info->u.pirq.pirq;
1883		gsi = info->u.pirq.gsi;
1884		irq = info->irq;
1885
1886		/* save/restore of PT devices doesn't work, so at this point the
1887		 * only devices present are GSI based emulated devices */
1888		if (!gsi)
1889			continue;
1890
1891		map_irq.domid = DOMID_SELF;
1892		map_irq.type = MAP_PIRQ_TYPE_GSI;
1893		map_irq.index = gsi;
1894		map_irq.pirq = pirq;
1895
1896		rc = HYPERVISOR_physdev_op(PHYSDEVOP_map_pirq, &map_irq);
1897		if (rc) {
1898			pr_warn("xen map irq failed gsi=%d irq=%d pirq=%d rc=%d\n",
1899				gsi, irq, pirq, rc);
1900			xen_free_irq(irq);
1901			continue;
1902		}
1903
1904		printk(KERN_DEBUG "xen: --> irq=%d, pirq=%d\n", irq, map_irq.pirq);
1905
1906		__startup_pirq(irq);
1907	}
1908}
1909
1910static void restore_cpu_virqs(unsigned int cpu)
1911{
1912	struct evtchn_bind_virq bind_virq;
1913	evtchn_port_t evtchn;
1914	int virq, irq;
1915
1916	for (virq = 0; virq < NR_VIRQS; virq++) {
1917		if ((irq = per_cpu(virq_to_irq, cpu)[virq]) == -1)
1918			continue;
1919
1920		BUG_ON(virq_from_irq(irq) != virq);
1921
1922		/* Get a new binding from Xen. */
1923		bind_virq.virq = virq;
1924		bind_virq.vcpu = xen_vcpu_nr(cpu);
1925		if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_virq,
1926						&bind_virq) != 0)
1927			BUG();
1928		evtchn = bind_virq.port;
1929
1930		/* Record the new mapping. */
1931		(void)xen_irq_info_virq_setup(cpu, irq, evtchn, virq);
1932		bind_evtchn_to_cpu(evtchn, cpu);
1933	}
1934}
1935
1936static void restore_cpu_ipis(unsigned int cpu)
1937{
1938	struct evtchn_bind_ipi bind_ipi;
1939	evtchn_port_t evtchn;
1940	int ipi, irq;
1941
1942	for (ipi = 0; ipi < XEN_NR_IPIS; ipi++) {
1943		if ((irq = per_cpu(ipi_to_irq, cpu)[ipi]) == -1)
1944			continue;
1945
1946		BUG_ON(ipi_from_irq(irq) != ipi);
1947
1948		/* Get a new binding from Xen. */
1949		bind_ipi.vcpu = xen_vcpu_nr(cpu);
1950		if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_ipi,
1951						&bind_ipi) != 0)
1952			BUG();
1953		evtchn = bind_ipi.port;
1954
1955		/* Record the new mapping. */
1956		(void)xen_irq_info_ipi_setup(cpu, irq, evtchn, ipi);
1957		bind_evtchn_to_cpu(evtchn, cpu);
1958	}
1959}
1960
1961/* Clear an irq's pending state, in preparation for polling on it */
1962void xen_clear_irq_pending(int irq)
1963{
1964	struct irq_info *info = info_for_irq(irq);
1965	evtchn_port_t evtchn = info ? info->evtchn : 0;
1966
1967	if (VALID_EVTCHN(evtchn))
1968		event_handler_exit(info);
1969}
1970EXPORT_SYMBOL(xen_clear_irq_pending);
1971void xen_set_irq_pending(int irq)
1972{
1973	evtchn_port_t evtchn = evtchn_from_irq(irq);
1974
1975	if (VALID_EVTCHN(evtchn))
1976		set_evtchn(evtchn);
1977}
1978
1979bool xen_test_irq_pending(int irq)
1980{
1981	evtchn_port_t evtchn = evtchn_from_irq(irq);
1982	bool ret = false;
1983
1984	if (VALID_EVTCHN(evtchn))
1985		ret = test_evtchn(evtchn);
1986
1987	return ret;
1988}
1989
1990/* Poll waiting for an irq to become pending with timeout.  In the usual case,
1991 * the irq will be disabled so it won't deliver an interrupt. */
1992void xen_poll_irq_timeout(int irq, u64 timeout)
1993{
1994	evtchn_port_t evtchn = evtchn_from_irq(irq);
1995
1996	if (VALID_EVTCHN(evtchn)) {
1997		struct sched_poll poll;
1998
1999		poll.nr_ports = 1;
2000		poll.timeout = timeout;
2001		set_xen_guest_handle(poll.ports, &evtchn);
2002
2003		if (HYPERVISOR_sched_op(SCHEDOP_poll, &poll) != 0)
2004			BUG();
2005	}
2006}
2007EXPORT_SYMBOL(xen_poll_irq_timeout);
2008/* Poll waiting for an irq to become pending.  In the usual case, the
2009 * irq will be disabled so it won't deliver an interrupt. */
2010void xen_poll_irq(int irq)
2011{
2012	xen_poll_irq_timeout(irq, 0 /* no timeout */);
2013}
2014
2015/* Check whether the IRQ line is shared with other guests. */
2016int xen_test_irq_shared(int irq)
2017{
2018	struct irq_info *info = info_for_irq(irq);
2019	struct physdev_irq_status_query irq_status;
2020
2021	if (WARN_ON(!info))
2022		return -ENOENT;
2023
2024	irq_status.irq = info->u.pirq.pirq;
2025
2026	if (HYPERVISOR_physdev_op(PHYSDEVOP_irq_status_query, &irq_status))
2027		return 0;
2028	return !(irq_status.flags & XENIRQSTAT_shared);
2029}
2030EXPORT_SYMBOL_GPL(xen_test_irq_shared);
2031
2032void xen_irq_resume(void)
2033{
2034	unsigned int cpu;
2035	struct irq_info *info;
2036
2037	/* New event-channel space is not 'live' yet. */
2038	xen_evtchn_resume();
2039
2040	/* No IRQ <-> event-channel mappings. */
2041	list_for_each_entry(info, &xen_irq_list_head, list)
2042		info->evtchn = 0; /* zap event-channel binding */
2043
2044	clear_evtchn_to_irq_all();
2045
2046	for_each_possible_cpu(cpu) {
2047		restore_cpu_virqs(cpu);
2048		restore_cpu_ipis(cpu);
2049	}
2050
2051	restore_pirqs();
2052}
2053
2054static struct irq_chip xen_dynamic_chip __read_mostly = {
2055	.name			= "xen-dyn",
2056
2057	.irq_disable		= disable_dynirq,
2058	.irq_mask		= disable_dynirq,
2059	.irq_unmask		= enable_dynirq,
2060
2061	.irq_ack		= ack_dynirq,
2062	.irq_mask_ack		= mask_ack_dynirq,
2063
2064	.irq_set_affinity	= set_affinity_irq,
2065	.irq_retrigger		= retrigger_dynirq,
2066};
2067
2068static struct irq_chip xen_lateeoi_chip __read_mostly = {
2069	/* The chip name needs to contain "xen-dyn" for irqbalance to work. */
2070	.name			= "xen-dyn-lateeoi",
2071
2072	.irq_disable		= disable_dynirq,
2073	.irq_mask		= disable_dynirq,
2074	.irq_unmask		= enable_dynirq,
2075
2076	.irq_ack		= lateeoi_ack_dynirq,
2077	.irq_mask_ack		= lateeoi_mask_ack_dynirq,
2078
2079	.irq_set_affinity	= set_affinity_irq,
2080	.irq_retrigger		= retrigger_dynirq,
2081};
2082
2083static struct irq_chip xen_pirq_chip __read_mostly = {
2084	.name			= "xen-pirq",
2085
2086	.irq_startup		= startup_pirq,
2087	.irq_shutdown		= shutdown_pirq,
2088	.irq_enable		= enable_pirq,
2089	.irq_disable		= disable_pirq,
2090
2091	.irq_mask		= disable_dynirq,
2092	.irq_unmask		= enable_dynirq,
2093
2094	.irq_ack		= eoi_pirq,
2095	.irq_eoi		= eoi_pirq,
2096	.irq_mask_ack		= mask_ack_pirq,
2097
2098	.irq_set_affinity	= set_affinity_irq,
2099
2100	.irq_retrigger		= retrigger_dynirq,
2101};
2102
2103static struct irq_chip xen_percpu_chip __read_mostly = {
2104	.name			= "xen-percpu",
2105
2106	.irq_disable		= disable_dynirq,
2107	.irq_mask		= disable_dynirq,
2108	.irq_unmask		= enable_dynirq,
2109
2110	.irq_ack		= ack_dynirq,
2111};
2112
2113#ifdef CONFIG_XEN_PVHVM
2114/* Vector callbacks are better than PCI interrupts to receive event
2115 * channel notifications because we can receive vector callbacks on any
2116 * vcpu and we don't need PCI support or APIC interactions. */
2117void xen_setup_callback_vector(void)
2118{
2119	uint64_t callback_via;
2120
2121	if (xen_have_vector_callback) {
2122		callback_via = HVM_CALLBACK_VECTOR(HYPERVISOR_CALLBACK_VECTOR);
2123		if (xen_set_callback_via(callback_via)) {
2124			pr_err("Request for Xen HVM callback vector failed\n");
2125			xen_have_vector_callback = 0;
2126		}
2127	}
2128}
2129
2130static __init void xen_alloc_callback_vector(void)
2131{
2132	if (!xen_have_vector_callback)
2133		return;
2134
2135	pr_info("Xen HVM callback vector for event delivery is enabled\n");
2136	alloc_intr_gate(HYPERVISOR_CALLBACK_VECTOR, asm_sysvec_xen_hvm_callback);
2137}
2138#else
2139void xen_setup_callback_vector(void) {}
2140static inline void xen_alloc_callback_vector(void) {}
2141#endif
2142
2143bool xen_fifo_events = true;
2144module_param_named(fifo_events, xen_fifo_events, bool, 0);
2145
2146static int xen_evtchn_cpu_prepare(unsigned int cpu)
2147{
2148	int ret = 0;
2149
2150	xen_cpu_init_eoi(cpu);
2151
2152	if (evtchn_ops->percpu_init)
2153		ret = evtchn_ops->percpu_init(cpu);
2154
2155	return ret;
2156}
2157
2158static int xen_evtchn_cpu_dead(unsigned int cpu)
2159{
2160	int ret = 0;
2161
2162	if (evtchn_ops->percpu_deinit)
2163		ret = evtchn_ops->percpu_deinit(cpu);
2164
2165	return ret;
2166}
2167
2168void __init xen_init_IRQ(void)
2169{
2170	int ret = -EINVAL;
2171	evtchn_port_t evtchn;
2172
2173	if (xen_fifo_events)
2174		ret = xen_evtchn_fifo_init();
2175	if (ret < 0) {
2176		xen_evtchn_2l_init();
2177		xen_fifo_events = false;
2178	}
2179
2180	xen_cpu_init_eoi(smp_processor_id());
2181
2182	cpuhp_setup_state_nocalls(CPUHP_XEN_EVTCHN_PREPARE,
2183				  "xen/evtchn:prepare",
2184				  xen_evtchn_cpu_prepare, xen_evtchn_cpu_dead);
2185
2186	evtchn_to_irq = kcalloc(EVTCHN_ROW(xen_evtchn_max_channels()),
2187				sizeof(*evtchn_to_irq), GFP_KERNEL);
2188	BUG_ON(!evtchn_to_irq);
2189
2190	/* No event channels are 'live' right now. */
2191	for (evtchn = 0; evtchn < xen_evtchn_nr_channels(); evtchn++)
2192		mask_evtchn(evtchn);
2193
2194	pirq_needs_eoi = pirq_needs_eoi_flag;
2195
2196#ifdef CONFIG_X86
2197	if (xen_pv_domain()) {
2198		if (xen_initial_domain())
2199			pci_xen_initial_domain();
2200	}
2201	if (xen_feature(XENFEAT_hvm_callback_vector)) {
2202		xen_setup_callback_vector();
2203		xen_alloc_callback_vector();
2204	}
2205
2206	if (xen_hvm_domain()) {
2207		native_init_IRQ();
2208		/* pci_xen_hvm_init must be called after native_init_IRQ so that
2209		 * __acpi_register_gsi can point at the right function */
2210		pci_xen_hvm_init();
2211	} else {
2212		int rc;
2213		struct physdev_pirq_eoi_gmfn eoi_gmfn;
2214
2215		pirq_eoi_map = (void *)__get_free_page(GFP_KERNEL|__GFP_ZERO);
2216		eoi_gmfn.gmfn = virt_to_gfn(pirq_eoi_map);
2217		rc = HYPERVISOR_physdev_op(PHYSDEVOP_pirq_eoi_gmfn_v2, &eoi_gmfn);
2218		if (rc != 0) {
2219			free_page((unsigned long) pirq_eoi_map);
2220			pirq_eoi_map = NULL;
2221		} else
2222			pirq_needs_eoi = pirq_check_eoi_map;
2223	}
2224#endif
2225}
2226