xref: /linux/drivers/xen/events/events_base.c (revision ca55b2fef3a9373fcfc30f82fd26bc7fccbda732)
1 /*
2  * Xen event channels
3  *
4  * Xen models interrupts with abstract event channels.  Because each
5  * domain gets 1024 event channels, but NR_IRQ is not that large, we
6  * must dynamically map irqs<->event channels.  The event channels
7  * interface with the rest of the kernel by defining a xen interrupt
8  * chip.  When an event is received, it is mapped to an irq and sent
9  * through the normal interrupt processing path.
10  *
11  * There are four kinds of events which can be mapped to an event
12  * channel:
13  *
14  * 1. Inter-domain notifications.  This includes all the virtual
15  *    device events, since they're driven by front-ends in another domain
16  *    (typically dom0).
17  * 2. VIRQs, typically used for timers.  These are per-cpu events.
18  * 3. IPIs.
19  * 4. PIRQs - Hardware interrupts.
20  *
21  * Jeremy Fitzhardinge <jeremy@xensource.com>, XenSource Inc, 2007
22  */
23 
24 #define pr_fmt(fmt) "xen:" KBUILD_MODNAME ": " fmt
25 
26 #include <linux/linkage.h>
27 #include <linux/interrupt.h>
28 #include <linux/irq.h>
29 #include <linux/module.h>
30 #include <linux/string.h>
31 #include <linux/bootmem.h>
32 #include <linux/slab.h>
33 #include <linux/irqnr.h>
34 #include <linux/pci.h>
35 
36 #ifdef CONFIG_X86
37 #include <asm/desc.h>
38 #include <asm/ptrace.h>
39 #include <asm/irq.h>
40 #include <asm/idle.h>
41 #include <asm/io_apic.h>
42 #include <asm/xen/pci.h>
43 #include <xen/page.h>
44 #endif
45 #include <asm/sync_bitops.h>
46 #include <asm/xen/hypercall.h>
47 #include <asm/xen/hypervisor.h>
48 
49 #include <xen/xen.h>
50 #include <xen/hvm.h>
51 #include <xen/xen-ops.h>
52 #include <xen/events.h>
53 #include <xen/interface/xen.h>
54 #include <xen/interface/event_channel.h>
55 #include <xen/interface/hvm/hvm_op.h>
56 #include <xen/interface/hvm/params.h>
57 #include <xen/interface/physdev.h>
58 #include <xen/interface/sched.h>
59 #include <xen/interface/vcpu.h>
60 #include <asm/hw_irq.h>
61 
62 #include "events_internal.h"
63 
64 const struct evtchn_ops *evtchn_ops;
65 
66 /*
67  * This lock protects updates to the following mapping and reference-count
68  * arrays. The lock does not need to be acquired to read the mapping tables.
69  */
70 static DEFINE_MUTEX(irq_mapping_update_lock);
71 
72 static LIST_HEAD(xen_irq_list_head);
73 
74 /* IRQ <-> VIRQ mapping. */
75 static DEFINE_PER_CPU(int [NR_VIRQS], virq_to_irq) = {[0 ... NR_VIRQS-1] = -1};
76 
77 /* IRQ <-> IPI mapping */
78 static DEFINE_PER_CPU(int [XEN_NR_IPIS], ipi_to_irq) = {[0 ... XEN_NR_IPIS-1] = -1};
79 
80 int **evtchn_to_irq;
81 #ifdef CONFIG_X86
82 static unsigned long *pirq_eoi_map;
83 #endif
84 static bool (*pirq_needs_eoi)(unsigned irq);
85 
86 #define EVTCHN_ROW(e)  (e / (PAGE_SIZE/sizeof(**evtchn_to_irq)))
87 #define EVTCHN_COL(e)  (e % (PAGE_SIZE/sizeof(**evtchn_to_irq)))
88 #define EVTCHN_PER_ROW (PAGE_SIZE / sizeof(**evtchn_to_irq))
89 
90 /* Xen will never allocate port zero for any purpose. */
91 #define VALID_EVTCHN(chn)	((chn) != 0)
92 
93 static struct irq_chip xen_dynamic_chip;
94 static struct irq_chip xen_percpu_chip;
95 static struct irq_chip xen_pirq_chip;
96 static void enable_dynirq(struct irq_data *data);
97 static void disable_dynirq(struct irq_data *data);
98 
99 static void clear_evtchn_to_irq_row(unsigned row)
100 {
101 	unsigned col;
102 
103 	for (col = 0; col < EVTCHN_PER_ROW; col++)
104 		evtchn_to_irq[row][col] = -1;
105 }
106 
107 static void clear_evtchn_to_irq_all(void)
108 {
109 	unsigned row;
110 
111 	for (row = 0; row < EVTCHN_ROW(xen_evtchn_max_channels()); row++) {
112 		if (evtchn_to_irq[row] == NULL)
113 			continue;
114 		clear_evtchn_to_irq_row(row);
115 	}
116 }
117 
118 static int set_evtchn_to_irq(unsigned evtchn, unsigned irq)
119 {
120 	unsigned row;
121 	unsigned col;
122 
123 	if (evtchn >= xen_evtchn_max_channels())
124 		return -EINVAL;
125 
126 	row = EVTCHN_ROW(evtchn);
127 	col = EVTCHN_COL(evtchn);
128 
129 	if (evtchn_to_irq[row] == NULL) {
130 		/* Unallocated irq entries return -1 anyway */
131 		if (irq == -1)
132 			return 0;
133 
134 		evtchn_to_irq[row] = (int *)get_zeroed_page(GFP_KERNEL);
135 		if (evtchn_to_irq[row] == NULL)
136 			return -ENOMEM;
137 
138 		clear_evtchn_to_irq_row(row);
139 	}
140 
141 	evtchn_to_irq[EVTCHN_ROW(evtchn)][EVTCHN_COL(evtchn)] = irq;
142 	return 0;
143 }
144 
145 int get_evtchn_to_irq(unsigned evtchn)
146 {
147 	if (evtchn >= xen_evtchn_max_channels())
148 		return -1;
149 	if (evtchn_to_irq[EVTCHN_ROW(evtchn)] == NULL)
150 		return -1;
151 	return evtchn_to_irq[EVTCHN_ROW(evtchn)][EVTCHN_COL(evtchn)];
152 }
153 
154 /* Get info for IRQ */
155 struct irq_info *info_for_irq(unsigned irq)
156 {
157 	return irq_get_handler_data(irq);
158 }
159 
160 /* Constructors for packed IRQ information. */
161 static int xen_irq_info_common_setup(struct irq_info *info,
162 				     unsigned irq,
163 				     enum xen_irq_type type,
164 				     unsigned evtchn,
165 				     unsigned short cpu)
166 {
167 	int ret;
168 
169 	BUG_ON(info->type != IRQT_UNBOUND && info->type != type);
170 
171 	info->type = type;
172 	info->irq = irq;
173 	info->evtchn = evtchn;
174 	info->cpu = cpu;
175 
176 	ret = set_evtchn_to_irq(evtchn, irq);
177 	if (ret < 0)
178 		return ret;
179 
180 	irq_clear_status_flags(irq, IRQ_NOREQUEST|IRQ_NOAUTOEN);
181 
182 	return xen_evtchn_port_setup(info);
183 }
184 
185 static int xen_irq_info_evtchn_setup(unsigned irq,
186 				     unsigned evtchn)
187 {
188 	struct irq_info *info = info_for_irq(irq);
189 
190 	return xen_irq_info_common_setup(info, irq, IRQT_EVTCHN, evtchn, 0);
191 }
192 
193 static int xen_irq_info_ipi_setup(unsigned cpu,
194 				  unsigned irq,
195 				  unsigned evtchn,
196 				  enum ipi_vector ipi)
197 {
198 	struct irq_info *info = info_for_irq(irq);
199 
200 	info->u.ipi = ipi;
201 
202 	per_cpu(ipi_to_irq, cpu)[ipi] = irq;
203 
204 	return xen_irq_info_common_setup(info, irq, IRQT_IPI, evtchn, 0);
205 }
206 
207 static int xen_irq_info_virq_setup(unsigned cpu,
208 				   unsigned irq,
209 				   unsigned evtchn,
210 				   unsigned virq)
211 {
212 	struct irq_info *info = info_for_irq(irq);
213 
214 	info->u.virq = virq;
215 
216 	per_cpu(virq_to_irq, cpu)[virq] = irq;
217 
218 	return xen_irq_info_common_setup(info, irq, IRQT_VIRQ, evtchn, 0);
219 }
220 
221 static int xen_irq_info_pirq_setup(unsigned irq,
222 				   unsigned evtchn,
223 				   unsigned pirq,
224 				   unsigned gsi,
225 				   uint16_t domid,
226 				   unsigned char flags)
227 {
228 	struct irq_info *info = info_for_irq(irq);
229 
230 	info->u.pirq.pirq = pirq;
231 	info->u.pirq.gsi = gsi;
232 	info->u.pirq.domid = domid;
233 	info->u.pirq.flags = flags;
234 
235 	return xen_irq_info_common_setup(info, irq, IRQT_PIRQ, evtchn, 0);
236 }
237 
238 static void xen_irq_info_cleanup(struct irq_info *info)
239 {
240 	set_evtchn_to_irq(info->evtchn, -1);
241 	info->evtchn = 0;
242 }
243 
244 /*
245  * Accessors for packed IRQ information.
246  */
247 unsigned int evtchn_from_irq(unsigned irq)
248 {
249 	if (unlikely(WARN(irq >= nr_irqs, "Invalid irq %d!\n", irq)))
250 		return 0;
251 
252 	return info_for_irq(irq)->evtchn;
253 }
254 
255 unsigned irq_from_evtchn(unsigned int evtchn)
256 {
257 	return get_evtchn_to_irq(evtchn);
258 }
259 EXPORT_SYMBOL_GPL(irq_from_evtchn);
260 
261 int irq_from_virq(unsigned int cpu, unsigned int virq)
262 {
263 	return per_cpu(virq_to_irq, cpu)[virq];
264 }
265 
266 static enum ipi_vector ipi_from_irq(unsigned irq)
267 {
268 	struct irq_info *info = info_for_irq(irq);
269 
270 	BUG_ON(info == NULL);
271 	BUG_ON(info->type != IRQT_IPI);
272 
273 	return info->u.ipi;
274 }
275 
276 static unsigned virq_from_irq(unsigned irq)
277 {
278 	struct irq_info *info = info_for_irq(irq);
279 
280 	BUG_ON(info == NULL);
281 	BUG_ON(info->type != IRQT_VIRQ);
282 
283 	return info->u.virq;
284 }
285 
286 static unsigned pirq_from_irq(unsigned irq)
287 {
288 	struct irq_info *info = info_for_irq(irq);
289 
290 	BUG_ON(info == NULL);
291 	BUG_ON(info->type != IRQT_PIRQ);
292 
293 	return info->u.pirq.pirq;
294 }
295 
296 static enum xen_irq_type type_from_irq(unsigned irq)
297 {
298 	return info_for_irq(irq)->type;
299 }
300 
301 unsigned cpu_from_irq(unsigned irq)
302 {
303 	return info_for_irq(irq)->cpu;
304 }
305 
306 unsigned int cpu_from_evtchn(unsigned int evtchn)
307 {
308 	int irq = get_evtchn_to_irq(evtchn);
309 	unsigned ret = 0;
310 
311 	if (irq != -1)
312 		ret = cpu_from_irq(irq);
313 
314 	return ret;
315 }
316 
317 #ifdef CONFIG_X86
318 static bool pirq_check_eoi_map(unsigned irq)
319 {
320 	return test_bit(pirq_from_irq(irq), pirq_eoi_map);
321 }
322 #endif
323 
324 static bool pirq_needs_eoi_flag(unsigned irq)
325 {
326 	struct irq_info *info = info_for_irq(irq);
327 	BUG_ON(info->type != IRQT_PIRQ);
328 
329 	return info->u.pirq.flags & PIRQ_NEEDS_EOI;
330 }
331 
332 static void bind_evtchn_to_cpu(unsigned int chn, unsigned int cpu)
333 {
334 	int irq = get_evtchn_to_irq(chn);
335 	struct irq_info *info = info_for_irq(irq);
336 
337 	BUG_ON(irq == -1);
338 #ifdef CONFIG_SMP
339 	cpumask_copy(irq_get_affinity_mask(irq), cpumask_of(cpu));
340 #endif
341 	xen_evtchn_port_bind_to_cpu(info, cpu);
342 
343 	info->cpu = cpu;
344 }
345 
346 static void xen_evtchn_mask_all(void)
347 {
348 	unsigned int evtchn;
349 
350 	for (evtchn = 0; evtchn < xen_evtchn_nr_channels(); evtchn++)
351 		mask_evtchn(evtchn);
352 }
353 
354 /**
355  * notify_remote_via_irq - send event to remote end of event channel via irq
356  * @irq: irq of event channel to send event to
357  *
358  * Unlike notify_remote_via_evtchn(), this is safe to use across
359  * save/restore. Notifications on a broken connection are silently
360  * dropped.
361  */
362 void notify_remote_via_irq(int irq)
363 {
364 	int evtchn = evtchn_from_irq(irq);
365 
366 	if (VALID_EVTCHN(evtchn))
367 		notify_remote_via_evtchn(evtchn);
368 }
369 EXPORT_SYMBOL_GPL(notify_remote_via_irq);
370 
371 static void xen_irq_init(unsigned irq)
372 {
373 	struct irq_info *info;
374 #ifdef CONFIG_SMP
375 	/* By default all event channels notify CPU#0. */
376 	cpumask_copy(irq_get_affinity_mask(irq), cpumask_of(0));
377 #endif
378 
379 	info = kzalloc(sizeof(*info), GFP_KERNEL);
380 	if (info == NULL)
381 		panic("Unable to allocate metadata for IRQ%d\n", irq);
382 
383 	info->type = IRQT_UNBOUND;
384 	info->refcnt = -1;
385 
386 	irq_set_handler_data(irq, info);
387 
388 	list_add_tail(&info->list, &xen_irq_list_head);
389 }
390 
391 static int __must_check xen_allocate_irqs_dynamic(int nvec)
392 {
393 	int i, irq = irq_alloc_descs(-1, 0, nvec, -1);
394 
395 	if (irq >= 0) {
396 		for (i = 0; i < nvec; i++)
397 			xen_irq_init(irq + i);
398 	}
399 
400 	return irq;
401 }
402 
403 static inline int __must_check xen_allocate_irq_dynamic(void)
404 {
405 
406 	return xen_allocate_irqs_dynamic(1);
407 }
408 
409 static int __must_check xen_allocate_irq_gsi(unsigned gsi)
410 {
411 	int irq;
412 
413 	/*
414 	 * A PV guest has no concept of a GSI (since it has no ACPI
415 	 * nor access to/knowledge of the physical APICs). Therefore
416 	 * all IRQs are dynamically allocated from the entire IRQ
417 	 * space.
418 	 */
419 	if (xen_pv_domain() && !xen_initial_domain())
420 		return xen_allocate_irq_dynamic();
421 
422 	/* Legacy IRQ descriptors are already allocated by the arch. */
423 	if (gsi < NR_IRQS_LEGACY)
424 		irq = gsi;
425 	else
426 		irq = irq_alloc_desc_at(gsi, -1);
427 
428 	xen_irq_init(irq);
429 
430 	return irq;
431 }
432 
433 static void xen_free_irq(unsigned irq)
434 {
435 	struct irq_info *info = irq_get_handler_data(irq);
436 
437 	if (WARN_ON(!info))
438 		return;
439 
440 	list_del(&info->list);
441 
442 	irq_set_handler_data(irq, NULL);
443 
444 	WARN_ON(info->refcnt > 0);
445 
446 	kfree(info);
447 
448 	/* Legacy IRQ descriptors are managed by the arch. */
449 	if (irq < NR_IRQS_LEGACY)
450 		return;
451 
452 	irq_free_desc(irq);
453 }
454 
455 static void xen_evtchn_close(unsigned int port)
456 {
457 	struct evtchn_close close;
458 
459 	close.port = port;
460 	if (HYPERVISOR_event_channel_op(EVTCHNOP_close, &close) != 0)
461 		BUG();
462 }
463 
464 static void pirq_query_unmask(int irq)
465 {
466 	struct physdev_irq_status_query irq_status;
467 	struct irq_info *info = info_for_irq(irq);
468 
469 	BUG_ON(info->type != IRQT_PIRQ);
470 
471 	irq_status.irq = pirq_from_irq(irq);
472 	if (HYPERVISOR_physdev_op(PHYSDEVOP_irq_status_query, &irq_status))
473 		irq_status.flags = 0;
474 
475 	info->u.pirq.flags &= ~PIRQ_NEEDS_EOI;
476 	if (irq_status.flags & XENIRQSTAT_needs_eoi)
477 		info->u.pirq.flags |= PIRQ_NEEDS_EOI;
478 }
479 
480 static void eoi_pirq(struct irq_data *data)
481 {
482 	int evtchn = evtchn_from_irq(data->irq);
483 	struct physdev_eoi eoi = { .irq = pirq_from_irq(data->irq) };
484 	int rc = 0;
485 
486 	irq_move_irq(data);
487 
488 	if (VALID_EVTCHN(evtchn))
489 		clear_evtchn(evtchn);
490 
491 	if (pirq_needs_eoi(data->irq)) {
492 		rc = HYPERVISOR_physdev_op(PHYSDEVOP_eoi, &eoi);
493 		WARN_ON(rc);
494 	}
495 }
496 
497 static void mask_ack_pirq(struct irq_data *data)
498 {
499 	disable_dynirq(data);
500 	eoi_pirq(data);
501 }
502 
503 static unsigned int __startup_pirq(unsigned int irq)
504 {
505 	struct evtchn_bind_pirq bind_pirq;
506 	struct irq_info *info = info_for_irq(irq);
507 	int evtchn = evtchn_from_irq(irq);
508 	int rc;
509 
510 	BUG_ON(info->type != IRQT_PIRQ);
511 
512 	if (VALID_EVTCHN(evtchn))
513 		goto out;
514 
515 	bind_pirq.pirq = pirq_from_irq(irq);
516 	/* NB. We are happy to share unless we are probing. */
517 	bind_pirq.flags = info->u.pirq.flags & PIRQ_SHAREABLE ?
518 					BIND_PIRQ__WILL_SHARE : 0;
519 	rc = HYPERVISOR_event_channel_op(EVTCHNOP_bind_pirq, &bind_pirq);
520 	if (rc != 0) {
521 		pr_warn("Failed to obtain physical IRQ %d\n", irq);
522 		return 0;
523 	}
524 	evtchn = bind_pirq.port;
525 
526 	pirq_query_unmask(irq);
527 
528 	rc = set_evtchn_to_irq(evtchn, irq);
529 	if (rc)
530 		goto err;
531 
532 	info->evtchn = evtchn;
533 	bind_evtchn_to_cpu(evtchn, 0);
534 
535 	rc = xen_evtchn_port_setup(info);
536 	if (rc)
537 		goto err;
538 
539 out:
540 	unmask_evtchn(evtchn);
541 	eoi_pirq(irq_get_irq_data(irq));
542 
543 	return 0;
544 
545 err:
546 	pr_err("irq%d: Failed to set port to irq mapping (%d)\n", irq, rc);
547 	xen_evtchn_close(evtchn);
548 	return 0;
549 }
550 
551 static unsigned int startup_pirq(struct irq_data *data)
552 {
553 	return __startup_pirq(data->irq);
554 }
555 
556 static void shutdown_pirq(struct irq_data *data)
557 {
558 	unsigned int irq = data->irq;
559 	struct irq_info *info = info_for_irq(irq);
560 	unsigned evtchn = evtchn_from_irq(irq);
561 
562 	BUG_ON(info->type != IRQT_PIRQ);
563 
564 	if (!VALID_EVTCHN(evtchn))
565 		return;
566 
567 	mask_evtchn(evtchn);
568 	xen_evtchn_close(evtchn);
569 	xen_irq_info_cleanup(info);
570 }
571 
572 static void enable_pirq(struct irq_data *data)
573 {
574 	startup_pirq(data);
575 }
576 
577 static void disable_pirq(struct irq_data *data)
578 {
579 	disable_dynirq(data);
580 }
581 
582 int xen_irq_from_gsi(unsigned gsi)
583 {
584 	struct irq_info *info;
585 
586 	list_for_each_entry(info, &xen_irq_list_head, list) {
587 		if (info->type != IRQT_PIRQ)
588 			continue;
589 
590 		if (info->u.pirq.gsi == gsi)
591 			return info->irq;
592 	}
593 
594 	return -1;
595 }
596 EXPORT_SYMBOL_GPL(xen_irq_from_gsi);
597 
598 static void __unbind_from_irq(unsigned int irq)
599 {
600 	int evtchn = evtchn_from_irq(irq);
601 	struct irq_info *info = irq_get_handler_data(irq);
602 
603 	if (info->refcnt > 0) {
604 		info->refcnt--;
605 		if (info->refcnt != 0)
606 			return;
607 	}
608 
609 	if (VALID_EVTCHN(evtchn)) {
610 		unsigned int cpu = cpu_from_irq(irq);
611 
612 		xen_evtchn_close(evtchn);
613 
614 		switch (type_from_irq(irq)) {
615 		case IRQT_VIRQ:
616 			per_cpu(virq_to_irq, cpu)[virq_from_irq(irq)] = -1;
617 			break;
618 		case IRQT_IPI:
619 			per_cpu(ipi_to_irq, cpu)[ipi_from_irq(irq)] = -1;
620 			break;
621 		default:
622 			break;
623 		}
624 
625 		xen_irq_info_cleanup(info);
626 	}
627 
628 	BUG_ON(info_for_irq(irq)->type == IRQT_UNBOUND);
629 
630 	xen_free_irq(irq);
631 }
632 
633 /*
634  * Do not make any assumptions regarding the relationship between the
635  * IRQ number returned here and the Xen pirq argument.
636  *
637  * Note: We don't assign an event channel until the irq actually started
638  * up.  Return an existing irq if we've already got one for the gsi.
639  *
640  * Shareable implies level triggered, not shareable implies edge
641  * triggered here.
642  */
643 int xen_bind_pirq_gsi_to_irq(unsigned gsi,
644 			     unsigned pirq, int shareable, char *name)
645 {
646 	int irq = -1;
647 	struct physdev_irq irq_op;
648 	int ret;
649 
650 	mutex_lock(&irq_mapping_update_lock);
651 
652 	irq = xen_irq_from_gsi(gsi);
653 	if (irq != -1) {
654 		pr_info("%s: returning irq %d for gsi %u\n",
655 			__func__, irq, gsi);
656 		goto out;
657 	}
658 
659 	irq = xen_allocate_irq_gsi(gsi);
660 	if (irq < 0)
661 		goto out;
662 
663 	irq_op.irq = irq;
664 	irq_op.vector = 0;
665 
666 	/* Only the privileged domain can do this. For non-priv, the pcifront
667 	 * driver provides a PCI bus that does the call to do exactly
668 	 * this in the priv domain. */
669 	if (xen_initial_domain() &&
670 	    HYPERVISOR_physdev_op(PHYSDEVOP_alloc_irq_vector, &irq_op)) {
671 		xen_free_irq(irq);
672 		irq = -ENOSPC;
673 		goto out;
674 	}
675 
676 	ret = xen_irq_info_pirq_setup(irq, 0, pirq, gsi, DOMID_SELF,
677 			       shareable ? PIRQ_SHAREABLE : 0);
678 	if (ret < 0) {
679 		__unbind_from_irq(irq);
680 		irq = ret;
681 		goto out;
682 	}
683 
684 	pirq_query_unmask(irq);
685 	/* We try to use the handler with the appropriate semantic for the
686 	 * type of interrupt: if the interrupt is an edge triggered
687 	 * interrupt we use handle_edge_irq.
688 	 *
689 	 * On the other hand if the interrupt is level triggered we use
690 	 * handle_fasteoi_irq like the native code does for this kind of
691 	 * interrupts.
692 	 *
693 	 * Depending on the Xen version, pirq_needs_eoi might return true
694 	 * not only for level triggered interrupts but for edge triggered
695 	 * interrupts too. In any case Xen always honors the eoi mechanism,
696 	 * not injecting any more pirqs of the same kind if the first one
697 	 * hasn't received an eoi yet. Therefore using the fasteoi handler
698 	 * is the right choice either way.
699 	 */
700 	if (shareable)
701 		irq_set_chip_and_handler_name(irq, &xen_pirq_chip,
702 				handle_fasteoi_irq, name);
703 	else
704 		irq_set_chip_and_handler_name(irq, &xen_pirq_chip,
705 				handle_edge_irq, name);
706 
707 out:
708 	mutex_unlock(&irq_mapping_update_lock);
709 
710 	return irq;
711 }
712 
713 #ifdef CONFIG_PCI_MSI
714 int xen_allocate_pirq_msi(struct pci_dev *dev, struct msi_desc *msidesc)
715 {
716 	int rc;
717 	struct physdev_get_free_pirq op_get_free_pirq;
718 
719 	op_get_free_pirq.type = MAP_PIRQ_TYPE_MSI;
720 	rc = HYPERVISOR_physdev_op(PHYSDEVOP_get_free_pirq, &op_get_free_pirq);
721 
722 	WARN_ONCE(rc == -ENOSYS,
723 		  "hypervisor does not support the PHYSDEVOP_get_free_pirq interface\n");
724 
725 	return rc ? -1 : op_get_free_pirq.pirq;
726 }
727 
728 int xen_bind_pirq_msi_to_irq(struct pci_dev *dev, struct msi_desc *msidesc,
729 			     int pirq, int nvec, const char *name, domid_t domid)
730 {
731 	int i, irq, ret;
732 
733 	mutex_lock(&irq_mapping_update_lock);
734 
735 	irq = xen_allocate_irqs_dynamic(nvec);
736 	if (irq < 0)
737 		goto out;
738 
739 	for (i = 0; i < nvec; i++) {
740 		irq_set_chip_and_handler_name(irq + i, &xen_pirq_chip, handle_edge_irq, name);
741 
742 		ret = xen_irq_info_pirq_setup(irq + i, 0, pirq + i, 0, domid,
743 					      i == 0 ? 0 : PIRQ_MSI_GROUP);
744 		if (ret < 0)
745 			goto error_irq;
746 	}
747 
748 	ret = irq_set_msi_desc(irq, msidesc);
749 	if (ret < 0)
750 		goto error_irq;
751 out:
752 	mutex_unlock(&irq_mapping_update_lock);
753 	return irq;
754 error_irq:
755 	for (; i >= 0; i--)
756 		__unbind_from_irq(irq + i);
757 	mutex_unlock(&irq_mapping_update_lock);
758 	return ret;
759 }
760 #endif
761 
762 int xen_destroy_irq(int irq)
763 {
764 	struct physdev_unmap_pirq unmap_irq;
765 	struct irq_info *info = info_for_irq(irq);
766 	int rc = -ENOENT;
767 
768 	mutex_lock(&irq_mapping_update_lock);
769 
770 	/*
771 	 * If trying to remove a vector in a MSI group different
772 	 * than the first one skip the PIRQ unmap unless this vector
773 	 * is the first one in the group.
774 	 */
775 	if (xen_initial_domain() && !(info->u.pirq.flags & PIRQ_MSI_GROUP)) {
776 		unmap_irq.pirq = info->u.pirq.pirq;
777 		unmap_irq.domid = info->u.pirq.domid;
778 		rc = HYPERVISOR_physdev_op(PHYSDEVOP_unmap_pirq, &unmap_irq);
779 		/* If another domain quits without making the pci_disable_msix
780 		 * call, the Xen hypervisor takes care of freeing the PIRQs
781 		 * (free_domain_pirqs).
782 		 */
783 		if ((rc == -ESRCH && info->u.pirq.domid != DOMID_SELF))
784 			pr_info("domain %d does not have %d anymore\n",
785 				info->u.pirq.domid, info->u.pirq.pirq);
786 		else if (rc) {
787 			pr_warn("unmap irq failed %d\n", rc);
788 			goto out;
789 		}
790 	}
791 
792 	xen_free_irq(irq);
793 
794 out:
795 	mutex_unlock(&irq_mapping_update_lock);
796 	return rc;
797 }
798 
799 int xen_irq_from_pirq(unsigned pirq)
800 {
801 	int irq;
802 
803 	struct irq_info *info;
804 
805 	mutex_lock(&irq_mapping_update_lock);
806 
807 	list_for_each_entry(info, &xen_irq_list_head, list) {
808 		if (info->type != IRQT_PIRQ)
809 			continue;
810 		irq = info->irq;
811 		if (info->u.pirq.pirq == pirq)
812 			goto out;
813 	}
814 	irq = -1;
815 out:
816 	mutex_unlock(&irq_mapping_update_lock);
817 
818 	return irq;
819 }
820 
821 
822 int xen_pirq_from_irq(unsigned irq)
823 {
824 	return pirq_from_irq(irq);
825 }
826 EXPORT_SYMBOL_GPL(xen_pirq_from_irq);
827 
828 int bind_evtchn_to_irq(unsigned int evtchn)
829 {
830 	int irq;
831 	int ret;
832 
833 	if (evtchn >= xen_evtchn_max_channels())
834 		return -ENOMEM;
835 
836 	mutex_lock(&irq_mapping_update_lock);
837 
838 	irq = get_evtchn_to_irq(evtchn);
839 
840 	if (irq == -1) {
841 		irq = xen_allocate_irq_dynamic();
842 		if (irq < 0)
843 			goto out;
844 
845 		irq_set_chip_and_handler_name(irq, &xen_dynamic_chip,
846 					      handle_edge_irq, "event");
847 
848 		ret = xen_irq_info_evtchn_setup(irq, evtchn);
849 		if (ret < 0) {
850 			__unbind_from_irq(irq);
851 			irq = ret;
852 			goto out;
853 		}
854 		/* New interdomain events are bound to VCPU 0. */
855 		bind_evtchn_to_cpu(evtchn, 0);
856 	} else {
857 		struct irq_info *info = info_for_irq(irq);
858 		WARN_ON(info == NULL || info->type != IRQT_EVTCHN);
859 	}
860 
861 out:
862 	mutex_unlock(&irq_mapping_update_lock);
863 
864 	return irq;
865 }
866 EXPORT_SYMBOL_GPL(bind_evtchn_to_irq);
867 
868 static int bind_ipi_to_irq(unsigned int ipi, unsigned int cpu)
869 {
870 	struct evtchn_bind_ipi bind_ipi;
871 	int evtchn, irq;
872 	int ret;
873 
874 	mutex_lock(&irq_mapping_update_lock);
875 
876 	irq = per_cpu(ipi_to_irq, cpu)[ipi];
877 
878 	if (irq == -1) {
879 		irq = xen_allocate_irq_dynamic();
880 		if (irq < 0)
881 			goto out;
882 
883 		irq_set_chip_and_handler_name(irq, &xen_percpu_chip,
884 					      handle_percpu_irq, "ipi");
885 
886 		bind_ipi.vcpu = cpu;
887 		if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_ipi,
888 						&bind_ipi) != 0)
889 			BUG();
890 		evtchn = bind_ipi.port;
891 
892 		ret = xen_irq_info_ipi_setup(cpu, irq, evtchn, ipi);
893 		if (ret < 0) {
894 			__unbind_from_irq(irq);
895 			irq = ret;
896 			goto out;
897 		}
898 		bind_evtchn_to_cpu(evtchn, cpu);
899 	} else {
900 		struct irq_info *info = info_for_irq(irq);
901 		WARN_ON(info == NULL || info->type != IRQT_IPI);
902 	}
903 
904  out:
905 	mutex_unlock(&irq_mapping_update_lock);
906 	return irq;
907 }
908 
909 int bind_interdomain_evtchn_to_irq(unsigned int remote_domain,
910 				   unsigned int remote_port)
911 {
912 	struct evtchn_bind_interdomain bind_interdomain;
913 	int err;
914 
915 	bind_interdomain.remote_dom  = remote_domain;
916 	bind_interdomain.remote_port = remote_port;
917 
918 	err = HYPERVISOR_event_channel_op(EVTCHNOP_bind_interdomain,
919 					  &bind_interdomain);
920 
921 	return err ? : bind_evtchn_to_irq(bind_interdomain.local_port);
922 }
923 EXPORT_SYMBOL_GPL(bind_interdomain_evtchn_to_irq);
924 
925 static int find_virq(unsigned int virq, unsigned int cpu)
926 {
927 	struct evtchn_status status;
928 	int port, rc = -ENOENT;
929 
930 	memset(&status, 0, sizeof(status));
931 	for (port = 0; port < xen_evtchn_max_channels(); port++) {
932 		status.dom = DOMID_SELF;
933 		status.port = port;
934 		rc = HYPERVISOR_event_channel_op(EVTCHNOP_status, &status);
935 		if (rc < 0)
936 			continue;
937 		if (status.status != EVTCHNSTAT_virq)
938 			continue;
939 		if (status.u.virq == virq && status.vcpu == cpu) {
940 			rc = port;
941 			break;
942 		}
943 	}
944 	return rc;
945 }
946 
947 /**
948  * xen_evtchn_nr_channels - number of usable event channel ports
949  *
950  * This may be less than the maximum supported by the current
951  * hypervisor ABI. Use xen_evtchn_max_channels() for the maximum
952  * supported.
953  */
954 unsigned xen_evtchn_nr_channels(void)
955 {
956         return evtchn_ops->nr_channels();
957 }
958 EXPORT_SYMBOL_GPL(xen_evtchn_nr_channels);
959 
960 int bind_virq_to_irq(unsigned int virq, unsigned int cpu, bool percpu)
961 {
962 	struct evtchn_bind_virq bind_virq;
963 	int evtchn, irq, ret;
964 
965 	mutex_lock(&irq_mapping_update_lock);
966 
967 	irq = per_cpu(virq_to_irq, cpu)[virq];
968 
969 	if (irq == -1) {
970 		irq = xen_allocate_irq_dynamic();
971 		if (irq < 0)
972 			goto out;
973 
974 		if (percpu)
975 			irq_set_chip_and_handler_name(irq, &xen_percpu_chip,
976 						      handle_percpu_irq, "virq");
977 		else
978 			irq_set_chip_and_handler_name(irq, &xen_dynamic_chip,
979 						      handle_edge_irq, "virq");
980 
981 		bind_virq.virq = virq;
982 		bind_virq.vcpu = cpu;
983 		ret = HYPERVISOR_event_channel_op(EVTCHNOP_bind_virq,
984 						&bind_virq);
985 		if (ret == 0)
986 			evtchn = bind_virq.port;
987 		else {
988 			if (ret == -EEXIST)
989 				ret = find_virq(virq, cpu);
990 			BUG_ON(ret < 0);
991 			evtchn = ret;
992 		}
993 
994 		ret = xen_irq_info_virq_setup(cpu, irq, evtchn, virq);
995 		if (ret < 0) {
996 			__unbind_from_irq(irq);
997 			irq = ret;
998 			goto out;
999 		}
1000 
1001 		bind_evtchn_to_cpu(evtchn, cpu);
1002 	} else {
1003 		struct irq_info *info = info_for_irq(irq);
1004 		WARN_ON(info == NULL || info->type != IRQT_VIRQ);
1005 	}
1006 
1007 out:
1008 	mutex_unlock(&irq_mapping_update_lock);
1009 
1010 	return irq;
1011 }
1012 
1013 static void unbind_from_irq(unsigned int irq)
1014 {
1015 	mutex_lock(&irq_mapping_update_lock);
1016 	__unbind_from_irq(irq);
1017 	mutex_unlock(&irq_mapping_update_lock);
1018 }
1019 
1020 int bind_evtchn_to_irqhandler(unsigned int evtchn,
1021 			      irq_handler_t handler,
1022 			      unsigned long irqflags,
1023 			      const char *devname, void *dev_id)
1024 {
1025 	int irq, retval;
1026 
1027 	irq = bind_evtchn_to_irq(evtchn);
1028 	if (irq < 0)
1029 		return irq;
1030 	retval = request_irq(irq, handler, irqflags, devname, dev_id);
1031 	if (retval != 0) {
1032 		unbind_from_irq(irq);
1033 		return retval;
1034 	}
1035 
1036 	return irq;
1037 }
1038 EXPORT_SYMBOL_GPL(bind_evtchn_to_irqhandler);
1039 
1040 int bind_interdomain_evtchn_to_irqhandler(unsigned int remote_domain,
1041 					  unsigned int remote_port,
1042 					  irq_handler_t handler,
1043 					  unsigned long irqflags,
1044 					  const char *devname,
1045 					  void *dev_id)
1046 {
1047 	int irq, retval;
1048 
1049 	irq = bind_interdomain_evtchn_to_irq(remote_domain, remote_port);
1050 	if (irq < 0)
1051 		return irq;
1052 
1053 	retval = request_irq(irq, handler, irqflags, devname, dev_id);
1054 	if (retval != 0) {
1055 		unbind_from_irq(irq);
1056 		return retval;
1057 	}
1058 
1059 	return irq;
1060 }
1061 EXPORT_SYMBOL_GPL(bind_interdomain_evtchn_to_irqhandler);
1062 
1063 int bind_virq_to_irqhandler(unsigned int virq, unsigned int cpu,
1064 			    irq_handler_t handler,
1065 			    unsigned long irqflags, const char *devname, void *dev_id)
1066 {
1067 	int irq, retval;
1068 
1069 	irq = bind_virq_to_irq(virq, cpu, irqflags & IRQF_PERCPU);
1070 	if (irq < 0)
1071 		return irq;
1072 	retval = request_irq(irq, handler, irqflags, devname, dev_id);
1073 	if (retval != 0) {
1074 		unbind_from_irq(irq);
1075 		return retval;
1076 	}
1077 
1078 	return irq;
1079 }
1080 EXPORT_SYMBOL_GPL(bind_virq_to_irqhandler);
1081 
1082 int bind_ipi_to_irqhandler(enum ipi_vector ipi,
1083 			   unsigned int cpu,
1084 			   irq_handler_t handler,
1085 			   unsigned long irqflags,
1086 			   const char *devname,
1087 			   void *dev_id)
1088 {
1089 	int irq, retval;
1090 
1091 	irq = bind_ipi_to_irq(ipi, cpu);
1092 	if (irq < 0)
1093 		return irq;
1094 
1095 	irqflags |= IRQF_NO_SUSPEND | IRQF_FORCE_RESUME | IRQF_EARLY_RESUME;
1096 	retval = request_irq(irq, handler, irqflags, devname, dev_id);
1097 	if (retval != 0) {
1098 		unbind_from_irq(irq);
1099 		return retval;
1100 	}
1101 
1102 	return irq;
1103 }
1104 
1105 void unbind_from_irqhandler(unsigned int irq, void *dev_id)
1106 {
1107 	struct irq_info *info = irq_get_handler_data(irq);
1108 
1109 	if (WARN_ON(!info))
1110 		return;
1111 	free_irq(irq, dev_id);
1112 	unbind_from_irq(irq);
1113 }
1114 EXPORT_SYMBOL_GPL(unbind_from_irqhandler);
1115 
1116 /**
1117  * xen_set_irq_priority() - set an event channel priority.
1118  * @irq:irq bound to an event channel.
1119  * @priority: priority between XEN_IRQ_PRIORITY_MAX and XEN_IRQ_PRIORITY_MIN.
1120  */
1121 int xen_set_irq_priority(unsigned irq, unsigned priority)
1122 {
1123 	struct evtchn_set_priority set_priority;
1124 
1125 	set_priority.port = evtchn_from_irq(irq);
1126 	set_priority.priority = priority;
1127 
1128 	return HYPERVISOR_event_channel_op(EVTCHNOP_set_priority,
1129 					   &set_priority);
1130 }
1131 EXPORT_SYMBOL_GPL(xen_set_irq_priority);
1132 
1133 int evtchn_make_refcounted(unsigned int evtchn)
1134 {
1135 	int irq = get_evtchn_to_irq(evtchn);
1136 	struct irq_info *info;
1137 
1138 	if (irq == -1)
1139 		return -ENOENT;
1140 
1141 	info = irq_get_handler_data(irq);
1142 
1143 	if (!info)
1144 		return -ENOENT;
1145 
1146 	WARN_ON(info->refcnt != -1);
1147 
1148 	info->refcnt = 1;
1149 
1150 	return 0;
1151 }
1152 EXPORT_SYMBOL_GPL(evtchn_make_refcounted);
1153 
1154 int evtchn_get(unsigned int evtchn)
1155 {
1156 	int irq;
1157 	struct irq_info *info;
1158 	int err = -ENOENT;
1159 
1160 	if (evtchn >= xen_evtchn_max_channels())
1161 		return -EINVAL;
1162 
1163 	mutex_lock(&irq_mapping_update_lock);
1164 
1165 	irq = get_evtchn_to_irq(evtchn);
1166 	if (irq == -1)
1167 		goto done;
1168 
1169 	info = irq_get_handler_data(irq);
1170 
1171 	if (!info)
1172 		goto done;
1173 
1174 	err = -EINVAL;
1175 	if (info->refcnt <= 0)
1176 		goto done;
1177 
1178 	info->refcnt++;
1179 	err = 0;
1180  done:
1181 	mutex_unlock(&irq_mapping_update_lock);
1182 
1183 	return err;
1184 }
1185 EXPORT_SYMBOL_GPL(evtchn_get);
1186 
1187 void evtchn_put(unsigned int evtchn)
1188 {
1189 	int irq = get_evtchn_to_irq(evtchn);
1190 	if (WARN_ON(irq == -1))
1191 		return;
1192 	unbind_from_irq(irq);
1193 }
1194 EXPORT_SYMBOL_GPL(evtchn_put);
1195 
1196 void xen_send_IPI_one(unsigned int cpu, enum ipi_vector vector)
1197 {
1198 	int irq;
1199 
1200 #ifdef CONFIG_X86
1201 	if (unlikely(vector == XEN_NMI_VECTOR)) {
1202 		int rc =  HYPERVISOR_vcpu_op(VCPUOP_send_nmi, cpu, NULL);
1203 		if (rc < 0)
1204 			printk(KERN_WARNING "Sending nmi to CPU%d failed (rc:%d)\n", cpu, rc);
1205 		return;
1206 	}
1207 #endif
1208 	irq = per_cpu(ipi_to_irq, cpu)[vector];
1209 	BUG_ON(irq < 0);
1210 	notify_remote_via_irq(irq);
1211 }
1212 
1213 static DEFINE_PER_CPU(unsigned, xed_nesting_count);
1214 
1215 static void __xen_evtchn_do_upcall(void)
1216 {
1217 	struct vcpu_info *vcpu_info = __this_cpu_read(xen_vcpu);
1218 	int cpu = get_cpu();
1219 	unsigned count;
1220 
1221 	do {
1222 		vcpu_info->evtchn_upcall_pending = 0;
1223 
1224 		if (__this_cpu_inc_return(xed_nesting_count) - 1)
1225 			goto out;
1226 
1227 		xen_evtchn_handle_events(cpu);
1228 
1229 		BUG_ON(!irqs_disabled());
1230 
1231 		count = __this_cpu_read(xed_nesting_count);
1232 		__this_cpu_write(xed_nesting_count, 0);
1233 	} while (count != 1 || vcpu_info->evtchn_upcall_pending);
1234 
1235 out:
1236 
1237 	put_cpu();
1238 }
1239 
1240 void xen_evtchn_do_upcall(struct pt_regs *regs)
1241 {
1242 	struct pt_regs *old_regs = set_irq_regs(regs);
1243 
1244 	irq_enter();
1245 #ifdef CONFIG_X86
1246 	exit_idle();
1247 	inc_irq_stat(irq_hv_callback_count);
1248 #endif
1249 
1250 	__xen_evtchn_do_upcall();
1251 
1252 	irq_exit();
1253 	set_irq_regs(old_regs);
1254 }
1255 
1256 void xen_hvm_evtchn_do_upcall(void)
1257 {
1258 	__xen_evtchn_do_upcall();
1259 }
1260 EXPORT_SYMBOL_GPL(xen_hvm_evtchn_do_upcall);
1261 
1262 /* Rebind a new event channel to an existing irq. */
1263 void rebind_evtchn_irq(int evtchn, int irq)
1264 {
1265 	struct irq_info *info = info_for_irq(irq);
1266 
1267 	if (WARN_ON(!info))
1268 		return;
1269 
1270 	/* Make sure the irq is masked, since the new event channel
1271 	   will also be masked. */
1272 	disable_irq(irq);
1273 
1274 	mutex_lock(&irq_mapping_update_lock);
1275 
1276 	/* After resume the irq<->evtchn mappings are all cleared out */
1277 	BUG_ON(get_evtchn_to_irq(evtchn) != -1);
1278 	/* Expect irq to have been bound before,
1279 	   so there should be a proper type */
1280 	BUG_ON(info->type == IRQT_UNBOUND);
1281 
1282 	(void)xen_irq_info_evtchn_setup(irq, evtchn);
1283 
1284 	mutex_unlock(&irq_mapping_update_lock);
1285 
1286         bind_evtchn_to_cpu(evtchn, info->cpu);
1287 	/* This will be deferred until interrupt is processed */
1288 	irq_set_affinity(irq, cpumask_of(info->cpu));
1289 
1290 	/* Unmask the event channel. */
1291 	enable_irq(irq);
1292 }
1293 
1294 /* Rebind an evtchn so that it gets delivered to a specific cpu */
1295 static int rebind_irq_to_cpu(unsigned irq, unsigned tcpu)
1296 {
1297 	struct evtchn_bind_vcpu bind_vcpu;
1298 	int evtchn = evtchn_from_irq(irq);
1299 	int masked;
1300 
1301 	if (!VALID_EVTCHN(evtchn))
1302 		return -1;
1303 
1304 	if (!xen_support_evtchn_rebind())
1305 		return -1;
1306 
1307 	/* Send future instances of this interrupt to other vcpu. */
1308 	bind_vcpu.port = evtchn;
1309 	bind_vcpu.vcpu = tcpu;
1310 
1311 	/*
1312 	 * Mask the event while changing the VCPU binding to prevent
1313 	 * it being delivered on an unexpected VCPU.
1314 	 */
1315 	masked = test_and_set_mask(evtchn);
1316 
1317 	/*
1318 	 * If this fails, it usually just indicates that we're dealing with a
1319 	 * virq or IPI channel, which don't actually need to be rebound. Ignore
1320 	 * it, but don't do the xenlinux-level rebind in that case.
1321 	 */
1322 	if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_vcpu, &bind_vcpu) >= 0)
1323 		bind_evtchn_to_cpu(evtchn, tcpu);
1324 
1325 	if (!masked)
1326 		unmask_evtchn(evtchn);
1327 
1328 	return 0;
1329 }
1330 
1331 static int set_affinity_irq(struct irq_data *data, const struct cpumask *dest,
1332 			    bool force)
1333 {
1334 	unsigned tcpu = cpumask_first_and(dest, cpu_online_mask);
1335 
1336 	return rebind_irq_to_cpu(data->irq, tcpu);
1337 }
1338 
1339 static void enable_dynirq(struct irq_data *data)
1340 {
1341 	int evtchn = evtchn_from_irq(data->irq);
1342 
1343 	if (VALID_EVTCHN(evtchn))
1344 		unmask_evtchn(evtchn);
1345 }
1346 
1347 static void disable_dynirq(struct irq_data *data)
1348 {
1349 	int evtchn = evtchn_from_irq(data->irq);
1350 
1351 	if (VALID_EVTCHN(evtchn))
1352 		mask_evtchn(evtchn);
1353 }
1354 
1355 static void ack_dynirq(struct irq_data *data)
1356 {
1357 	int evtchn = evtchn_from_irq(data->irq);
1358 
1359 	irq_move_irq(data);
1360 
1361 	if (VALID_EVTCHN(evtchn))
1362 		clear_evtchn(evtchn);
1363 }
1364 
1365 static void mask_ack_dynirq(struct irq_data *data)
1366 {
1367 	disable_dynirq(data);
1368 	ack_dynirq(data);
1369 }
1370 
1371 static int retrigger_dynirq(struct irq_data *data)
1372 {
1373 	unsigned int evtchn = evtchn_from_irq(data->irq);
1374 	int masked;
1375 
1376 	if (!VALID_EVTCHN(evtchn))
1377 		return 0;
1378 
1379 	masked = test_and_set_mask(evtchn);
1380 	set_evtchn(evtchn);
1381 	if (!masked)
1382 		unmask_evtchn(evtchn);
1383 
1384 	return 1;
1385 }
1386 
1387 static void restore_pirqs(void)
1388 {
1389 	int pirq, rc, irq, gsi;
1390 	struct physdev_map_pirq map_irq;
1391 	struct irq_info *info;
1392 
1393 	list_for_each_entry(info, &xen_irq_list_head, list) {
1394 		if (info->type != IRQT_PIRQ)
1395 			continue;
1396 
1397 		pirq = info->u.pirq.pirq;
1398 		gsi = info->u.pirq.gsi;
1399 		irq = info->irq;
1400 
1401 		/* save/restore of PT devices doesn't work, so at this point the
1402 		 * only devices present are GSI based emulated devices */
1403 		if (!gsi)
1404 			continue;
1405 
1406 		map_irq.domid = DOMID_SELF;
1407 		map_irq.type = MAP_PIRQ_TYPE_GSI;
1408 		map_irq.index = gsi;
1409 		map_irq.pirq = pirq;
1410 
1411 		rc = HYPERVISOR_physdev_op(PHYSDEVOP_map_pirq, &map_irq);
1412 		if (rc) {
1413 			pr_warn("xen map irq failed gsi=%d irq=%d pirq=%d rc=%d\n",
1414 				gsi, irq, pirq, rc);
1415 			xen_free_irq(irq);
1416 			continue;
1417 		}
1418 
1419 		printk(KERN_DEBUG "xen: --> irq=%d, pirq=%d\n", irq, map_irq.pirq);
1420 
1421 		__startup_pirq(irq);
1422 	}
1423 }
1424 
1425 static void restore_cpu_virqs(unsigned int cpu)
1426 {
1427 	struct evtchn_bind_virq bind_virq;
1428 	int virq, irq, evtchn;
1429 
1430 	for (virq = 0; virq < NR_VIRQS; virq++) {
1431 		if ((irq = per_cpu(virq_to_irq, cpu)[virq]) == -1)
1432 			continue;
1433 
1434 		BUG_ON(virq_from_irq(irq) != virq);
1435 
1436 		/* Get a new binding from Xen. */
1437 		bind_virq.virq = virq;
1438 		bind_virq.vcpu = cpu;
1439 		if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_virq,
1440 						&bind_virq) != 0)
1441 			BUG();
1442 		evtchn = bind_virq.port;
1443 
1444 		/* Record the new mapping. */
1445 		(void)xen_irq_info_virq_setup(cpu, irq, evtchn, virq);
1446 		bind_evtchn_to_cpu(evtchn, cpu);
1447 	}
1448 }
1449 
1450 static void restore_cpu_ipis(unsigned int cpu)
1451 {
1452 	struct evtchn_bind_ipi bind_ipi;
1453 	int ipi, irq, evtchn;
1454 
1455 	for (ipi = 0; ipi < XEN_NR_IPIS; ipi++) {
1456 		if ((irq = per_cpu(ipi_to_irq, cpu)[ipi]) == -1)
1457 			continue;
1458 
1459 		BUG_ON(ipi_from_irq(irq) != ipi);
1460 
1461 		/* Get a new binding from Xen. */
1462 		bind_ipi.vcpu = cpu;
1463 		if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_ipi,
1464 						&bind_ipi) != 0)
1465 			BUG();
1466 		evtchn = bind_ipi.port;
1467 
1468 		/* Record the new mapping. */
1469 		(void)xen_irq_info_ipi_setup(cpu, irq, evtchn, ipi);
1470 		bind_evtchn_to_cpu(evtchn, cpu);
1471 	}
1472 }
1473 
1474 /* Clear an irq's pending state, in preparation for polling on it */
1475 void xen_clear_irq_pending(int irq)
1476 {
1477 	int evtchn = evtchn_from_irq(irq);
1478 
1479 	if (VALID_EVTCHN(evtchn))
1480 		clear_evtchn(evtchn);
1481 }
1482 EXPORT_SYMBOL(xen_clear_irq_pending);
1483 void xen_set_irq_pending(int irq)
1484 {
1485 	int evtchn = evtchn_from_irq(irq);
1486 
1487 	if (VALID_EVTCHN(evtchn))
1488 		set_evtchn(evtchn);
1489 }
1490 
1491 bool xen_test_irq_pending(int irq)
1492 {
1493 	int evtchn = evtchn_from_irq(irq);
1494 	bool ret = false;
1495 
1496 	if (VALID_EVTCHN(evtchn))
1497 		ret = test_evtchn(evtchn);
1498 
1499 	return ret;
1500 }
1501 
1502 /* Poll waiting for an irq to become pending with timeout.  In the usual case,
1503  * the irq will be disabled so it won't deliver an interrupt. */
1504 void xen_poll_irq_timeout(int irq, u64 timeout)
1505 {
1506 	evtchn_port_t evtchn = evtchn_from_irq(irq);
1507 
1508 	if (VALID_EVTCHN(evtchn)) {
1509 		struct sched_poll poll;
1510 
1511 		poll.nr_ports = 1;
1512 		poll.timeout = timeout;
1513 		set_xen_guest_handle(poll.ports, &evtchn);
1514 
1515 		if (HYPERVISOR_sched_op(SCHEDOP_poll, &poll) != 0)
1516 			BUG();
1517 	}
1518 }
1519 EXPORT_SYMBOL(xen_poll_irq_timeout);
1520 /* Poll waiting for an irq to become pending.  In the usual case, the
1521  * irq will be disabled so it won't deliver an interrupt. */
1522 void xen_poll_irq(int irq)
1523 {
1524 	xen_poll_irq_timeout(irq, 0 /* no timeout */);
1525 }
1526 
1527 /* Check whether the IRQ line is shared with other guests. */
1528 int xen_test_irq_shared(int irq)
1529 {
1530 	struct irq_info *info = info_for_irq(irq);
1531 	struct physdev_irq_status_query irq_status;
1532 
1533 	if (WARN_ON(!info))
1534 		return -ENOENT;
1535 
1536 	irq_status.irq = info->u.pirq.pirq;
1537 
1538 	if (HYPERVISOR_physdev_op(PHYSDEVOP_irq_status_query, &irq_status))
1539 		return 0;
1540 	return !(irq_status.flags & XENIRQSTAT_shared);
1541 }
1542 EXPORT_SYMBOL_GPL(xen_test_irq_shared);
1543 
1544 void xen_irq_resume(void)
1545 {
1546 	unsigned int cpu;
1547 	struct irq_info *info;
1548 
1549 	/* New event-channel space is not 'live' yet. */
1550 	xen_evtchn_mask_all();
1551 	xen_evtchn_resume();
1552 
1553 	/* No IRQ <-> event-channel mappings. */
1554 	list_for_each_entry(info, &xen_irq_list_head, list)
1555 		info->evtchn = 0; /* zap event-channel binding */
1556 
1557 	clear_evtchn_to_irq_all();
1558 
1559 	for_each_possible_cpu(cpu) {
1560 		restore_cpu_virqs(cpu);
1561 		restore_cpu_ipis(cpu);
1562 	}
1563 
1564 	restore_pirqs();
1565 }
1566 
1567 static struct irq_chip xen_dynamic_chip __read_mostly = {
1568 	.name			= "xen-dyn",
1569 
1570 	.irq_disable		= disable_dynirq,
1571 	.irq_mask		= disable_dynirq,
1572 	.irq_unmask		= enable_dynirq,
1573 
1574 	.irq_ack		= ack_dynirq,
1575 	.irq_mask_ack		= mask_ack_dynirq,
1576 
1577 	.irq_set_affinity	= set_affinity_irq,
1578 	.irq_retrigger		= retrigger_dynirq,
1579 };
1580 
1581 static struct irq_chip xen_pirq_chip __read_mostly = {
1582 	.name			= "xen-pirq",
1583 
1584 	.irq_startup		= startup_pirq,
1585 	.irq_shutdown		= shutdown_pirq,
1586 	.irq_enable		= enable_pirq,
1587 	.irq_disable		= disable_pirq,
1588 
1589 	.irq_mask		= disable_dynirq,
1590 	.irq_unmask		= enable_dynirq,
1591 
1592 	.irq_ack		= eoi_pirq,
1593 	.irq_eoi		= eoi_pirq,
1594 	.irq_mask_ack		= mask_ack_pirq,
1595 
1596 	.irq_set_affinity	= set_affinity_irq,
1597 
1598 	.irq_retrigger		= retrigger_dynirq,
1599 };
1600 
1601 static struct irq_chip xen_percpu_chip __read_mostly = {
1602 	.name			= "xen-percpu",
1603 
1604 	.irq_disable		= disable_dynirq,
1605 	.irq_mask		= disable_dynirq,
1606 	.irq_unmask		= enable_dynirq,
1607 
1608 	.irq_ack		= ack_dynirq,
1609 };
1610 
1611 int xen_set_callback_via(uint64_t via)
1612 {
1613 	struct xen_hvm_param a;
1614 	a.domid = DOMID_SELF;
1615 	a.index = HVM_PARAM_CALLBACK_IRQ;
1616 	a.value = via;
1617 	return HYPERVISOR_hvm_op(HVMOP_set_param, &a);
1618 }
1619 EXPORT_SYMBOL_GPL(xen_set_callback_via);
1620 
1621 #ifdef CONFIG_XEN_PVHVM
1622 /* Vector callbacks are better than PCI interrupts to receive event
1623  * channel notifications because we can receive vector callbacks on any
1624  * vcpu and we don't need PCI support or APIC interactions. */
1625 void xen_callback_vector(void)
1626 {
1627 	int rc;
1628 	uint64_t callback_via;
1629 	if (xen_have_vector_callback) {
1630 		callback_via = HVM_CALLBACK_VECTOR(HYPERVISOR_CALLBACK_VECTOR);
1631 		rc = xen_set_callback_via(callback_via);
1632 		if (rc) {
1633 			pr_err("Request for Xen HVM callback vector failed\n");
1634 			xen_have_vector_callback = 0;
1635 			return;
1636 		}
1637 		pr_info("Xen HVM callback vector for event delivery is enabled\n");
1638 		/* in the restore case the vector has already been allocated */
1639 		if (!test_bit(HYPERVISOR_CALLBACK_VECTOR, used_vectors))
1640 			alloc_intr_gate(HYPERVISOR_CALLBACK_VECTOR,
1641 					xen_hvm_callback_vector);
1642 	}
1643 }
1644 #else
1645 void xen_callback_vector(void) {}
1646 #endif
1647 
1648 #undef MODULE_PARAM_PREFIX
1649 #define MODULE_PARAM_PREFIX "xen."
1650 
1651 static bool fifo_events = true;
1652 module_param(fifo_events, bool, 0);
1653 
1654 void __init xen_init_IRQ(void)
1655 {
1656 	int ret = -EINVAL;
1657 
1658 	if (fifo_events)
1659 		ret = xen_evtchn_fifo_init();
1660 	if (ret < 0)
1661 		xen_evtchn_2l_init();
1662 
1663 	evtchn_to_irq = kcalloc(EVTCHN_ROW(xen_evtchn_max_channels()),
1664 				sizeof(*evtchn_to_irq), GFP_KERNEL);
1665 	BUG_ON(!evtchn_to_irq);
1666 
1667 	/* No event channels are 'live' right now. */
1668 	xen_evtchn_mask_all();
1669 
1670 	pirq_needs_eoi = pirq_needs_eoi_flag;
1671 
1672 #ifdef CONFIG_X86
1673 	if (xen_pv_domain()) {
1674 		irq_ctx_init(smp_processor_id());
1675 		if (xen_initial_domain())
1676 			pci_xen_initial_domain();
1677 	}
1678 	if (xen_feature(XENFEAT_hvm_callback_vector))
1679 		xen_callback_vector();
1680 
1681 	if (xen_hvm_domain()) {
1682 		native_init_IRQ();
1683 		/* pci_xen_hvm_init must be called after native_init_IRQ so that
1684 		 * __acpi_register_gsi can point at the right function */
1685 		pci_xen_hvm_init();
1686 	} else {
1687 		int rc;
1688 		struct physdev_pirq_eoi_gmfn eoi_gmfn;
1689 
1690 		pirq_eoi_map = (void *)__get_free_page(GFP_KERNEL|__GFP_ZERO);
1691 		eoi_gmfn.gmfn = virt_to_gfn(pirq_eoi_map);
1692 		rc = HYPERVISOR_physdev_op(PHYSDEVOP_pirq_eoi_gmfn_v2, &eoi_gmfn);
1693 		/* TODO: No PVH support for PIRQ EOI */
1694 		if (rc != 0) {
1695 			free_page((unsigned long) pirq_eoi_map);
1696 			pirq_eoi_map = NULL;
1697 		} else
1698 			pirq_needs_eoi = pirq_check_eoi_map;
1699 	}
1700 #endif
1701 }
1702