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.
11 * There are four kinds of events which can be mapped to an event
14 * 1. Inter-domain notifications. This includes all the virtual
15 * device events, since they're driven by front-ends in another domain
17 * 2. VIRQs, typically used for timers. These are per-cpu events.
19 * 4. PIRQs - Hardware interrupts.
21 * Jeremy Fitzhardinge <jeremy@xensource.com>, XenSource Inc, 2007
24 #define pr_fmt(fmt) "xen:" KBUILD_MODNAME ": " fmt
26 #include <linux/linkage.h>
27 #include <linux/interrupt.h>
28 #include <linux/irq.h>
29 #include <linux/moduleparam.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>
38 #include <asm/ptrace.h>
41 #include <asm/io_apic.h>
42 #include <asm/i8259.h>
43 #include <asm/xen/pci.h>
45 #include <asm/sync_bitops.h>
46 #include <asm/xen/hypercall.h>
47 #include <asm/xen/hypervisor.h>
52 #include <xen/xen-ops.h>
53 #include <xen/events.h>
54 #include <xen/interface/xen.h>
55 #include <xen/interface/event_channel.h>
56 #include <xen/interface/hvm/hvm_op.h>
57 #include <xen/interface/hvm/params.h>
58 #include <xen/interface/physdev.h>
59 #include <xen/interface/sched.h>
60 #include <xen/interface/vcpu.h>
61 #include <asm/hw_irq.h>
63 #include "events_internal.h"
65 const struct evtchn_ops
*evtchn_ops
;
68 * This lock protects updates to the following mapping and reference-count
69 * arrays. The lock does not need to be acquired to read the mapping tables.
71 static DEFINE_MUTEX(irq_mapping_update_lock
);
73 static LIST_HEAD(xen_irq_list_head
);
75 /* IRQ <-> VIRQ mapping. */
76 static DEFINE_PER_CPU(int [NR_VIRQS
], virq_to_irq
) = {[0 ... NR_VIRQS
-1] = -1};
78 /* IRQ <-> IPI mapping */
79 static DEFINE_PER_CPU(int [XEN_NR_IPIS
], ipi_to_irq
) = {[0 ... XEN_NR_IPIS
-1] = -1};
83 static unsigned long *pirq_eoi_map
;
85 static bool (*pirq_needs_eoi
)(unsigned irq
);
87 #define EVTCHN_ROW(e) (e / (PAGE_SIZE/sizeof(**evtchn_to_irq)))
88 #define EVTCHN_COL(e) (e % (PAGE_SIZE/sizeof(**evtchn_to_irq)))
89 #define EVTCHN_PER_ROW (PAGE_SIZE / sizeof(**evtchn_to_irq))
91 /* Xen will never allocate port zero for any purpose. */
92 #define VALID_EVTCHN(chn) ((chn) != 0)
94 static struct irq_chip xen_dynamic_chip
;
95 static struct irq_chip xen_percpu_chip
;
96 static struct irq_chip xen_pirq_chip
;
97 static void enable_dynirq(struct irq_data
*data
);
98 static void disable_dynirq(struct irq_data
*data
);
100 static void clear_evtchn_to_irq_row(unsigned row
)
104 for (col
= 0; col
< EVTCHN_PER_ROW
; col
++)
105 evtchn_to_irq
[row
][col
] = -1;
108 static void clear_evtchn_to_irq_all(void)
112 for (row
= 0; row
< EVTCHN_ROW(xen_evtchn_max_channels()); row
++) {
113 if (evtchn_to_irq
[row
] == NULL
)
115 clear_evtchn_to_irq_row(row
);
119 static int set_evtchn_to_irq(unsigned evtchn
, unsigned irq
)
124 if (evtchn
>= xen_evtchn_max_channels())
127 row
= EVTCHN_ROW(evtchn
);
128 col
= EVTCHN_COL(evtchn
);
130 if (evtchn_to_irq
[row
] == NULL
) {
131 /* Unallocated irq entries return -1 anyway */
135 evtchn_to_irq
[row
] = (int *)get_zeroed_page(GFP_KERNEL
);
136 if (evtchn_to_irq
[row
] == NULL
)
139 clear_evtchn_to_irq_row(row
);
142 evtchn_to_irq
[EVTCHN_ROW(evtchn
)][EVTCHN_COL(evtchn
)] = irq
;
146 int get_evtchn_to_irq(unsigned evtchn
)
148 if (evtchn
>= xen_evtchn_max_channels())
150 if (evtchn_to_irq
[EVTCHN_ROW(evtchn
)] == NULL
)
152 return evtchn_to_irq
[EVTCHN_ROW(evtchn
)][EVTCHN_COL(evtchn
)];
155 /* Get info for IRQ */
156 struct irq_info
*info_for_irq(unsigned irq
)
158 return irq_get_handler_data(irq
);
161 /* Constructors for packed IRQ information. */
162 static int xen_irq_info_common_setup(struct irq_info
*info
,
164 enum xen_irq_type type
,
170 BUG_ON(info
->type
!= IRQT_UNBOUND
&& info
->type
!= type
);
174 info
->evtchn
= evtchn
;
177 ret
= set_evtchn_to_irq(evtchn
, irq
);
181 irq_clear_status_flags(irq
, IRQ_NOREQUEST
|IRQ_NOAUTOEN
);
183 return xen_evtchn_port_setup(info
);
186 static int xen_irq_info_evtchn_setup(unsigned irq
,
189 struct irq_info
*info
= info_for_irq(irq
);
191 return xen_irq_info_common_setup(info
, irq
, IRQT_EVTCHN
, evtchn
, 0);
194 static int xen_irq_info_ipi_setup(unsigned cpu
,
199 struct irq_info
*info
= info_for_irq(irq
);
203 per_cpu(ipi_to_irq
, cpu
)[ipi
] = irq
;
205 return xen_irq_info_common_setup(info
, irq
, IRQT_IPI
, evtchn
, 0);
208 static int xen_irq_info_virq_setup(unsigned cpu
,
213 struct irq_info
*info
= info_for_irq(irq
);
217 per_cpu(virq_to_irq
, cpu
)[virq
] = irq
;
219 return xen_irq_info_common_setup(info
, irq
, IRQT_VIRQ
, evtchn
, 0);
222 static int xen_irq_info_pirq_setup(unsigned irq
,
229 struct irq_info
*info
= info_for_irq(irq
);
231 info
->u
.pirq
.pirq
= pirq
;
232 info
->u
.pirq
.gsi
= gsi
;
233 info
->u
.pirq
.domid
= domid
;
234 info
->u
.pirq
.flags
= flags
;
236 return xen_irq_info_common_setup(info
, irq
, IRQT_PIRQ
, evtchn
, 0);
239 static void xen_irq_info_cleanup(struct irq_info
*info
)
241 set_evtchn_to_irq(info
->evtchn
, -1);
246 * Accessors for packed IRQ information.
248 unsigned int evtchn_from_irq(unsigned irq
)
250 if (unlikely(WARN(irq
>= nr_irqs
, "Invalid irq %d!\n", irq
)))
253 return info_for_irq(irq
)->evtchn
;
256 unsigned irq_from_evtchn(unsigned int evtchn
)
258 return get_evtchn_to_irq(evtchn
);
260 EXPORT_SYMBOL_GPL(irq_from_evtchn
);
262 int irq_from_virq(unsigned int cpu
, unsigned int virq
)
264 return per_cpu(virq_to_irq
, cpu
)[virq
];
267 static enum ipi_vector
ipi_from_irq(unsigned irq
)
269 struct irq_info
*info
= info_for_irq(irq
);
271 BUG_ON(info
== NULL
);
272 BUG_ON(info
->type
!= IRQT_IPI
);
277 static unsigned virq_from_irq(unsigned irq
)
279 struct irq_info
*info
= info_for_irq(irq
);
281 BUG_ON(info
== NULL
);
282 BUG_ON(info
->type
!= IRQT_VIRQ
);
287 static unsigned pirq_from_irq(unsigned irq
)
289 struct irq_info
*info
= info_for_irq(irq
);
291 BUG_ON(info
== NULL
);
292 BUG_ON(info
->type
!= IRQT_PIRQ
);
294 return info
->u
.pirq
.pirq
;
297 static enum xen_irq_type
type_from_irq(unsigned irq
)
299 return info_for_irq(irq
)->type
;
302 unsigned cpu_from_irq(unsigned irq
)
304 return info_for_irq(irq
)->cpu
;
307 unsigned int cpu_from_evtchn(unsigned int evtchn
)
309 int irq
= get_evtchn_to_irq(evtchn
);
313 ret
= cpu_from_irq(irq
);
319 static bool pirq_check_eoi_map(unsigned irq
)
321 return test_bit(pirq_from_irq(irq
), pirq_eoi_map
);
325 static bool pirq_needs_eoi_flag(unsigned irq
)
327 struct irq_info
*info
= info_for_irq(irq
);
328 BUG_ON(info
->type
!= IRQT_PIRQ
);
330 return info
->u
.pirq
.flags
& PIRQ_NEEDS_EOI
;
333 static void bind_evtchn_to_cpu(unsigned int chn
, unsigned int cpu
)
335 int irq
= get_evtchn_to_irq(chn
);
336 struct irq_info
*info
= info_for_irq(irq
);
340 cpumask_copy(irq_get_affinity_mask(irq
), cpumask_of(cpu
));
342 xen_evtchn_port_bind_to_cpu(info
, cpu
);
347 static void xen_evtchn_mask_all(void)
351 for (evtchn
= 0; evtchn
< xen_evtchn_nr_channels(); evtchn
++)
356 * notify_remote_via_irq - send event to remote end of event channel via irq
357 * @irq: irq of event channel to send event to
359 * Unlike notify_remote_via_evtchn(), this is safe to use across
360 * save/restore. Notifications on a broken connection are silently
363 void notify_remote_via_irq(int irq
)
365 int evtchn
= evtchn_from_irq(irq
);
367 if (VALID_EVTCHN(evtchn
))
368 notify_remote_via_evtchn(evtchn
);
370 EXPORT_SYMBOL_GPL(notify_remote_via_irq
);
372 static void xen_irq_init(unsigned irq
)
374 struct irq_info
*info
;
376 /* By default all event channels notify CPU#0. */
377 cpumask_copy(irq_get_affinity_mask(irq
), cpumask_of(0));
380 info
= kzalloc(sizeof(*info
), GFP_KERNEL
);
382 panic("Unable to allocate metadata for IRQ%d\n", irq
);
384 info
->type
= IRQT_UNBOUND
;
387 irq_set_handler_data(irq
, info
);
389 list_add_tail(&info
->list
, &xen_irq_list_head
);
392 static int __must_check
xen_allocate_irqs_dynamic(int nvec
)
394 int i
, irq
= irq_alloc_descs(-1, 0, nvec
, -1);
397 for (i
= 0; i
< nvec
; i
++)
398 xen_irq_init(irq
+ i
);
404 static inline int __must_check
xen_allocate_irq_dynamic(void)
407 return xen_allocate_irqs_dynamic(1);
410 static int __must_check
xen_allocate_irq_gsi(unsigned gsi
)
415 * A PV guest has no concept of a GSI (since it has no ACPI
416 * nor access to/knowledge of the physical APICs). Therefore
417 * all IRQs are dynamically allocated from the entire IRQ
420 if (xen_pv_domain() && !xen_initial_domain())
421 return xen_allocate_irq_dynamic();
423 /* Legacy IRQ descriptors are already allocated by the arch. */
424 if (gsi
< nr_legacy_irqs())
427 irq
= irq_alloc_desc_at(gsi
, -1);
434 static void xen_free_irq(unsigned irq
)
436 struct irq_info
*info
= irq_get_handler_data(irq
);
441 list_del(&info
->list
);
443 irq_set_handler_data(irq
, NULL
);
445 WARN_ON(info
->refcnt
> 0);
449 /* Legacy IRQ descriptors are managed by the arch. */
450 if (irq
< nr_legacy_irqs())
456 static void xen_evtchn_close(unsigned int port
)
458 struct evtchn_close close
;
461 if (HYPERVISOR_event_channel_op(EVTCHNOP_close
, &close
) != 0)
465 static void pirq_query_unmask(int irq
)
467 struct physdev_irq_status_query irq_status
;
468 struct irq_info
*info
= info_for_irq(irq
);
470 BUG_ON(info
->type
!= IRQT_PIRQ
);
472 irq_status
.irq
= pirq_from_irq(irq
);
473 if (HYPERVISOR_physdev_op(PHYSDEVOP_irq_status_query
, &irq_status
))
474 irq_status
.flags
= 0;
476 info
->u
.pirq
.flags
&= ~PIRQ_NEEDS_EOI
;
477 if (irq_status
.flags
& XENIRQSTAT_needs_eoi
)
478 info
->u
.pirq
.flags
|= PIRQ_NEEDS_EOI
;
481 static void eoi_pirq(struct irq_data
*data
)
483 int evtchn
= evtchn_from_irq(data
->irq
);
484 struct physdev_eoi eoi
= { .irq
= pirq_from_irq(data
->irq
) };
487 if (!VALID_EVTCHN(evtchn
))
490 if (unlikely(irqd_is_setaffinity_pending(data
))) {
491 int masked
= test_and_set_mask(evtchn
);
493 clear_evtchn(evtchn
);
495 irq_move_masked_irq(data
);
498 unmask_evtchn(evtchn
);
500 clear_evtchn(evtchn
);
502 if (pirq_needs_eoi(data
->irq
)) {
503 rc
= HYPERVISOR_physdev_op(PHYSDEVOP_eoi
, &eoi
);
508 static void mask_ack_pirq(struct irq_data
*data
)
510 disable_dynirq(data
);
514 static unsigned int __startup_pirq(unsigned int irq
)
516 struct evtchn_bind_pirq bind_pirq
;
517 struct irq_info
*info
= info_for_irq(irq
);
518 int evtchn
= evtchn_from_irq(irq
);
521 BUG_ON(info
->type
!= IRQT_PIRQ
);
523 if (VALID_EVTCHN(evtchn
))
526 bind_pirq
.pirq
= pirq_from_irq(irq
);
527 /* NB. We are happy to share unless we are probing. */
528 bind_pirq
.flags
= info
->u
.pirq
.flags
& PIRQ_SHAREABLE
?
529 BIND_PIRQ__WILL_SHARE
: 0;
530 rc
= HYPERVISOR_event_channel_op(EVTCHNOP_bind_pirq
, &bind_pirq
);
532 pr_warn("Failed to obtain physical IRQ %d\n", irq
);
535 evtchn
= bind_pirq
.port
;
537 pirq_query_unmask(irq
);
539 rc
= set_evtchn_to_irq(evtchn
, irq
);
543 info
->evtchn
= evtchn
;
544 bind_evtchn_to_cpu(evtchn
, 0);
546 rc
= xen_evtchn_port_setup(info
);
551 unmask_evtchn(evtchn
);
552 eoi_pirq(irq_get_irq_data(irq
));
557 pr_err("irq%d: Failed to set port to irq mapping (%d)\n", irq
, rc
);
558 xen_evtchn_close(evtchn
);
562 static unsigned int startup_pirq(struct irq_data
*data
)
564 return __startup_pirq(data
->irq
);
567 static void shutdown_pirq(struct irq_data
*data
)
569 unsigned int irq
= data
->irq
;
570 struct irq_info
*info
= info_for_irq(irq
);
571 unsigned evtchn
= evtchn_from_irq(irq
);
573 BUG_ON(info
->type
!= IRQT_PIRQ
);
575 if (!VALID_EVTCHN(evtchn
))
579 xen_evtchn_close(evtchn
);
580 xen_irq_info_cleanup(info
);
583 static void enable_pirq(struct irq_data
*data
)
588 static void disable_pirq(struct irq_data
*data
)
590 disable_dynirq(data
);
593 int xen_irq_from_gsi(unsigned gsi
)
595 struct irq_info
*info
;
597 list_for_each_entry(info
, &xen_irq_list_head
, list
) {
598 if (info
->type
!= IRQT_PIRQ
)
601 if (info
->u
.pirq
.gsi
== gsi
)
607 EXPORT_SYMBOL_GPL(xen_irq_from_gsi
);
609 static void __unbind_from_irq(unsigned int irq
)
611 int evtchn
= evtchn_from_irq(irq
);
612 struct irq_info
*info
= irq_get_handler_data(irq
);
614 if (info
->refcnt
> 0) {
616 if (info
->refcnt
!= 0)
620 if (VALID_EVTCHN(evtchn
)) {
621 unsigned int cpu
= cpu_from_irq(irq
);
623 xen_evtchn_close(evtchn
);
625 switch (type_from_irq(irq
)) {
627 per_cpu(virq_to_irq
, cpu
)[virq_from_irq(irq
)] = -1;
630 per_cpu(ipi_to_irq
, cpu
)[ipi_from_irq(irq
)] = -1;
636 xen_irq_info_cleanup(info
);
639 BUG_ON(info_for_irq(irq
)->type
== IRQT_UNBOUND
);
645 * Do not make any assumptions regarding the relationship between the
646 * IRQ number returned here and the Xen pirq argument.
648 * Note: We don't assign an event channel until the irq actually started
649 * up. Return an existing irq if we've already got one for the gsi.
651 * Shareable implies level triggered, not shareable implies edge
654 int xen_bind_pirq_gsi_to_irq(unsigned gsi
,
655 unsigned pirq
, int shareable
, char *name
)
658 struct physdev_irq irq_op
;
661 mutex_lock(&irq_mapping_update_lock
);
663 irq
= xen_irq_from_gsi(gsi
);
665 pr_info("%s: returning irq %d for gsi %u\n",
670 irq
= xen_allocate_irq_gsi(gsi
);
677 /* Only the privileged domain can do this. For non-priv, the pcifront
678 * driver provides a PCI bus that does the call to do exactly
679 * this in the priv domain. */
680 if (xen_initial_domain() &&
681 HYPERVISOR_physdev_op(PHYSDEVOP_alloc_irq_vector
, &irq_op
)) {
687 ret
= xen_irq_info_pirq_setup(irq
, 0, pirq
, gsi
, DOMID_SELF
,
688 shareable
? PIRQ_SHAREABLE
: 0);
690 __unbind_from_irq(irq
);
695 pirq_query_unmask(irq
);
696 /* We try to use the handler with the appropriate semantic for the
697 * type of interrupt: if the interrupt is an edge triggered
698 * interrupt we use handle_edge_irq.
700 * On the other hand if the interrupt is level triggered we use
701 * handle_fasteoi_irq like the native code does for this kind of
704 * Depending on the Xen version, pirq_needs_eoi might return true
705 * not only for level triggered interrupts but for edge triggered
706 * interrupts too. In any case Xen always honors the eoi mechanism,
707 * not injecting any more pirqs of the same kind if the first one
708 * hasn't received an eoi yet. Therefore using the fasteoi handler
709 * is the right choice either way.
712 irq_set_chip_and_handler_name(irq
, &xen_pirq_chip
,
713 handle_fasteoi_irq
, name
);
715 irq_set_chip_and_handler_name(irq
, &xen_pirq_chip
,
716 handle_edge_irq
, name
);
719 mutex_unlock(&irq_mapping_update_lock
);
724 #ifdef CONFIG_PCI_MSI
725 int xen_allocate_pirq_msi(struct pci_dev
*dev
, struct msi_desc
*msidesc
)
728 struct physdev_get_free_pirq op_get_free_pirq
;
730 op_get_free_pirq
.type
= MAP_PIRQ_TYPE_MSI
;
731 rc
= HYPERVISOR_physdev_op(PHYSDEVOP_get_free_pirq
, &op_get_free_pirq
);
733 WARN_ONCE(rc
== -ENOSYS
,
734 "hypervisor does not support the PHYSDEVOP_get_free_pirq interface\n");
736 return rc
? -1 : op_get_free_pirq
.pirq
;
739 int xen_bind_pirq_msi_to_irq(struct pci_dev
*dev
, struct msi_desc
*msidesc
,
740 int pirq
, int nvec
, const char *name
, domid_t domid
)
744 mutex_lock(&irq_mapping_update_lock
);
746 irq
= xen_allocate_irqs_dynamic(nvec
);
750 for (i
= 0; i
< nvec
; i
++) {
751 irq_set_chip_and_handler_name(irq
+ i
, &xen_pirq_chip
, handle_edge_irq
, name
);
753 ret
= xen_irq_info_pirq_setup(irq
+ i
, 0, pirq
+ i
, 0, domid
,
754 i
== 0 ? 0 : PIRQ_MSI_GROUP
);
759 ret
= irq_set_msi_desc(irq
, msidesc
);
763 mutex_unlock(&irq_mapping_update_lock
);
767 __unbind_from_irq(irq
+ i
);
768 mutex_unlock(&irq_mapping_update_lock
);
773 int xen_destroy_irq(int irq
)
775 struct physdev_unmap_pirq unmap_irq
;
776 struct irq_info
*info
= info_for_irq(irq
);
779 mutex_lock(&irq_mapping_update_lock
);
782 * If trying to remove a vector in a MSI group different
783 * than the first one skip the PIRQ unmap unless this vector
784 * is the first one in the group.
786 if (xen_initial_domain() && !(info
->u
.pirq
.flags
& PIRQ_MSI_GROUP
)) {
787 unmap_irq
.pirq
= info
->u
.pirq
.pirq
;
788 unmap_irq
.domid
= info
->u
.pirq
.domid
;
789 rc
= HYPERVISOR_physdev_op(PHYSDEVOP_unmap_pirq
, &unmap_irq
);
790 /* If another domain quits without making the pci_disable_msix
791 * call, the Xen hypervisor takes care of freeing the PIRQs
792 * (free_domain_pirqs).
794 if ((rc
== -ESRCH
&& info
->u
.pirq
.domid
!= DOMID_SELF
))
795 pr_info("domain %d does not have %d anymore\n",
796 info
->u
.pirq
.domid
, info
->u
.pirq
.pirq
);
798 pr_warn("unmap irq failed %d\n", rc
);
806 mutex_unlock(&irq_mapping_update_lock
);
810 int xen_irq_from_pirq(unsigned pirq
)
814 struct irq_info
*info
;
816 mutex_lock(&irq_mapping_update_lock
);
818 list_for_each_entry(info
, &xen_irq_list_head
, list
) {
819 if (info
->type
!= IRQT_PIRQ
)
822 if (info
->u
.pirq
.pirq
== pirq
)
827 mutex_unlock(&irq_mapping_update_lock
);
833 int xen_pirq_from_irq(unsigned irq
)
835 return pirq_from_irq(irq
);
837 EXPORT_SYMBOL_GPL(xen_pirq_from_irq
);
839 int bind_evtchn_to_irq(unsigned int evtchn
)
844 if (evtchn
>= xen_evtchn_max_channels())
847 mutex_lock(&irq_mapping_update_lock
);
849 irq
= get_evtchn_to_irq(evtchn
);
852 irq
= xen_allocate_irq_dynamic();
856 irq_set_chip_and_handler_name(irq
, &xen_dynamic_chip
,
857 handle_edge_irq
, "event");
859 ret
= xen_irq_info_evtchn_setup(irq
, evtchn
);
861 __unbind_from_irq(irq
);
865 /* New interdomain events are bound to VCPU 0. */
866 bind_evtchn_to_cpu(evtchn
, 0);
868 struct irq_info
*info
= info_for_irq(irq
);
869 WARN_ON(info
== NULL
|| info
->type
!= IRQT_EVTCHN
);
873 mutex_unlock(&irq_mapping_update_lock
);
877 EXPORT_SYMBOL_GPL(bind_evtchn_to_irq
);
879 static int bind_ipi_to_irq(unsigned int ipi
, unsigned int cpu
)
881 struct evtchn_bind_ipi bind_ipi
;
885 mutex_lock(&irq_mapping_update_lock
);
887 irq
= per_cpu(ipi_to_irq
, cpu
)[ipi
];
890 irq
= xen_allocate_irq_dynamic();
894 irq_set_chip_and_handler_name(irq
, &xen_percpu_chip
,
895 handle_percpu_irq
, "ipi");
898 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_ipi
,
901 evtchn
= bind_ipi
.port
;
903 ret
= xen_irq_info_ipi_setup(cpu
, irq
, evtchn
, ipi
);
905 __unbind_from_irq(irq
);
909 bind_evtchn_to_cpu(evtchn
, cpu
);
911 struct irq_info
*info
= info_for_irq(irq
);
912 WARN_ON(info
== NULL
|| info
->type
!= IRQT_IPI
);
916 mutex_unlock(&irq_mapping_update_lock
);
920 int bind_interdomain_evtchn_to_irq(unsigned int remote_domain
,
921 unsigned int remote_port
)
923 struct evtchn_bind_interdomain bind_interdomain
;
926 bind_interdomain
.remote_dom
= remote_domain
;
927 bind_interdomain
.remote_port
= remote_port
;
929 err
= HYPERVISOR_event_channel_op(EVTCHNOP_bind_interdomain
,
932 return err
? : bind_evtchn_to_irq(bind_interdomain
.local_port
);
934 EXPORT_SYMBOL_GPL(bind_interdomain_evtchn_to_irq
);
936 static int find_virq(unsigned int virq
, unsigned int cpu
)
938 struct evtchn_status status
;
939 int port
, rc
= -ENOENT
;
941 memset(&status
, 0, sizeof(status
));
942 for (port
= 0; port
< xen_evtchn_max_channels(); port
++) {
943 status
.dom
= DOMID_SELF
;
945 rc
= HYPERVISOR_event_channel_op(EVTCHNOP_status
, &status
);
948 if (status
.status
!= EVTCHNSTAT_virq
)
950 if (status
.u
.virq
== virq
&& status
.vcpu
== cpu
) {
959 * xen_evtchn_nr_channels - number of usable event channel ports
961 * This may be less than the maximum supported by the current
962 * hypervisor ABI. Use xen_evtchn_max_channels() for the maximum
965 unsigned xen_evtchn_nr_channels(void)
967 return evtchn_ops
->nr_channels();
969 EXPORT_SYMBOL_GPL(xen_evtchn_nr_channels
);
971 int bind_virq_to_irq(unsigned int virq
, unsigned int cpu
, bool percpu
)
973 struct evtchn_bind_virq bind_virq
;
974 int evtchn
, irq
, ret
;
976 mutex_lock(&irq_mapping_update_lock
);
978 irq
= per_cpu(virq_to_irq
, cpu
)[virq
];
981 irq
= xen_allocate_irq_dynamic();
986 irq_set_chip_and_handler_name(irq
, &xen_percpu_chip
,
987 handle_percpu_irq
, "virq");
989 irq_set_chip_and_handler_name(irq
, &xen_dynamic_chip
,
990 handle_edge_irq
, "virq");
992 bind_virq
.virq
= virq
;
993 bind_virq
.vcpu
= cpu
;
994 ret
= HYPERVISOR_event_channel_op(EVTCHNOP_bind_virq
,
997 evtchn
= bind_virq
.port
;
1000 ret
= find_virq(virq
, cpu
);
1005 ret
= xen_irq_info_virq_setup(cpu
, irq
, evtchn
, virq
);
1007 __unbind_from_irq(irq
);
1012 bind_evtchn_to_cpu(evtchn
, cpu
);
1014 struct irq_info
*info
= info_for_irq(irq
);
1015 WARN_ON(info
== NULL
|| info
->type
!= IRQT_VIRQ
);
1019 mutex_unlock(&irq_mapping_update_lock
);
1024 static void unbind_from_irq(unsigned int irq
)
1026 mutex_lock(&irq_mapping_update_lock
);
1027 __unbind_from_irq(irq
);
1028 mutex_unlock(&irq_mapping_update_lock
);
1031 int bind_evtchn_to_irqhandler(unsigned int evtchn
,
1032 irq_handler_t handler
,
1033 unsigned long irqflags
,
1034 const char *devname
, void *dev_id
)
1038 irq
= bind_evtchn_to_irq(evtchn
);
1041 retval
= request_irq(irq
, handler
, irqflags
, devname
, dev_id
);
1043 unbind_from_irq(irq
);
1049 EXPORT_SYMBOL_GPL(bind_evtchn_to_irqhandler
);
1051 int bind_interdomain_evtchn_to_irqhandler(unsigned int remote_domain
,
1052 unsigned int remote_port
,
1053 irq_handler_t handler
,
1054 unsigned long irqflags
,
1055 const char *devname
,
1060 irq
= bind_interdomain_evtchn_to_irq(remote_domain
, remote_port
);
1064 retval
= request_irq(irq
, handler
, irqflags
, devname
, dev_id
);
1066 unbind_from_irq(irq
);
1072 EXPORT_SYMBOL_GPL(bind_interdomain_evtchn_to_irqhandler
);
1074 int bind_virq_to_irqhandler(unsigned int virq
, unsigned int cpu
,
1075 irq_handler_t handler
,
1076 unsigned long irqflags
, const char *devname
, void *dev_id
)
1080 irq
= bind_virq_to_irq(virq
, cpu
, irqflags
& IRQF_PERCPU
);
1083 retval
= request_irq(irq
, handler
, irqflags
, devname
, dev_id
);
1085 unbind_from_irq(irq
);
1091 EXPORT_SYMBOL_GPL(bind_virq_to_irqhandler
);
1093 int bind_ipi_to_irqhandler(enum ipi_vector ipi
,
1095 irq_handler_t handler
,
1096 unsigned long irqflags
,
1097 const char *devname
,
1102 irq
= bind_ipi_to_irq(ipi
, cpu
);
1106 irqflags
|= IRQF_NO_SUSPEND
| IRQF_FORCE_RESUME
| IRQF_EARLY_RESUME
;
1107 retval
= request_irq(irq
, handler
, irqflags
, devname
, dev_id
);
1109 unbind_from_irq(irq
);
1116 void unbind_from_irqhandler(unsigned int irq
, void *dev_id
)
1118 struct irq_info
*info
= irq_get_handler_data(irq
);
1122 free_irq(irq
, dev_id
);
1123 unbind_from_irq(irq
);
1125 EXPORT_SYMBOL_GPL(unbind_from_irqhandler
);
1128 * xen_set_irq_priority() - set an event channel priority.
1129 * @irq:irq bound to an event channel.
1130 * @priority: priority between XEN_IRQ_PRIORITY_MAX and XEN_IRQ_PRIORITY_MIN.
1132 int xen_set_irq_priority(unsigned irq
, unsigned priority
)
1134 struct evtchn_set_priority set_priority
;
1136 set_priority
.port
= evtchn_from_irq(irq
);
1137 set_priority
.priority
= priority
;
1139 return HYPERVISOR_event_channel_op(EVTCHNOP_set_priority
,
1142 EXPORT_SYMBOL_GPL(xen_set_irq_priority
);
1144 int evtchn_make_refcounted(unsigned int evtchn
)
1146 int irq
= get_evtchn_to_irq(evtchn
);
1147 struct irq_info
*info
;
1152 info
= irq_get_handler_data(irq
);
1157 WARN_ON(info
->refcnt
!= -1);
1163 EXPORT_SYMBOL_GPL(evtchn_make_refcounted
);
1165 int evtchn_get(unsigned int evtchn
)
1168 struct irq_info
*info
;
1171 if (evtchn
>= xen_evtchn_max_channels())
1174 mutex_lock(&irq_mapping_update_lock
);
1176 irq
= get_evtchn_to_irq(evtchn
);
1180 info
= irq_get_handler_data(irq
);
1186 if (info
->refcnt
<= 0)
1192 mutex_unlock(&irq_mapping_update_lock
);
1196 EXPORT_SYMBOL_GPL(evtchn_get
);
1198 void evtchn_put(unsigned int evtchn
)
1200 int irq
= get_evtchn_to_irq(evtchn
);
1201 if (WARN_ON(irq
== -1))
1203 unbind_from_irq(irq
);
1205 EXPORT_SYMBOL_GPL(evtchn_put
);
1207 void xen_send_IPI_one(unsigned int cpu
, enum ipi_vector vector
)
1212 if (unlikely(vector
== XEN_NMI_VECTOR
)) {
1213 int rc
= HYPERVISOR_vcpu_op(VCPUOP_send_nmi
, cpu
, NULL
);
1215 printk(KERN_WARNING
"Sending nmi to CPU%d failed (rc:%d)\n", cpu
, rc
);
1219 irq
= per_cpu(ipi_to_irq
, cpu
)[vector
];
1221 notify_remote_via_irq(irq
);
1224 static DEFINE_PER_CPU(unsigned, xed_nesting_count
);
1226 static void __xen_evtchn_do_upcall(void)
1228 struct vcpu_info
*vcpu_info
= __this_cpu_read(xen_vcpu
);
1229 int cpu
= get_cpu();
1233 vcpu_info
->evtchn_upcall_pending
= 0;
1235 if (__this_cpu_inc_return(xed_nesting_count
) - 1)
1238 xen_evtchn_handle_events(cpu
);
1240 BUG_ON(!irqs_disabled());
1242 count
= __this_cpu_read(xed_nesting_count
);
1243 __this_cpu_write(xed_nesting_count
, 0);
1244 } while (count
!= 1 || vcpu_info
->evtchn_upcall_pending
);
1251 void xen_evtchn_do_upcall(struct pt_regs
*regs
)
1253 struct pt_regs
*old_regs
= set_irq_regs(regs
);
1258 inc_irq_stat(irq_hv_callback_count
);
1261 __xen_evtchn_do_upcall();
1264 set_irq_regs(old_regs
);
1267 void xen_hvm_evtchn_do_upcall(void)
1269 __xen_evtchn_do_upcall();
1271 EXPORT_SYMBOL_GPL(xen_hvm_evtchn_do_upcall
);
1273 /* Rebind a new event channel to an existing irq. */
1274 void rebind_evtchn_irq(int evtchn
, int irq
)
1276 struct irq_info
*info
= info_for_irq(irq
);
1281 /* Make sure the irq is masked, since the new event channel
1282 will also be masked. */
1285 mutex_lock(&irq_mapping_update_lock
);
1287 /* After resume the irq<->evtchn mappings are all cleared out */
1288 BUG_ON(get_evtchn_to_irq(evtchn
) != -1);
1289 /* Expect irq to have been bound before,
1290 so there should be a proper type */
1291 BUG_ON(info
->type
== IRQT_UNBOUND
);
1293 (void)xen_irq_info_evtchn_setup(irq
, evtchn
);
1295 mutex_unlock(&irq_mapping_update_lock
);
1297 bind_evtchn_to_cpu(evtchn
, info
->cpu
);
1298 /* This will be deferred until interrupt is processed */
1299 irq_set_affinity(irq
, cpumask_of(info
->cpu
));
1301 /* Unmask the event channel. */
1305 /* Rebind an evtchn so that it gets delivered to a specific cpu */
1306 static int rebind_irq_to_cpu(unsigned irq
, unsigned tcpu
)
1308 struct evtchn_bind_vcpu bind_vcpu
;
1309 int evtchn
= evtchn_from_irq(irq
);
1312 if (!VALID_EVTCHN(evtchn
))
1315 if (!xen_support_evtchn_rebind())
1318 /* Send future instances of this interrupt to other vcpu. */
1319 bind_vcpu
.port
= evtchn
;
1320 bind_vcpu
.vcpu
= tcpu
;
1323 * Mask the event while changing the VCPU binding to prevent
1324 * it being delivered on an unexpected VCPU.
1326 masked
= test_and_set_mask(evtchn
);
1329 * If this fails, it usually just indicates that we're dealing with a
1330 * virq or IPI channel, which don't actually need to be rebound. Ignore
1331 * it, but don't do the xenlinux-level rebind in that case.
1333 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_vcpu
, &bind_vcpu
) >= 0)
1334 bind_evtchn_to_cpu(evtchn
, tcpu
);
1337 unmask_evtchn(evtchn
);
1342 static int set_affinity_irq(struct irq_data
*data
, const struct cpumask
*dest
,
1345 unsigned tcpu
= cpumask_first_and(dest
, cpu_online_mask
);
1347 return rebind_irq_to_cpu(data
->irq
, tcpu
);
1350 static void enable_dynirq(struct irq_data
*data
)
1352 int evtchn
= evtchn_from_irq(data
->irq
);
1354 if (VALID_EVTCHN(evtchn
))
1355 unmask_evtchn(evtchn
);
1358 static void disable_dynirq(struct irq_data
*data
)
1360 int evtchn
= evtchn_from_irq(data
->irq
);
1362 if (VALID_EVTCHN(evtchn
))
1363 mask_evtchn(evtchn
);
1366 static void ack_dynirq(struct irq_data
*data
)
1368 int evtchn
= evtchn_from_irq(data
->irq
);
1370 if (!VALID_EVTCHN(evtchn
))
1373 if (unlikely(irqd_is_setaffinity_pending(data
))) {
1374 int masked
= test_and_set_mask(evtchn
);
1376 clear_evtchn(evtchn
);
1378 irq_move_masked_irq(data
);
1381 unmask_evtchn(evtchn
);
1383 clear_evtchn(evtchn
);
1386 static void mask_ack_dynirq(struct irq_data
*data
)
1388 disable_dynirq(data
);
1392 static int retrigger_dynirq(struct irq_data
*data
)
1394 unsigned int evtchn
= evtchn_from_irq(data
->irq
);
1397 if (!VALID_EVTCHN(evtchn
))
1400 masked
= test_and_set_mask(evtchn
);
1403 unmask_evtchn(evtchn
);
1408 static void restore_pirqs(void)
1410 int pirq
, rc
, irq
, gsi
;
1411 struct physdev_map_pirq map_irq
;
1412 struct irq_info
*info
;
1414 list_for_each_entry(info
, &xen_irq_list_head
, list
) {
1415 if (info
->type
!= IRQT_PIRQ
)
1418 pirq
= info
->u
.pirq
.pirq
;
1419 gsi
= info
->u
.pirq
.gsi
;
1422 /* save/restore of PT devices doesn't work, so at this point the
1423 * only devices present are GSI based emulated devices */
1427 map_irq
.domid
= DOMID_SELF
;
1428 map_irq
.type
= MAP_PIRQ_TYPE_GSI
;
1429 map_irq
.index
= gsi
;
1430 map_irq
.pirq
= pirq
;
1432 rc
= HYPERVISOR_physdev_op(PHYSDEVOP_map_pirq
, &map_irq
);
1434 pr_warn("xen map irq failed gsi=%d irq=%d pirq=%d rc=%d\n",
1435 gsi
, irq
, pirq
, rc
);
1440 printk(KERN_DEBUG
"xen: --> irq=%d, pirq=%d\n", irq
, map_irq
.pirq
);
1442 __startup_pirq(irq
);
1446 static void restore_cpu_virqs(unsigned int cpu
)
1448 struct evtchn_bind_virq bind_virq
;
1449 int virq
, irq
, evtchn
;
1451 for (virq
= 0; virq
< NR_VIRQS
; virq
++) {
1452 if ((irq
= per_cpu(virq_to_irq
, cpu
)[virq
]) == -1)
1455 BUG_ON(virq_from_irq(irq
) != virq
);
1457 /* Get a new binding from Xen. */
1458 bind_virq
.virq
= virq
;
1459 bind_virq
.vcpu
= cpu
;
1460 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_virq
,
1463 evtchn
= bind_virq
.port
;
1465 /* Record the new mapping. */
1466 (void)xen_irq_info_virq_setup(cpu
, irq
, evtchn
, virq
);
1467 bind_evtchn_to_cpu(evtchn
, cpu
);
1471 static void restore_cpu_ipis(unsigned int cpu
)
1473 struct evtchn_bind_ipi bind_ipi
;
1474 int ipi
, irq
, evtchn
;
1476 for (ipi
= 0; ipi
< XEN_NR_IPIS
; ipi
++) {
1477 if ((irq
= per_cpu(ipi_to_irq
, cpu
)[ipi
]) == -1)
1480 BUG_ON(ipi_from_irq(irq
) != ipi
);
1482 /* Get a new binding from Xen. */
1483 bind_ipi
.vcpu
= cpu
;
1484 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_ipi
,
1487 evtchn
= bind_ipi
.port
;
1489 /* Record the new mapping. */
1490 (void)xen_irq_info_ipi_setup(cpu
, irq
, evtchn
, ipi
);
1491 bind_evtchn_to_cpu(evtchn
, cpu
);
1495 /* Clear an irq's pending state, in preparation for polling on it */
1496 void xen_clear_irq_pending(int irq
)
1498 int evtchn
= evtchn_from_irq(irq
);
1500 if (VALID_EVTCHN(evtchn
))
1501 clear_evtchn(evtchn
);
1503 EXPORT_SYMBOL(xen_clear_irq_pending
);
1504 void xen_set_irq_pending(int irq
)
1506 int evtchn
= evtchn_from_irq(irq
);
1508 if (VALID_EVTCHN(evtchn
))
1512 bool xen_test_irq_pending(int irq
)
1514 int evtchn
= evtchn_from_irq(irq
);
1517 if (VALID_EVTCHN(evtchn
))
1518 ret
= test_evtchn(evtchn
);
1523 /* Poll waiting for an irq to become pending with timeout. In the usual case,
1524 * the irq will be disabled so it won't deliver an interrupt. */
1525 void xen_poll_irq_timeout(int irq
, u64 timeout
)
1527 evtchn_port_t evtchn
= evtchn_from_irq(irq
);
1529 if (VALID_EVTCHN(evtchn
)) {
1530 struct sched_poll poll
;
1533 poll
.timeout
= timeout
;
1534 set_xen_guest_handle(poll
.ports
, &evtchn
);
1536 if (HYPERVISOR_sched_op(SCHEDOP_poll
, &poll
) != 0)
1540 EXPORT_SYMBOL(xen_poll_irq_timeout
);
1541 /* Poll waiting for an irq to become pending. In the usual case, the
1542 * irq will be disabled so it won't deliver an interrupt. */
1543 void xen_poll_irq(int irq
)
1545 xen_poll_irq_timeout(irq
, 0 /* no timeout */);
1548 /* Check whether the IRQ line is shared with other guests. */
1549 int xen_test_irq_shared(int irq
)
1551 struct irq_info
*info
= info_for_irq(irq
);
1552 struct physdev_irq_status_query irq_status
;
1557 irq_status
.irq
= info
->u
.pirq
.pirq
;
1559 if (HYPERVISOR_physdev_op(PHYSDEVOP_irq_status_query
, &irq_status
))
1561 return !(irq_status
.flags
& XENIRQSTAT_shared
);
1563 EXPORT_SYMBOL_GPL(xen_test_irq_shared
);
1565 void xen_irq_resume(void)
1568 struct irq_info
*info
;
1570 /* New event-channel space is not 'live' yet. */
1571 xen_evtchn_mask_all();
1572 xen_evtchn_resume();
1574 /* No IRQ <-> event-channel mappings. */
1575 list_for_each_entry(info
, &xen_irq_list_head
, list
)
1576 info
->evtchn
= 0; /* zap event-channel binding */
1578 clear_evtchn_to_irq_all();
1580 for_each_possible_cpu(cpu
) {
1581 restore_cpu_virqs(cpu
);
1582 restore_cpu_ipis(cpu
);
1588 static struct irq_chip xen_dynamic_chip __read_mostly
= {
1591 .irq_disable
= disable_dynirq
,
1592 .irq_mask
= disable_dynirq
,
1593 .irq_unmask
= enable_dynirq
,
1595 .irq_ack
= ack_dynirq
,
1596 .irq_mask_ack
= mask_ack_dynirq
,
1598 .irq_set_affinity
= set_affinity_irq
,
1599 .irq_retrigger
= retrigger_dynirq
,
1602 static struct irq_chip xen_pirq_chip __read_mostly
= {
1605 .irq_startup
= startup_pirq
,
1606 .irq_shutdown
= shutdown_pirq
,
1607 .irq_enable
= enable_pirq
,
1608 .irq_disable
= disable_pirq
,
1610 .irq_mask
= disable_dynirq
,
1611 .irq_unmask
= enable_dynirq
,
1613 .irq_ack
= eoi_pirq
,
1614 .irq_eoi
= eoi_pirq
,
1615 .irq_mask_ack
= mask_ack_pirq
,
1617 .irq_set_affinity
= set_affinity_irq
,
1619 .irq_retrigger
= retrigger_dynirq
,
1622 static struct irq_chip xen_percpu_chip __read_mostly
= {
1623 .name
= "xen-percpu",
1625 .irq_disable
= disable_dynirq
,
1626 .irq_mask
= disable_dynirq
,
1627 .irq_unmask
= enable_dynirq
,
1629 .irq_ack
= ack_dynirq
,
1632 int xen_set_callback_via(uint64_t via
)
1634 struct xen_hvm_param a
;
1635 a
.domid
= DOMID_SELF
;
1636 a
.index
= HVM_PARAM_CALLBACK_IRQ
;
1638 return HYPERVISOR_hvm_op(HVMOP_set_param
, &a
);
1640 EXPORT_SYMBOL_GPL(xen_set_callback_via
);
1642 #ifdef CONFIG_XEN_PVHVM
1643 /* Vector callbacks are better than PCI interrupts to receive event
1644 * channel notifications because we can receive vector callbacks on any
1645 * vcpu and we don't need PCI support or APIC interactions. */
1646 void xen_callback_vector(void)
1649 uint64_t callback_via
;
1650 if (xen_have_vector_callback
) {
1651 callback_via
= HVM_CALLBACK_VECTOR(HYPERVISOR_CALLBACK_VECTOR
);
1652 rc
= xen_set_callback_via(callback_via
);
1654 pr_err("Request for Xen HVM callback vector failed\n");
1655 xen_have_vector_callback
= 0;
1658 pr_info("Xen HVM callback vector for event delivery is enabled\n");
1659 /* in the restore case the vector has already been allocated */
1660 if (!test_bit(HYPERVISOR_CALLBACK_VECTOR
, used_vectors
))
1661 alloc_intr_gate(HYPERVISOR_CALLBACK_VECTOR
,
1662 xen_hvm_callback_vector
);
1666 void xen_callback_vector(void) {}
1669 #undef MODULE_PARAM_PREFIX
1670 #define MODULE_PARAM_PREFIX "xen."
1672 static bool fifo_events
= true;
1673 module_param(fifo_events
, bool, 0);
1675 void __init
xen_init_IRQ(void)
1680 ret
= xen_evtchn_fifo_init();
1682 xen_evtchn_2l_init();
1684 evtchn_to_irq
= kcalloc(EVTCHN_ROW(xen_evtchn_max_channels()),
1685 sizeof(*evtchn_to_irq
), GFP_KERNEL
);
1686 BUG_ON(!evtchn_to_irq
);
1688 /* No event channels are 'live' right now. */
1689 xen_evtchn_mask_all();
1691 pirq_needs_eoi
= pirq_needs_eoi_flag
;
1694 if (xen_pv_domain()) {
1695 irq_ctx_init(smp_processor_id());
1696 if (xen_initial_domain())
1697 pci_xen_initial_domain();
1699 if (xen_feature(XENFEAT_hvm_callback_vector
))
1700 xen_callback_vector();
1702 if (xen_hvm_domain()) {
1704 /* pci_xen_hvm_init must be called after native_init_IRQ so that
1705 * __acpi_register_gsi can point at the right function */
1709 struct physdev_pirq_eoi_gmfn eoi_gmfn
;
1711 pirq_eoi_map
= (void *)__get_free_page(GFP_KERNEL
|__GFP_ZERO
);
1712 eoi_gmfn
.gmfn
= virt_to_gfn(pirq_eoi_map
);
1713 rc
= HYPERVISOR_physdev_op(PHYSDEVOP_pirq_eoi_gmfn_v2
, &eoi_gmfn
);
1714 /* TODO: No PVH support for PIRQ EOI */
1716 free_page((unsigned long) pirq_eoi_map
);
1717 pirq_eoi_map
= NULL
;
1719 pirq_needs_eoi
= pirq_check_eoi_map
;