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>
40 #include <asm/io_apic.h>
41 #include <asm/i8259.h>
42 #include <asm/xen/pci.h>
44 #include <asm/sync_bitops.h>
45 #include <asm/xen/hypercall.h>
46 #include <asm/xen/hypervisor.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>
62 #include "events_internal.h"
64 const struct evtchn_ops
*evtchn_ops
;
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.
70 static DEFINE_MUTEX(irq_mapping_update_lock
);
72 static LIST_HEAD(xen_irq_list_head
);
74 /* IRQ <-> VIRQ mapping. */
75 static DEFINE_PER_CPU(int [NR_VIRQS
], virq_to_irq
) = {[0 ... NR_VIRQS
-1] = -1};
77 /* IRQ <-> IPI mapping */
78 static DEFINE_PER_CPU(int [XEN_NR_IPIS
], ipi_to_irq
) = {[0 ... XEN_NR_IPIS
-1] = -1};
82 static unsigned long *pirq_eoi_map
;
84 static bool (*pirq_needs_eoi
)(unsigned irq
);
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))
90 /* Xen will never allocate port zero for any purpose. */
91 #define VALID_EVTCHN(chn) ((chn) != 0)
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
);
99 static void clear_evtchn_to_irq_row(unsigned row
)
103 for (col
= 0; col
< EVTCHN_PER_ROW
; col
++)
104 evtchn_to_irq
[row
][col
] = -1;
107 static void clear_evtchn_to_irq_all(void)
111 for (row
= 0; row
< EVTCHN_ROW(xen_evtchn_max_channels()); row
++) {
112 if (evtchn_to_irq
[row
] == NULL
)
114 clear_evtchn_to_irq_row(row
);
118 static int set_evtchn_to_irq(unsigned evtchn
, unsigned irq
)
123 if (evtchn
>= xen_evtchn_max_channels())
126 row
= EVTCHN_ROW(evtchn
);
127 col
= EVTCHN_COL(evtchn
);
129 if (evtchn_to_irq
[row
] == NULL
) {
130 /* Unallocated irq entries return -1 anyway */
134 evtchn_to_irq
[row
] = (int *)get_zeroed_page(GFP_KERNEL
);
135 if (evtchn_to_irq
[row
] == NULL
)
138 clear_evtchn_to_irq_row(row
);
141 evtchn_to_irq
[row
][col
] = irq
;
145 int get_evtchn_to_irq(unsigned evtchn
)
147 if (evtchn
>= xen_evtchn_max_channels())
149 if (evtchn_to_irq
[EVTCHN_ROW(evtchn
)] == NULL
)
151 return evtchn_to_irq
[EVTCHN_ROW(evtchn
)][EVTCHN_COL(evtchn
)];
154 /* Get info for IRQ */
155 struct irq_info
*info_for_irq(unsigned irq
)
157 return irq_get_handler_data(irq
);
160 /* Constructors for packed IRQ information. */
161 static int xen_irq_info_common_setup(struct irq_info
*info
,
163 enum xen_irq_type type
,
169 BUG_ON(info
->type
!= IRQT_UNBOUND
&& info
->type
!= type
);
173 info
->evtchn
= evtchn
;
176 ret
= set_evtchn_to_irq(evtchn
, irq
);
180 irq_clear_status_flags(irq
, IRQ_NOREQUEST
|IRQ_NOAUTOEN
);
182 return xen_evtchn_port_setup(info
);
185 static int xen_irq_info_evtchn_setup(unsigned irq
,
188 struct irq_info
*info
= info_for_irq(irq
);
190 return xen_irq_info_common_setup(info
, irq
, IRQT_EVTCHN
, evtchn
, 0);
193 static int xen_irq_info_ipi_setup(unsigned cpu
,
198 struct irq_info
*info
= info_for_irq(irq
);
202 per_cpu(ipi_to_irq
, cpu
)[ipi
] = irq
;
204 return xen_irq_info_common_setup(info
, irq
, IRQT_IPI
, evtchn
, 0);
207 static int xen_irq_info_virq_setup(unsigned cpu
,
212 struct irq_info
*info
= info_for_irq(irq
);
216 per_cpu(virq_to_irq
, cpu
)[virq
] = irq
;
218 return xen_irq_info_common_setup(info
, irq
, IRQT_VIRQ
, evtchn
, 0);
221 static int xen_irq_info_pirq_setup(unsigned irq
,
228 struct irq_info
*info
= info_for_irq(irq
);
230 info
->u
.pirq
.pirq
= pirq
;
231 info
->u
.pirq
.gsi
= gsi
;
232 info
->u
.pirq
.domid
= domid
;
233 info
->u
.pirq
.flags
= flags
;
235 return xen_irq_info_common_setup(info
, irq
, IRQT_PIRQ
, evtchn
, 0);
238 static void xen_irq_info_cleanup(struct irq_info
*info
)
240 set_evtchn_to_irq(info
->evtchn
, -1);
245 * Accessors for packed IRQ information.
247 unsigned int evtchn_from_irq(unsigned irq
)
249 if (unlikely(WARN(irq
>= nr_irqs
, "Invalid irq %d!\n", irq
)))
252 return info_for_irq(irq
)->evtchn
;
255 unsigned irq_from_evtchn(unsigned int evtchn
)
257 return get_evtchn_to_irq(evtchn
);
259 EXPORT_SYMBOL_GPL(irq_from_evtchn
);
261 int irq_from_virq(unsigned int cpu
, unsigned int virq
)
263 return per_cpu(virq_to_irq
, cpu
)[virq
];
266 static enum ipi_vector
ipi_from_irq(unsigned irq
)
268 struct irq_info
*info
= info_for_irq(irq
);
270 BUG_ON(info
== NULL
);
271 BUG_ON(info
->type
!= IRQT_IPI
);
276 static unsigned virq_from_irq(unsigned irq
)
278 struct irq_info
*info
= info_for_irq(irq
);
280 BUG_ON(info
== NULL
);
281 BUG_ON(info
->type
!= IRQT_VIRQ
);
286 static unsigned pirq_from_irq(unsigned irq
)
288 struct irq_info
*info
= info_for_irq(irq
);
290 BUG_ON(info
== NULL
);
291 BUG_ON(info
->type
!= IRQT_PIRQ
);
293 return info
->u
.pirq
.pirq
;
296 static enum xen_irq_type
type_from_irq(unsigned irq
)
298 return info_for_irq(irq
)->type
;
301 unsigned cpu_from_irq(unsigned irq
)
303 return info_for_irq(irq
)->cpu
;
306 unsigned int cpu_from_evtchn(unsigned int evtchn
)
308 int irq
= get_evtchn_to_irq(evtchn
);
312 ret
= cpu_from_irq(irq
);
318 static bool pirq_check_eoi_map(unsigned irq
)
320 return test_bit(pirq_from_irq(irq
), pirq_eoi_map
);
324 static bool pirq_needs_eoi_flag(unsigned irq
)
326 struct irq_info
*info
= info_for_irq(irq
);
327 BUG_ON(info
->type
!= IRQT_PIRQ
);
329 return info
->u
.pirq
.flags
& PIRQ_NEEDS_EOI
;
332 static void bind_evtchn_to_cpu(unsigned int chn
, unsigned int cpu
)
334 int irq
= get_evtchn_to_irq(chn
);
335 struct irq_info
*info
= info_for_irq(irq
);
339 cpumask_copy(irq_get_affinity_mask(irq
), cpumask_of(cpu
));
341 xen_evtchn_port_bind_to_cpu(info
, cpu
);
347 * notify_remote_via_irq - send event to remote end of event channel via irq
348 * @irq: irq of event channel to send event to
350 * Unlike notify_remote_via_evtchn(), this is safe to use across
351 * save/restore. Notifications on a broken connection are silently
354 void notify_remote_via_irq(int irq
)
356 int evtchn
= evtchn_from_irq(irq
);
358 if (VALID_EVTCHN(evtchn
))
359 notify_remote_via_evtchn(evtchn
);
361 EXPORT_SYMBOL_GPL(notify_remote_via_irq
);
363 static void xen_irq_init(unsigned irq
)
365 struct irq_info
*info
;
367 /* By default all event channels notify CPU#0. */
368 cpumask_copy(irq_get_affinity_mask(irq
), cpumask_of(0));
371 info
= kzalloc(sizeof(*info
), GFP_KERNEL
);
373 panic("Unable to allocate metadata for IRQ%d\n", irq
);
375 info
->type
= IRQT_UNBOUND
;
378 irq_set_handler_data(irq
, info
);
380 list_add_tail(&info
->list
, &xen_irq_list_head
);
383 static int __must_check
xen_allocate_irqs_dynamic(int nvec
)
385 int i
, irq
= irq_alloc_descs(-1, 0, nvec
, -1);
388 for (i
= 0; i
< nvec
; i
++)
389 xen_irq_init(irq
+ i
);
395 static inline int __must_check
xen_allocate_irq_dynamic(void)
398 return xen_allocate_irqs_dynamic(1);
401 static int __must_check
xen_allocate_irq_gsi(unsigned gsi
)
406 * A PV guest has no concept of a GSI (since it has no ACPI
407 * nor access to/knowledge of the physical APICs). Therefore
408 * all IRQs are dynamically allocated from the entire IRQ
411 if (xen_pv_domain() && !xen_initial_domain())
412 return xen_allocate_irq_dynamic();
414 /* Legacy IRQ descriptors are already allocated by the arch. */
415 if (gsi
< nr_legacy_irqs())
418 irq
= irq_alloc_desc_at(gsi
, -1);
425 static void xen_free_irq(unsigned irq
)
427 struct irq_info
*info
= irq_get_handler_data(irq
);
432 list_del(&info
->list
);
434 irq_set_handler_data(irq
, NULL
);
436 WARN_ON(info
->refcnt
> 0);
440 /* Legacy IRQ descriptors are managed by the arch. */
441 if (irq
< nr_legacy_irqs())
447 static void xen_evtchn_close(unsigned int port
)
449 struct evtchn_close close
;
452 if (HYPERVISOR_event_channel_op(EVTCHNOP_close
, &close
) != 0)
456 static void pirq_query_unmask(int irq
)
458 struct physdev_irq_status_query irq_status
;
459 struct irq_info
*info
= info_for_irq(irq
);
461 BUG_ON(info
->type
!= IRQT_PIRQ
);
463 irq_status
.irq
= pirq_from_irq(irq
);
464 if (HYPERVISOR_physdev_op(PHYSDEVOP_irq_status_query
, &irq_status
))
465 irq_status
.flags
= 0;
467 info
->u
.pirq
.flags
&= ~PIRQ_NEEDS_EOI
;
468 if (irq_status
.flags
& XENIRQSTAT_needs_eoi
)
469 info
->u
.pirq
.flags
|= PIRQ_NEEDS_EOI
;
472 static void eoi_pirq(struct irq_data
*data
)
474 int evtchn
= evtchn_from_irq(data
->irq
);
475 struct physdev_eoi eoi
= { .irq
= pirq_from_irq(data
->irq
) };
478 if (!VALID_EVTCHN(evtchn
))
481 if (unlikely(irqd_is_setaffinity_pending(data
)) &&
482 likely(!irqd_irq_disabled(data
))) {
483 int masked
= test_and_set_mask(evtchn
);
485 clear_evtchn(evtchn
);
487 irq_move_masked_irq(data
);
490 unmask_evtchn(evtchn
);
492 clear_evtchn(evtchn
);
494 if (pirq_needs_eoi(data
->irq
)) {
495 rc
= HYPERVISOR_physdev_op(PHYSDEVOP_eoi
, &eoi
);
500 static void mask_ack_pirq(struct irq_data
*data
)
502 disable_dynirq(data
);
506 static unsigned int __startup_pirq(unsigned int irq
)
508 struct evtchn_bind_pirq bind_pirq
;
509 struct irq_info
*info
= info_for_irq(irq
);
510 int evtchn
= evtchn_from_irq(irq
);
513 BUG_ON(info
->type
!= IRQT_PIRQ
);
515 if (VALID_EVTCHN(evtchn
))
518 bind_pirq
.pirq
= pirq_from_irq(irq
);
519 /* NB. We are happy to share unless we are probing. */
520 bind_pirq
.flags
= info
->u
.pirq
.flags
& PIRQ_SHAREABLE
?
521 BIND_PIRQ__WILL_SHARE
: 0;
522 rc
= HYPERVISOR_event_channel_op(EVTCHNOP_bind_pirq
, &bind_pirq
);
524 pr_warn("Failed to obtain physical IRQ %d\n", irq
);
527 evtchn
= bind_pirq
.port
;
529 pirq_query_unmask(irq
);
531 rc
= set_evtchn_to_irq(evtchn
, irq
);
535 info
->evtchn
= evtchn
;
536 bind_evtchn_to_cpu(evtchn
, 0);
538 rc
= xen_evtchn_port_setup(info
);
543 unmask_evtchn(evtchn
);
544 eoi_pirq(irq_get_irq_data(irq
));
549 pr_err("irq%d: Failed to set port to irq mapping (%d)\n", irq
, rc
);
550 xen_evtchn_close(evtchn
);
554 static unsigned int startup_pirq(struct irq_data
*data
)
556 return __startup_pirq(data
->irq
);
559 static void shutdown_pirq(struct irq_data
*data
)
561 unsigned int irq
= data
->irq
;
562 struct irq_info
*info
= info_for_irq(irq
);
563 unsigned evtchn
= evtchn_from_irq(irq
);
565 BUG_ON(info
->type
!= IRQT_PIRQ
);
567 if (!VALID_EVTCHN(evtchn
))
571 xen_evtchn_close(evtchn
);
572 xen_irq_info_cleanup(info
);
575 static void enable_pirq(struct irq_data
*data
)
580 static void disable_pirq(struct irq_data
*data
)
582 disable_dynirq(data
);
585 int xen_irq_from_gsi(unsigned gsi
)
587 struct irq_info
*info
;
589 list_for_each_entry(info
, &xen_irq_list_head
, list
) {
590 if (info
->type
!= IRQT_PIRQ
)
593 if (info
->u
.pirq
.gsi
== gsi
)
599 EXPORT_SYMBOL_GPL(xen_irq_from_gsi
);
601 static void __unbind_from_irq(unsigned int irq
)
603 int evtchn
= evtchn_from_irq(irq
);
604 struct irq_info
*info
= irq_get_handler_data(irq
);
606 if (info
->refcnt
> 0) {
608 if (info
->refcnt
!= 0)
612 if (VALID_EVTCHN(evtchn
)) {
613 unsigned int cpu
= cpu_from_irq(irq
);
615 xen_evtchn_close(evtchn
);
617 switch (type_from_irq(irq
)) {
619 per_cpu(virq_to_irq
, cpu
)[virq_from_irq(irq
)] = -1;
622 per_cpu(ipi_to_irq
, cpu
)[ipi_from_irq(irq
)] = -1;
628 xen_irq_info_cleanup(info
);
635 * Do not make any assumptions regarding the relationship between the
636 * IRQ number returned here and the Xen pirq argument.
638 * Note: We don't assign an event channel until the irq actually started
639 * up. Return an existing irq if we've already got one for the gsi.
641 * Shareable implies level triggered, not shareable implies edge
644 int xen_bind_pirq_gsi_to_irq(unsigned gsi
,
645 unsigned pirq
, int shareable
, char *name
)
648 struct physdev_irq irq_op
;
651 mutex_lock(&irq_mapping_update_lock
);
653 irq
= xen_irq_from_gsi(gsi
);
655 pr_info("%s: returning irq %d for gsi %u\n",
660 irq
= xen_allocate_irq_gsi(gsi
);
667 /* Only the privileged domain can do this. For non-priv, the pcifront
668 * driver provides a PCI bus that does the call to do exactly
669 * this in the priv domain. */
670 if (xen_initial_domain() &&
671 HYPERVISOR_physdev_op(PHYSDEVOP_alloc_irq_vector
, &irq_op
)) {
677 ret
= xen_irq_info_pirq_setup(irq
, 0, pirq
, gsi
, DOMID_SELF
,
678 shareable
? PIRQ_SHAREABLE
: 0);
680 __unbind_from_irq(irq
);
685 pirq_query_unmask(irq
);
686 /* We try to use the handler with the appropriate semantic for the
687 * type of interrupt: if the interrupt is an edge triggered
688 * interrupt we use handle_edge_irq.
690 * On the other hand if the interrupt is level triggered we use
691 * handle_fasteoi_irq like the native code does for this kind of
694 * Depending on the Xen version, pirq_needs_eoi might return true
695 * not only for level triggered interrupts but for edge triggered
696 * interrupts too. In any case Xen always honors the eoi mechanism,
697 * not injecting any more pirqs of the same kind if the first one
698 * hasn't received an eoi yet. Therefore using the fasteoi handler
699 * is the right choice either way.
702 irq_set_chip_and_handler_name(irq
, &xen_pirq_chip
,
703 handle_fasteoi_irq
, name
);
705 irq_set_chip_and_handler_name(irq
, &xen_pirq_chip
,
706 handle_edge_irq
, name
);
709 mutex_unlock(&irq_mapping_update_lock
);
714 #ifdef CONFIG_PCI_MSI
715 int xen_allocate_pirq_msi(struct pci_dev
*dev
, struct msi_desc
*msidesc
)
718 struct physdev_get_free_pirq op_get_free_pirq
;
720 op_get_free_pirq
.type
= MAP_PIRQ_TYPE_MSI
;
721 rc
= HYPERVISOR_physdev_op(PHYSDEVOP_get_free_pirq
, &op_get_free_pirq
);
723 WARN_ONCE(rc
== -ENOSYS
,
724 "hypervisor does not support the PHYSDEVOP_get_free_pirq interface\n");
726 return rc
? -1 : op_get_free_pirq
.pirq
;
729 int xen_bind_pirq_msi_to_irq(struct pci_dev
*dev
, struct msi_desc
*msidesc
,
730 int pirq
, int nvec
, const char *name
, domid_t domid
)
734 mutex_lock(&irq_mapping_update_lock
);
736 irq
= xen_allocate_irqs_dynamic(nvec
);
740 for (i
= 0; i
< nvec
; i
++) {
741 irq_set_chip_and_handler_name(irq
+ i
, &xen_pirq_chip
, handle_edge_irq
, name
);
743 ret
= xen_irq_info_pirq_setup(irq
+ i
, 0, pirq
+ i
, 0, domid
,
744 i
== 0 ? 0 : PIRQ_MSI_GROUP
);
749 ret
= irq_set_msi_desc(irq
, msidesc
);
753 mutex_unlock(&irq_mapping_update_lock
);
757 __unbind_from_irq(irq
+ nvec
);
758 mutex_unlock(&irq_mapping_update_lock
);
763 int xen_destroy_irq(int irq
)
765 struct physdev_unmap_pirq unmap_irq
;
766 struct irq_info
*info
= info_for_irq(irq
);
769 mutex_lock(&irq_mapping_update_lock
);
772 * If trying to remove a vector in a MSI group different
773 * than the first one skip the PIRQ unmap unless this vector
774 * is the first one in the group.
776 if (xen_initial_domain() && !(info
->u
.pirq
.flags
& PIRQ_MSI_GROUP
)) {
777 unmap_irq
.pirq
= info
->u
.pirq
.pirq
;
778 unmap_irq
.domid
= info
->u
.pirq
.domid
;
779 rc
= HYPERVISOR_physdev_op(PHYSDEVOP_unmap_pirq
, &unmap_irq
);
780 /* If another domain quits without making the pci_disable_msix
781 * call, the Xen hypervisor takes care of freeing the PIRQs
782 * (free_domain_pirqs).
784 if ((rc
== -ESRCH
&& info
->u
.pirq
.domid
!= DOMID_SELF
))
785 pr_info("domain %d does not have %d anymore\n",
786 info
->u
.pirq
.domid
, info
->u
.pirq
.pirq
);
788 pr_warn("unmap irq failed %d\n", rc
);
796 mutex_unlock(&irq_mapping_update_lock
);
800 int xen_irq_from_pirq(unsigned pirq
)
804 struct irq_info
*info
;
806 mutex_lock(&irq_mapping_update_lock
);
808 list_for_each_entry(info
, &xen_irq_list_head
, list
) {
809 if (info
->type
!= IRQT_PIRQ
)
812 if (info
->u
.pirq
.pirq
== pirq
)
817 mutex_unlock(&irq_mapping_update_lock
);
823 int xen_pirq_from_irq(unsigned irq
)
825 return pirq_from_irq(irq
);
827 EXPORT_SYMBOL_GPL(xen_pirq_from_irq
);
829 int bind_evtchn_to_irq(unsigned int evtchn
)
834 if (evtchn
>= xen_evtchn_max_channels())
837 mutex_lock(&irq_mapping_update_lock
);
839 irq
= get_evtchn_to_irq(evtchn
);
842 irq
= xen_allocate_irq_dynamic();
846 irq_set_chip_and_handler_name(irq
, &xen_dynamic_chip
,
847 handle_edge_irq
, "event");
849 ret
= xen_irq_info_evtchn_setup(irq
, evtchn
);
851 __unbind_from_irq(irq
);
855 /* New interdomain events are bound to VCPU 0. */
856 bind_evtchn_to_cpu(evtchn
, 0);
858 struct irq_info
*info
= info_for_irq(irq
);
859 WARN_ON(info
== NULL
|| info
->type
!= IRQT_EVTCHN
);
863 mutex_unlock(&irq_mapping_update_lock
);
867 EXPORT_SYMBOL_GPL(bind_evtchn_to_irq
);
869 static int bind_ipi_to_irq(unsigned int ipi
, unsigned int cpu
)
871 struct evtchn_bind_ipi bind_ipi
;
875 mutex_lock(&irq_mapping_update_lock
);
877 irq
= per_cpu(ipi_to_irq
, cpu
)[ipi
];
880 irq
= xen_allocate_irq_dynamic();
884 irq_set_chip_and_handler_name(irq
, &xen_percpu_chip
,
885 handle_percpu_irq
, "ipi");
887 bind_ipi
.vcpu
= xen_vcpu_nr(cpu
);
888 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_ipi
,
891 evtchn
= bind_ipi
.port
;
893 ret
= xen_irq_info_ipi_setup(cpu
, irq
, evtchn
, ipi
);
895 __unbind_from_irq(irq
);
899 bind_evtchn_to_cpu(evtchn
, cpu
);
901 struct irq_info
*info
= info_for_irq(irq
);
902 WARN_ON(info
== NULL
|| info
->type
!= IRQT_IPI
);
906 mutex_unlock(&irq_mapping_update_lock
);
910 int bind_interdomain_evtchn_to_irq(unsigned int remote_domain
,
911 unsigned int remote_port
)
913 struct evtchn_bind_interdomain bind_interdomain
;
916 bind_interdomain
.remote_dom
= remote_domain
;
917 bind_interdomain
.remote_port
= remote_port
;
919 err
= HYPERVISOR_event_channel_op(EVTCHNOP_bind_interdomain
,
922 return err
? : bind_evtchn_to_irq(bind_interdomain
.local_port
);
924 EXPORT_SYMBOL_GPL(bind_interdomain_evtchn_to_irq
);
926 static int find_virq(unsigned int virq
, unsigned int cpu
)
928 struct evtchn_status status
;
929 int port
, rc
= -ENOENT
;
931 memset(&status
, 0, sizeof(status
));
932 for (port
= 0; port
< xen_evtchn_max_channels(); port
++) {
933 status
.dom
= DOMID_SELF
;
935 rc
= HYPERVISOR_event_channel_op(EVTCHNOP_status
, &status
);
938 if (status
.status
!= EVTCHNSTAT_virq
)
940 if (status
.u
.virq
== virq
&& status
.vcpu
== xen_vcpu_nr(cpu
)) {
949 * xen_evtchn_nr_channels - number of usable event channel ports
951 * This may be less than the maximum supported by the current
952 * hypervisor ABI. Use xen_evtchn_max_channels() for the maximum
955 unsigned xen_evtchn_nr_channels(void)
957 return evtchn_ops
->nr_channels();
959 EXPORT_SYMBOL_GPL(xen_evtchn_nr_channels
);
961 int bind_virq_to_irq(unsigned int virq
, unsigned int cpu
, bool percpu
)
963 struct evtchn_bind_virq bind_virq
;
964 int evtchn
, irq
, ret
;
966 mutex_lock(&irq_mapping_update_lock
);
968 irq
= per_cpu(virq_to_irq
, cpu
)[virq
];
971 irq
= xen_allocate_irq_dynamic();
976 irq_set_chip_and_handler_name(irq
, &xen_percpu_chip
,
977 handle_percpu_irq
, "virq");
979 irq_set_chip_and_handler_name(irq
, &xen_dynamic_chip
,
980 handle_edge_irq
, "virq");
982 bind_virq
.virq
= virq
;
983 bind_virq
.vcpu
= xen_vcpu_nr(cpu
);
984 ret
= HYPERVISOR_event_channel_op(EVTCHNOP_bind_virq
,
987 evtchn
= bind_virq
.port
;
990 ret
= find_virq(virq
, cpu
);
995 ret
= xen_irq_info_virq_setup(cpu
, irq
, evtchn
, virq
);
997 __unbind_from_irq(irq
);
1002 bind_evtchn_to_cpu(evtchn
, cpu
);
1004 struct irq_info
*info
= info_for_irq(irq
);
1005 WARN_ON(info
== NULL
|| info
->type
!= IRQT_VIRQ
);
1009 mutex_unlock(&irq_mapping_update_lock
);
1014 static void unbind_from_irq(unsigned int irq
)
1016 mutex_lock(&irq_mapping_update_lock
);
1017 __unbind_from_irq(irq
);
1018 mutex_unlock(&irq_mapping_update_lock
);
1021 int bind_evtchn_to_irqhandler(unsigned int evtchn
,
1022 irq_handler_t handler
,
1023 unsigned long irqflags
,
1024 const char *devname
, void *dev_id
)
1028 irq
= bind_evtchn_to_irq(evtchn
);
1031 retval
= request_irq(irq
, handler
, irqflags
, devname
, dev_id
);
1033 unbind_from_irq(irq
);
1039 EXPORT_SYMBOL_GPL(bind_evtchn_to_irqhandler
);
1041 int bind_interdomain_evtchn_to_irqhandler(unsigned int remote_domain
,
1042 unsigned int remote_port
,
1043 irq_handler_t handler
,
1044 unsigned long irqflags
,
1045 const char *devname
,
1050 irq
= bind_interdomain_evtchn_to_irq(remote_domain
, remote_port
);
1054 retval
= request_irq(irq
, handler
, irqflags
, devname
, dev_id
);
1056 unbind_from_irq(irq
);
1062 EXPORT_SYMBOL_GPL(bind_interdomain_evtchn_to_irqhandler
);
1064 int bind_virq_to_irqhandler(unsigned int virq
, unsigned int cpu
,
1065 irq_handler_t handler
,
1066 unsigned long irqflags
, const char *devname
, void *dev_id
)
1070 irq
= bind_virq_to_irq(virq
, cpu
, irqflags
& IRQF_PERCPU
);
1073 retval
= request_irq(irq
, handler
, irqflags
, devname
, dev_id
);
1075 unbind_from_irq(irq
);
1081 EXPORT_SYMBOL_GPL(bind_virq_to_irqhandler
);
1083 int bind_ipi_to_irqhandler(enum ipi_vector ipi
,
1085 irq_handler_t handler
,
1086 unsigned long irqflags
,
1087 const char *devname
,
1092 irq
= bind_ipi_to_irq(ipi
, cpu
);
1096 irqflags
|= IRQF_NO_SUSPEND
| IRQF_FORCE_RESUME
| IRQF_EARLY_RESUME
;
1097 retval
= request_irq(irq
, handler
, irqflags
, devname
, dev_id
);
1099 unbind_from_irq(irq
);
1106 void unbind_from_irqhandler(unsigned int irq
, void *dev_id
)
1108 struct irq_info
*info
= irq_get_handler_data(irq
);
1112 free_irq(irq
, dev_id
);
1113 unbind_from_irq(irq
);
1115 EXPORT_SYMBOL_GPL(unbind_from_irqhandler
);
1118 * xen_set_irq_priority() - set an event channel priority.
1119 * @irq:irq bound to an event channel.
1120 * @priority: priority between XEN_IRQ_PRIORITY_MAX and XEN_IRQ_PRIORITY_MIN.
1122 int xen_set_irq_priority(unsigned irq
, unsigned priority
)
1124 struct evtchn_set_priority set_priority
;
1126 set_priority
.port
= evtchn_from_irq(irq
);
1127 set_priority
.priority
= priority
;
1129 return HYPERVISOR_event_channel_op(EVTCHNOP_set_priority
,
1132 EXPORT_SYMBOL_GPL(xen_set_irq_priority
);
1134 int evtchn_make_refcounted(unsigned int evtchn
)
1136 int irq
= get_evtchn_to_irq(evtchn
);
1137 struct irq_info
*info
;
1142 info
= irq_get_handler_data(irq
);
1147 WARN_ON(info
->refcnt
!= -1);
1153 EXPORT_SYMBOL_GPL(evtchn_make_refcounted
);
1155 int evtchn_get(unsigned int evtchn
)
1158 struct irq_info
*info
;
1161 if (evtchn
>= xen_evtchn_max_channels())
1164 mutex_lock(&irq_mapping_update_lock
);
1166 irq
= get_evtchn_to_irq(evtchn
);
1170 info
= irq_get_handler_data(irq
);
1176 if (info
->refcnt
<= 0)
1182 mutex_unlock(&irq_mapping_update_lock
);
1186 EXPORT_SYMBOL_GPL(evtchn_get
);
1188 void evtchn_put(unsigned int evtchn
)
1190 int irq
= get_evtchn_to_irq(evtchn
);
1191 if (WARN_ON(irq
== -1))
1193 unbind_from_irq(irq
);
1195 EXPORT_SYMBOL_GPL(evtchn_put
);
1197 void xen_send_IPI_one(unsigned int cpu
, enum ipi_vector vector
)
1202 if (unlikely(vector
== XEN_NMI_VECTOR
)) {
1203 int rc
= HYPERVISOR_vcpu_op(VCPUOP_send_nmi
, xen_vcpu_nr(cpu
),
1206 printk(KERN_WARNING
"Sending nmi to CPU%d failed (rc:%d)\n", cpu
, rc
);
1210 irq
= per_cpu(ipi_to_irq
, cpu
)[vector
];
1212 notify_remote_via_irq(irq
);
1215 static DEFINE_PER_CPU(unsigned, xed_nesting_count
);
1217 static void __xen_evtchn_do_upcall(void)
1219 struct vcpu_info
*vcpu_info
= __this_cpu_read(xen_vcpu
);
1220 int cpu
= get_cpu();
1224 vcpu_info
->evtchn_upcall_pending
= 0;
1226 if (__this_cpu_inc_return(xed_nesting_count
) - 1)
1229 xen_evtchn_handle_events(cpu
);
1231 BUG_ON(!irqs_disabled());
1233 count
= __this_cpu_read(xed_nesting_count
);
1234 __this_cpu_write(xed_nesting_count
, 0);
1235 } while (count
!= 1 || vcpu_info
->evtchn_upcall_pending
);
1242 void xen_evtchn_do_upcall(struct pt_regs
*regs
)
1244 struct pt_regs
*old_regs
= set_irq_regs(regs
);
1248 inc_irq_stat(irq_hv_callback_count
);
1251 __xen_evtchn_do_upcall();
1254 set_irq_regs(old_regs
);
1257 void xen_hvm_evtchn_do_upcall(void)
1259 __xen_evtchn_do_upcall();
1261 EXPORT_SYMBOL_GPL(xen_hvm_evtchn_do_upcall
);
1263 /* Rebind a new event channel to an existing irq. */
1264 void rebind_evtchn_irq(int evtchn
, int irq
)
1266 struct irq_info
*info
= info_for_irq(irq
);
1271 /* Make sure the irq is masked, since the new event channel
1272 will also be masked. */
1275 mutex_lock(&irq_mapping_update_lock
);
1277 /* After resume the irq<->evtchn mappings are all cleared out */
1278 BUG_ON(get_evtchn_to_irq(evtchn
) != -1);
1279 /* Expect irq to have been bound before,
1280 so there should be a proper type */
1281 BUG_ON(info
->type
== IRQT_UNBOUND
);
1283 (void)xen_irq_info_evtchn_setup(irq
, evtchn
);
1285 mutex_unlock(&irq_mapping_update_lock
);
1287 bind_evtchn_to_cpu(evtchn
, info
->cpu
);
1288 /* This will be deferred until interrupt is processed */
1289 irq_set_affinity(irq
, cpumask_of(info
->cpu
));
1291 /* Unmask the event channel. */
1295 /* Rebind an evtchn so that it gets delivered to a specific cpu */
1296 int xen_rebind_evtchn_to_cpu(int evtchn
, unsigned tcpu
)
1298 struct evtchn_bind_vcpu bind_vcpu
;
1301 if (!VALID_EVTCHN(evtchn
))
1304 if (!xen_support_evtchn_rebind())
1307 /* Send future instances of this interrupt to other vcpu. */
1308 bind_vcpu
.port
= evtchn
;
1309 bind_vcpu
.vcpu
= xen_vcpu_nr(tcpu
);
1312 * Mask the event while changing the VCPU binding to prevent
1313 * it being delivered on an unexpected VCPU.
1315 masked
= test_and_set_mask(evtchn
);
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.
1322 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_vcpu
, &bind_vcpu
) >= 0)
1323 bind_evtchn_to_cpu(evtchn
, tcpu
);
1326 unmask_evtchn(evtchn
);
1330 EXPORT_SYMBOL_GPL(xen_rebind_evtchn_to_cpu
);
1332 static int set_affinity_irq(struct irq_data
*data
, const struct cpumask
*dest
,
1335 unsigned tcpu
= cpumask_first_and(dest
, cpu_online_mask
);
1336 int ret
= xen_rebind_evtchn_to_cpu(evtchn_from_irq(data
->irq
), tcpu
);
1339 irq_data_update_effective_affinity(data
, cpumask_of(tcpu
));
1344 static void enable_dynirq(struct irq_data
*data
)
1346 int evtchn
= evtchn_from_irq(data
->irq
);
1348 if (VALID_EVTCHN(evtchn
))
1349 unmask_evtchn(evtchn
);
1352 static void disable_dynirq(struct irq_data
*data
)
1354 int evtchn
= evtchn_from_irq(data
->irq
);
1356 if (VALID_EVTCHN(evtchn
))
1357 mask_evtchn(evtchn
);
1360 static void ack_dynirq(struct irq_data
*data
)
1362 int evtchn
= evtchn_from_irq(data
->irq
);
1364 if (!VALID_EVTCHN(evtchn
))
1367 if (unlikely(irqd_is_setaffinity_pending(data
)) &&
1368 likely(!irqd_irq_disabled(data
))) {
1369 int masked
= test_and_set_mask(evtchn
);
1371 clear_evtchn(evtchn
);
1373 irq_move_masked_irq(data
);
1376 unmask_evtchn(evtchn
);
1378 clear_evtchn(evtchn
);
1381 static void mask_ack_dynirq(struct irq_data
*data
)
1383 disable_dynirq(data
);
1387 static int retrigger_dynirq(struct irq_data
*data
)
1389 unsigned int evtchn
= evtchn_from_irq(data
->irq
);
1392 if (!VALID_EVTCHN(evtchn
))
1395 masked
= test_and_set_mask(evtchn
);
1398 unmask_evtchn(evtchn
);
1403 static void restore_pirqs(void)
1405 int pirq
, rc
, irq
, gsi
;
1406 struct physdev_map_pirq map_irq
;
1407 struct irq_info
*info
;
1409 list_for_each_entry(info
, &xen_irq_list_head
, list
) {
1410 if (info
->type
!= IRQT_PIRQ
)
1413 pirq
= info
->u
.pirq
.pirq
;
1414 gsi
= info
->u
.pirq
.gsi
;
1417 /* save/restore of PT devices doesn't work, so at this point the
1418 * only devices present are GSI based emulated devices */
1422 map_irq
.domid
= DOMID_SELF
;
1423 map_irq
.type
= MAP_PIRQ_TYPE_GSI
;
1424 map_irq
.index
= gsi
;
1425 map_irq
.pirq
= pirq
;
1427 rc
= HYPERVISOR_physdev_op(PHYSDEVOP_map_pirq
, &map_irq
);
1429 pr_warn("xen map irq failed gsi=%d irq=%d pirq=%d rc=%d\n",
1430 gsi
, irq
, pirq
, rc
);
1435 printk(KERN_DEBUG
"xen: --> irq=%d, pirq=%d\n", irq
, map_irq
.pirq
);
1437 __startup_pirq(irq
);
1441 static void restore_cpu_virqs(unsigned int cpu
)
1443 struct evtchn_bind_virq bind_virq
;
1444 int virq
, irq
, evtchn
;
1446 for (virq
= 0; virq
< NR_VIRQS
; virq
++) {
1447 if ((irq
= per_cpu(virq_to_irq
, cpu
)[virq
]) == -1)
1450 BUG_ON(virq_from_irq(irq
) != virq
);
1452 /* Get a new binding from Xen. */
1453 bind_virq
.virq
= virq
;
1454 bind_virq
.vcpu
= xen_vcpu_nr(cpu
);
1455 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_virq
,
1458 evtchn
= bind_virq
.port
;
1460 /* Record the new mapping. */
1461 (void)xen_irq_info_virq_setup(cpu
, irq
, evtchn
, virq
);
1462 bind_evtchn_to_cpu(evtchn
, cpu
);
1466 static void restore_cpu_ipis(unsigned int cpu
)
1468 struct evtchn_bind_ipi bind_ipi
;
1469 int ipi
, irq
, evtchn
;
1471 for (ipi
= 0; ipi
< XEN_NR_IPIS
; ipi
++) {
1472 if ((irq
= per_cpu(ipi_to_irq
, cpu
)[ipi
]) == -1)
1475 BUG_ON(ipi_from_irq(irq
) != ipi
);
1477 /* Get a new binding from Xen. */
1478 bind_ipi
.vcpu
= xen_vcpu_nr(cpu
);
1479 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_ipi
,
1482 evtchn
= bind_ipi
.port
;
1484 /* Record the new mapping. */
1485 (void)xen_irq_info_ipi_setup(cpu
, irq
, evtchn
, ipi
);
1486 bind_evtchn_to_cpu(evtchn
, cpu
);
1490 /* Clear an irq's pending state, in preparation for polling on it */
1491 void xen_clear_irq_pending(int irq
)
1493 int evtchn
= evtchn_from_irq(irq
);
1495 if (VALID_EVTCHN(evtchn
))
1496 clear_evtchn(evtchn
);
1498 EXPORT_SYMBOL(xen_clear_irq_pending
);
1499 void xen_set_irq_pending(int irq
)
1501 int evtchn
= evtchn_from_irq(irq
);
1503 if (VALID_EVTCHN(evtchn
))
1507 bool xen_test_irq_pending(int irq
)
1509 int evtchn
= evtchn_from_irq(irq
);
1512 if (VALID_EVTCHN(evtchn
))
1513 ret
= test_evtchn(evtchn
);
1518 /* Poll waiting for an irq to become pending with timeout. In the usual case,
1519 * the irq will be disabled so it won't deliver an interrupt. */
1520 void xen_poll_irq_timeout(int irq
, u64 timeout
)
1522 evtchn_port_t evtchn
= evtchn_from_irq(irq
);
1524 if (VALID_EVTCHN(evtchn
)) {
1525 struct sched_poll poll
;
1528 poll
.timeout
= timeout
;
1529 set_xen_guest_handle(poll
.ports
, &evtchn
);
1531 if (HYPERVISOR_sched_op(SCHEDOP_poll
, &poll
) != 0)
1535 EXPORT_SYMBOL(xen_poll_irq_timeout
);
1536 /* Poll waiting for an irq to become pending. In the usual case, the
1537 * irq will be disabled so it won't deliver an interrupt. */
1538 void xen_poll_irq(int irq
)
1540 xen_poll_irq_timeout(irq
, 0 /* no timeout */);
1543 /* Check whether the IRQ line is shared with other guests. */
1544 int xen_test_irq_shared(int irq
)
1546 struct irq_info
*info
= info_for_irq(irq
);
1547 struct physdev_irq_status_query irq_status
;
1552 irq_status
.irq
= info
->u
.pirq
.pirq
;
1554 if (HYPERVISOR_physdev_op(PHYSDEVOP_irq_status_query
, &irq_status
))
1556 return !(irq_status
.flags
& XENIRQSTAT_shared
);
1558 EXPORT_SYMBOL_GPL(xen_test_irq_shared
);
1560 void xen_irq_resume(void)
1563 struct irq_info
*info
;
1565 /* New event-channel space is not 'live' yet. */
1566 xen_evtchn_resume();
1568 /* No IRQ <-> event-channel mappings. */
1569 list_for_each_entry(info
, &xen_irq_list_head
, list
)
1570 info
->evtchn
= 0; /* zap event-channel binding */
1572 clear_evtchn_to_irq_all();
1574 for_each_possible_cpu(cpu
) {
1575 restore_cpu_virqs(cpu
);
1576 restore_cpu_ipis(cpu
);
1582 static struct irq_chip xen_dynamic_chip __read_mostly
= {
1585 .irq_disable
= disable_dynirq
,
1586 .irq_mask
= disable_dynirq
,
1587 .irq_unmask
= enable_dynirq
,
1589 .irq_ack
= ack_dynirq
,
1590 .irq_mask_ack
= mask_ack_dynirq
,
1592 .irq_set_affinity
= set_affinity_irq
,
1593 .irq_retrigger
= retrigger_dynirq
,
1596 static struct irq_chip xen_pirq_chip __read_mostly
= {
1599 .irq_startup
= startup_pirq
,
1600 .irq_shutdown
= shutdown_pirq
,
1601 .irq_enable
= enable_pirq
,
1602 .irq_disable
= disable_pirq
,
1604 .irq_mask
= disable_dynirq
,
1605 .irq_unmask
= enable_dynirq
,
1607 .irq_ack
= eoi_pirq
,
1608 .irq_eoi
= eoi_pirq
,
1609 .irq_mask_ack
= mask_ack_pirq
,
1611 .irq_set_affinity
= set_affinity_irq
,
1613 .irq_retrigger
= retrigger_dynirq
,
1616 static struct irq_chip xen_percpu_chip __read_mostly
= {
1617 .name
= "xen-percpu",
1619 .irq_disable
= disable_dynirq
,
1620 .irq_mask
= disable_dynirq
,
1621 .irq_unmask
= enable_dynirq
,
1623 .irq_ack
= ack_dynirq
,
1626 int xen_set_callback_via(uint64_t via
)
1628 struct xen_hvm_param a
;
1629 a
.domid
= DOMID_SELF
;
1630 a
.index
= HVM_PARAM_CALLBACK_IRQ
;
1632 return HYPERVISOR_hvm_op(HVMOP_set_param
, &a
);
1634 EXPORT_SYMBOL_GPL(xen_set_callback_via
);
1636 #ifdef CONFIG_XEN_PVHVM
1637 /* Vector callbacks are better than PCI interrupts to receive event
1638 * channel notifications because we can receive vector callbacks on any
1639 * vcpu and we don't need PCI support or APIC interactions. */
1640 void xen_callback_vector(void)
1643 uint64_t callback_via
;
1645 if (xen_have_vector_callback
) {
1646 callback_via
= HVM_CALLBACK_VECTOR(HYPERVISOR_CALLBACK_VECTOR
);
1647 rc
= xen_set_callback_via(callback_via
);
1649 pr_err("Request for Xen HVM callback vector failed\n");
1650 xen_have_vector_callback
= 0;
1653 pr_info_once("Xen HVM callback vector for event delivery is enabled\n");
1654 alloc_intr_gate(HYPERVISOR_CALLBACK_VECTOR
,
1655 xen_hvm_callback_vector
);
1659 void xen_callback_vector(void) {}
1662 #undef MODULE_PARAM_PREFIX
1663 #define MODULE_PARAM_PREFIX "xen."
1665 static bool fifo_events
= true;
1666 module_param(fifo_events
, bool, 0);
1668 void __init
xen_init_IRQ(void)
1671 unsigned int evtchn
;
1674 ret
= xen_evtchn_fifo_init();
1676 xen_evtchn_2l_init();
1678 evtchn_to_irq
= kcalloc(EVTCHN_ROW(xen_evtchn_max_channels()),
1679 sizeof(*evtchn_to_irq
), GFP_KERNEL
);
1680 BUG_ON(!evtchn_to_irq
);
1682 /* No event channels are 'live' right now. */
1683 for (evtchn
= 0; evtchn
< xen_evtchn_nr_channels(); evtchn
++)
1684 mask_evtchn(evtchn
);
1686 pirq_needs_eoi
= pirq_needs_eoi_flag
;
1689 if (xen_pv_domain()) {
1690 irq_ctx_init(smp_processor_id());
1691 if (xen_initial_domain())
1692 pci_xen_initial_domain();
1694 if (xen_feature(XENFEAT_hvm_callback_vector
))
1695 xen_callback_vector();
1697 if (xen_hvm_domain()) {
1699 /* pci_xen_hvm_init must be called after native_init_IRQ so that
1700 * __acpi_register_gsi can point at the right function */
1704 struct physdev_pirq_eoi_gmfn eoi_gmfn
;
1706 pirq_eoi_map
= (void *)__get_free_page(GFP_KERNEL
|__GFP_ZERO
);
1707 eoi_gmfn
.gmfn
= virt_to_gfn(pirq_eoi_map
);
1708 rc
= HYPERVISOR_physdev_op(PHYSDEVOP_pirq_eoi_gmfn_v2
, &eoi_gmfn
);
1710 free_page((unsigned long) pirq_eoi_map
);
1711 pirq_eoi_map
= NULL
;
1713 pirq_needs_eoi
= pirq_check_eoi_map
;