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/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>
38 #include <asm/ptrace.h>
41 #include <asm/io_apic.h>
42 #include <asm/xen/pci.h>
45 #include <asm/sync_bitops.h>
46 #include <asm/xen/hypercall.h>
47 #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
[EVTCHN_ROW(evtchn
)][EVTCHN_COL(evtchn
)] = 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_irq_data(irq
)->affinity
, cpumask_of(cpu
));
341 xen_evtchn_port_bind_to_cpu(info
, cpu
);
346 static void xen_evtchn_mask_all(void)
350 for (evtchn
= 0; evtchn
< xen_evtchn_nr_channels(); evtchn
++)
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
358 * Unlike notify_remote_via_evtchn(), this is safe to use across
359 * save/restore. Notifications on a broken connection are silently
362 void notify_remote_via_irq(int irq
)
364 int evtchn
= evtchn_from_irq(irq
);
366 if (VALID_EVTCHN(evtchn
))
367 notify_remote_via_evtchn(evtchn
);
369 EXPORT_SYMBOL_GPL(notify_remote_via_irq
);
371 static void xen_irq_init(unsigned irq
)
373 struct irq_info
*info
;
375 /* By default all event channels notify CPU#0. */
376 cpumask_copy(irq_get_irq_data(irq
)->affinity
, cpumask_of(0));
379 info
= kzalloc(sizeof(*info
), GFP_KERNEL
);
381 panic("Unable to allocate metadata for IRQ%d\n", irq
);
383 info
->type
= IRQT_UNBOUND
;
386 irq_set_handler_data(irq
, info
);
388 list_add_tail(&info
->list
, &xen_irq_list_head
);
391 static int __must_check
xen_allocate_irqs_dynamic(int nvec
)
393 int i
, irq
= irq_alloc_descs(-1, 0, nvec
, -1);
396 for (i
= 0; i
< nvec
; i
++)
397 xen_irq_init(irq
+ i
);
403 static inline int __must_check
xen_allocate_irq_dynamic(void)
406 return xen_allocate_irqs_dynamic(1);
409 static int __must_check
xen_allocate_irq_gsi(unsigned gsi
)
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
419 if (xen_pv_domain() && !xen_initial_domain())
420 return xen_allocate_irq_dynamic();
422 /* Legacy IRQ descriptors are already allocated by the arch. */
423 if (gsi
< NR_IRQS_LEGACY
)
426 irq
= irq_alloc_desc_at(gsi
, -1);
433 static void xen_free_irq(unsigned irq
)
435 struct irq_info
*info
= irq_get_handler_data(irq
);
440 list_del(&info
->list
);
442 irq_set_handler_data(irq
, NULL
);
444 WARN_ON(info
->refcnt
> 0);
448 /* Legacy IRQ descriptors are managed by the arch. */
449 if (irq
< NR_IRQS_LEGACY
)
455 static void xen_evtchn_close(unsigned int port
, unsigned int cpu
)
457 struct evtchn_close close
;
459 xen_evtchn_op_close(port
, cpu
);
462 if (HYPERVISOR_event_channel_op(EVTCHNOP_close
, &close
) != 0)
466 static void pirq_query_unmask(int irq
)
468 struct physdev_irq_status_query irq_status
;
469 struct irq_info
*info
= info_for_irq(irq
);
471 BUG_ON(info
->type
!= IRQT_PIRQ
);
473 irq_status
.irq
= pirq_from_irq(irq
);
474 if (HYPERVISOR_physdev_op(PHYSDEVOP_irq_status_query
, &irq_status
))
475 irq_status
.flags
= 0;
477 info
->u
.pirq
.flags
&= ~PIRQ_NEEDS_EOI
;
478 if (irq_status
.flags
& XENIRQSTAT_needs_eoi
)
479 info
->u
.pirq
.flags
|= PIRQ_NEEDS_EOI
;
482 static void eoi_pirq(struct irq_data
*data
)
484 int evtchn
= evtchn_from_irq(data
->irq
);
485 struct physdev_eoi eoi
= { .irq
= pirq_from_irq(data
->irq
) };
490 if (VALID_EVTCHN(evtchn
))
491 clear_evtchn(evtchn
);
493 if (pirq_needs_eoi(data
->irq
)) {
494 rc
= HYPERVISOR_physdev_op(PHYSDEVOP_eoi
, &eoi
);
499 static void mask_ack_pirq(struct irq_data
*data
)
501 disable_dynirq(data
);
505 static unsigned int __startup_pirq(unsigned int irq
)
507 struct evtchn_bind_pirq bind_pirq
;
508 struct irq_info
*info
= info_for_irq(irq
);
509 int evtchn
= evtchn_from_irq(irq
);
512 BUG_ON(info
->type
!= IRQT_PIRQ
);
514 if (VALID_EVTCHN(evtchn
))
517 bind_pirq
.pirq
= pirq_from_irq(irq
);
518 /* NB. We are happy to share unless we are probing. */
519 bind_pirq
.flags
= info
->u
.pirq
.flags
& PIRQ_SHAREABLE
?
520 BIND_PIRQ__WILL_SHARE
: 0;
521 rc
= HYPERVISOR_event_channel_op(EVTCHNOP_bind_pirq
, &bind_pirq
);
523 pr_warn("Failed to obtain physical IRQ %d\n", irq
);
526 evtchn
= bind_pirq
.port
;
528 pirq_query_unmask(irq
);
530 rc
= set_evtchn_to_irq(evtchn
, irq
);
534 info
->evtchn
= evtchn
;
535 bind_evtchn_to_cpu(evtchn
, 0);
537 rc
= xen_evtchn_port_setup(info
);
542 unmask_evtchn(evtchn
);
543 eoi_pirq(irq_get_irq_data(irq
));
548 pr_err("irq%d: Failed to set port to irq mapping (%d)\n", irq
, rc
);
549 xen_evtchn_close(evtchn
, NR_CPUS
);
553 static unsigned int startup_pirq(struct irq_data
*data
)
555 return __startup_pirq(data
->irq
);
558 static void shutdown_pirq(struct irq_data
*data
)
560 unsigned int irq
= data
->irq
;
561 struct irq_info
*info
= info_for_irq(irq
);
562 unsigned evtchn
= evtchn_from_irq(irq
);
564 BUG_ON(info
->type
!= IRQT_PIRQ
);
566 if (!VALID_EVTCHN(evtchn
))
570 xen_evtchn_close(evtchn
, cpu_from_evtchn(evtchn
));
571 xen_irq_info_cleanup(info
);
574 static void enable_pirq(struct irq_data
*data
)
579 static void disable_pirq(struct irq_data
*data
)
581 disable_dynirq(data
);
584 int xen_irq_from_gsi(unsigned gsi
)
586 struct irq_info
*info
;
588 list_for_each_entry(info
, &xen_irq_list_head
, list
) {
589 if (info
->type
!= IRQT_PIRQ
)
592 if (info
->u
.pirq
.gsi
== gsi
)
598 EXPORT_SYMBOL_GPL(xen_irq_from_gsi
);
600 static void __unbind_from_irq(unsigned int irq
)
602 int evtchn
= evtchn_from_irq(irq
);
603 struct irq_info
*info
= irq_get_handler_data(irq
);
605 if (info
->refcnt
> 0) {
607 if (info
->refcnt
!= 0)
611 if (VALID_EVTCHN(evtchn
)) {
612 unsigned int cpu
= cpu_from_irq(irq
);
614 xen_evtchn_close(evtchn
, cpu
);
616 switch (type_from_irq(irq
)) {
618 per_cpu(virq_to_irq
, cpu
)[virq_from_irq(irq
)] = -1;
621 per_cpu(ipi_to_irq
, cpu
)[ipi_from_irq(irq
)] = -1;
627 xen_irq_info_cleanup(info
);
630 BUG_ON(info_for_irq(irq
)->type
== IRQT_UNBOUND
);
636 * Do not make any assumptions regarding the relationship between the
637 * IRQ number returned here and the Xen pirq argument.
639 * Note: We don't assign an event channel until the irq actually started
640 * up. Return an existing irq if we've already got one for the gsi.
642 * Shareable implies level triggered, not shareable implies edge
645 int xen_bind_pirq_gsi_to_irq(unsigned gsi
,
646 unsigned pirq
, int shareable
, char *name
)
649 struct physdev_irq irq_op
;
652 mutex_lock(&irq_mapping_update_lock
);
654 irq
= xen_irq_from_gsi(gsi
);
656 pr_info("%s: returning irq %d for gsi %u\n",
661 irq
= xen_allocate_irq_gsi(gsi
);
668 /* Only the privileged domain can do this. For non-priv, the pcifront
669 * driver provides a PCI bus that does the call to do exactly
670 * this in the priv domain. */
671 if (xen_initial_domain() &&
672 HYPERVISOR_physdev_op(PHYSDEVOP_alloc_irq_vector
, &irq_op
)) {
678 ret
= xen_irq_info_pirq_setup(irq
, 0, pirq
, gsi
, DOMID_SELF
,
679 shareable
? PIRQ_SHAREABLE
: 0);
681 __unbind_from_irq(irq
);
686 pirq_query_unmask(irq
);
687 /* We try to use the handler with the appropriate semantic for the
688 * type of interrupt: if the interrupt is an edge triggered
689 * interrupt we use handle_edge_irq.
691 * On the other hand if the interrupt is level triggered we use
692 * handle_fasteoi_irq like the native code does for this kind of
695 * Depending on the Xen version, pirq_needs_eoi might return true
696 * not only for level triggered interrupts but for edge triggered
697 * interrupts too. In any case Xen always honors the eoi mechanism,
698 * not injecting any more pirqs of the same kind if the first one
699 * hasn't received an eoi yet. Therefore using the fasteoi handler
700 * is the right choice either way.
703 irq_set_chip_and_handler_name(irq
, &xen_pirq_chip
,
704 handle_fasteoi_irq
, name
);
706 irq_set_chip_and_handler_name(irq
, &xen_pirq_chip
,
707 handle_edge_irq
, name
);
710 mutex_unlock(&irq_mapping_update_lock
);
715 #ifdef CONFIG_PCI_MSI
716 int xen_allocate_pirq_msi(struct pci_dev
*dev
, struct msi_desc
*msidesc
)
719 struct physdev_get_free_pirq op_get_free_pirq
;
721 op_get_free_pirq
.type
= MAP_PIRQ_TYPE_MSI
;
722 rc
= HYPERVISOR_physdev_op(PHYSDEVOP_get_free_pirq
, &op_get_free_pirq
);
724 WARN_ONCE(rc
== -ENOSYS
,
725 "hypervisor does not support the PHYSDEVOP_get_free_pirq interface\n");
727 return rc
? -1 : op_get_free_pirq
.pirq
;
730 int xen_bind_pirq_msi_to_irq(struct pci_dev
*dev
, struct msi_desc
*msidesc
,
731 int pirq
, int nvec
, const char *name
, domid_t domid
)
735 mutex_lock(&irq_mapping_update_lock
);
737 irq
= xen_allocate_irqs_dynamic(nvec
);
741 for (i
= 0; i
< nvec
; i
++) {
742 irq_set_chip_and_handler_name(irq
+ i
, &xen_pirq_chip
, handle_edge_irq
, name
);
744 ret
= xen_irq_info_pirq_setup(irq
+ i
, 0, pirq
+ i
, 0, domid
,
745 i
== 0 ? 0 : PIRQ_MSI_GROUP
);
750 ret
= irq_set_msi_desc(irq
, msidesc
);
754 mutex_unlock(&irq_mapping_update_lock
);
758 __unbind_from_irq(irq
+ i
);
759 mutex_unlock(&irq_mapping_update_lock
);
764 int xen_destroy_irq(int irq
)
766 struct physdev_unmap_pirq unmap_irq
;
767 struct irq_info
*info
= info_for_irq(irq
);
770 mutex_lock(&irq_mapping_update_lock
);
773 * If trying to remove a vector in a MSI group different
774 * than the first one skip the PIRQ unmap unless this vector
775 * is the first one in the group.
777 if (xen_initial_domain() && !(info
->u
.pirq
.flags
& PIRQ_MSI_GROUP
)) {
778 unmap_irq
.pirq
= info
->u
.pirq
.pirq
;
779 unmap_irq
.domid
= info
->u
.pirq
.domid
;
780 rc
= HYPERVISOR_physdev_op(PHYSDEVOP_unmap_pirq
, &unmap_irq
);
781 /* If another domain quits without making the pci_disable_msix
782 * call, the Xen hypervisor takes care of freeing the PIRQs
783 * (free_domain_pirqs).
785 if ((rc
== -ESRCH
&& info
->u
.pirq
.domid
!= DOMID_SELF
))
786 pr_info("domain %d does not have %d anymore\n",
787 info
->u
.pirq
.domid
, info
->u
.pirq
.pirq
);
789 pr_warn("unmap irq failed %d\n", rc
);
797 mutex_unlock(&irq_mapping_update_lock
);
801 int xen_irq_from_pirq(unsigned pirq
)
805 struct irq_info
*info
;
807 mutex_lock(&irq_mapping_update_lock
);
809 list_for_each_entry(info
, &xen_irq_list_head
, list
) {
810 if (info
->type
!= IRQT_PIRQ
)
813 if (info
->u
.pirq
.pirq
== pirq
)
818 mutex_unlock(&irq_mapping_update_lock
);
824 int xen_pirq_from_irq(unsigned irq
)
826 return pirq_from_irq(irq
);
828 EXPORT_SYMBOL_GPL(xen_pirq_from_irq
);
830 int bind_evtchn_to_irq(unsigned int evtchn
)
835 if (evtchn
>= xen_evtchn_max_channels())
838 mutex_lock(&irq_mapping_update_lock
);
840 irq
= get_evtchn_to_irq(evtchn
);
843 irq
= xen_allocate_irq_dynamic();
847 irq_set_chip_and_handler_name(irq
, &xen_dynamic_chip
,
848 handle_edge_irq
, "event");
850 ret
= xen_irq_info_evtchn_setup(irq
, evtchn
);
852 __unbind_from_irq(irq
);
856 /* New interdomain events are bound to VCPU 0. */
857 bind_evtchn_to_cpu(evtchn
, 0);
859 struct irq_info
*info
= info_for_irq(irq
);
860 WARN_ON(info
== NULL
|| info
->type
!= IRQT_EVTCHN
);
864 mutex_unlock(&irq_mapping_update_lock
);
868 EXPORT_SYMBOL_GPL(bind_evtchn_to_irq
);
870 static int bind_ipi_to_irq(unsigned int ipi
, unsigned int cpu
)
872 struct evtchn_bind_ipi bind_ipi
;
876 mutex_lock(&irq_mapping_update_lock
);
878 irq
= per_cpu(ipi_to_irq
, cpu
)[ipi
];
881 irq
= xen_allocate_irq_dynamic();
885 irq_set_chip_and_handler_name(irq
, &xen_percpu_chip
,
886 handle_percpu_irq
, "ipi");
889 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_ipi
,
892 evtchn
= bind_ipi
.port
;
894 ret
= xen_irq_info_ipi_setup(cpu
, irq
, evtchn
, ipi
);
896 __unbind_from_irq(irq
);
900 bind_evtchn_to_cpu(evtchn
, cpu
);
902 struct irq_info
*info
= info_for_irq(irq
);
903 WARN_ON(info
== NULL
|| info
->type
!= IRQT_IPI
);
907 mutex_unlock(&irq_mapping_update_lock
);
911 int bind_interdomain_evtchn_to_irq(unsigned int remote_domain
,
912 unsigned int remote_port
)
914 struct evtchn_bind_interdomain bind_interdomain
;
917 bind_interdomain
.remote_dom
= remote_domain
;
918 bind_interdomain
.remote_port
= remote_port
;
920 err
= HYPERVISOR_event_channel_op(EVTCHNOP_bind_interdomain
,
923 return err
? : bind_evtchn_to_irq(bind_interdomain
.local_port
);
925 EXPORT_SYMBOL_GPL(bind_interdomain_evtchn_to_irq
);
927 static int find_virq(unsigned int virq
, unsigned int cpu
)
929 struct evtchn_status status
;
930 int port
, rc
= -ENOENT
;
932 memset(&status
, 0, sizeof(status
));
933 for (port
= 0; port
< xen_evtchn_max_channels(); port
++) {
934 status
.dom
= DOMID_SELF
;
936 rc
= HYPERVISOR_event_channel_op(EVTCHNOP_status
, &status
);
939 if (status
.status
!= EVTCHNSTAT_virq
)
941 if (status
.u
.virq
== virq
&& status
.vcpu
== cpu
) {
950 * xen_evtchn_nr_channels - number of usable event channel ports
952 * This may be less than the maximum supported by the current
953 * hypervisor ABI. Use xen_evtchn_max_channels() for the maximum
956 unsigned xen_evtchn_nr_channels(void)
958 return evtchn_ops
->nr_channels();
960 EXPORT_SYMBOL_GPL(xen_evtchn_nr_channels
);
962 int bind_virq_to_irq(unsigned int virq
, unsigned int cpu
, bool percpu
)
964 struct evtchn_bind_virq bind_virq
;
965 int evtchn
, irq
, ret
;
967 mutex_lock(&irq_mapping_update_lock
);
969 irq
= per_cpu(virq_to_irq
, cpu
)[virq
];
972 irq
= xen_allocate_irq_dynamic();
977 irq_set_chip_and_handler_name(irq
, &xen_percpu_chip
,
978 handle_percpu_irq
, "virq");
980 irq_set_chip_and_handler_name(irq
, &xen_dynamic_chip
,
981 handle_edge_irq
, "virq");
983 bind_virq
.virq
= virq
;
984 bind_virq
.vcpu
= cpu
;
985 ret
= HYPERVISOR_event_channel_op(EVTCHNOP_bind_virq
,
988 evtchn
= bind_virq
.port
;
991 ret
= find_virq(virq
, cpu
);
996 ret
= xen_irq_info_virq_setup(cpu
, irq
, evtchn
, virq
);
998 __unbind_from_irq(irq
);
1003 bind_evtchn_to_cpu(evtchn
, cpu
);
1005 struct irq_info
*info
= info_for_irq(irq
);
1006 WARN_ON(info
== NULL
|| info
->type
!= IRQT_VIRQ
);
1010 mutex_unlock(&irq_mapping_update_lock
);
1015 static void unbind_from_irq(unsigned int irq
)
1017 mutex_lock(&irq_mapping_update_lock
);
1018 __unbind_from_irq(irq
);
1019 mutex_unlock(&irq_mapping_update_lock
);
1022 int bind_evtchn_to_irqhandler(unsigned int evtchn
,
1023 irq_handler_t handler
,
1024 unsigned long irqflags
,
1025 const char *devname
, void *dev_id
)
1029 irq
= bind_evtchn_to_irq(evtchn
);
1032 retval
= request_irq(irq
, handler
, irqflags
, devname
, dev_id
);
1034 unbind_from_irq(irq
);
1040 EXPORT_SYMBOL_GPL(bind_evtchn_to_irqhandler
);
1042 int bind_interdomain_evtchn_to_irqhandler(unsigned int remote_domain
,
1043 unsigned int remote_port
,
1044 irq_handler_t handler
,
1045 unsigned long irqflags
,
1046 const char *devname
,
1051 irq
= bind_interdomain_evtchn_to_irq(remote_domain
, remote_port
);
1055 retval
= request_irq(irq
, handler
, irqflags
, devname
, dev_id
);
1057 unbind_from_irq(irq
);
1063 EXPORT_SYMBOL_GPL(bind_interdomain_evtchn_to_irqhandler
);
1065 int bind_virq_to_irqhandler(unsigned int virq
, unsigned int cpu
,
1066 irq_handler_t handler
,
1067 unsigned long irqflags
, const char *devname
, void *dev_id
)
1071 irq
= bind_virq_to_irq(virq
, cpu
, irqflags
& IRQF_PERCPU
);
1074 retval
= request_irq(irq
, handler
, irqflags
, devname
, dev_id
);
1076 unbind_from_irq(irq
);
1082 EXPORT_SYMBOL_GPL(bind_virq_to_irqhandler
);
1084 int bind_ipi_to_irqhandler(enum ipi_vector ipi
,
1086 irq_handler_t handler
,
1087 unsigned long irqflags
,
1088 const char *devname
,
1093 irq
= bind_ipi_to_irq(ipi
, cpu
);
1097 irqflags
|= IRQF_NO_SUSPEND
| IRQF_FORCE_RESUME
| IRQF_EARLY_RESUME
;
1098 retval
= request_irq(irq
, handler
, irqflags
, devname
, dev_id
);
1100 unbind_from_irq(irq
);
1107 void unbind_from_irqhandler(unsigned int irq
, void *dev_id
)
1109 struct irq_info
*info
= irq_get_handler_data(irq
);
1113 free_irq(irq
, dev_id
);
1114 unbind_from_irq(irq
);
1116 EXPORT_SYMBOL_GPL(unbind_from_irqhandler
);
1119 * xen_set_irq_priority() - set an event channel priority.
1120 * @irq:irq bound to an event channel.
1121 * @priority: priority between XEN_IRQ_PRIORITY_MAX and XEN_IRQ_PRIORITY_MIN.
1123 int xen_set_irq_priority(unsigned irq
, unsigned priority
)
1125 struct evtchn_set_priority set_priority
;
1127 set_priority
.port
= evtchn_from_irq(irq
);
1128 set_priority
.priority
= priority
;
1130 return HYPERVISOR_event_channel_op(EVTCHNOP_set_priority
,
1133 EXPORT_SYMBOL_GPL(xen_set_irq_priority
);
1135 int evtchn_make_refcounted(unsigned int evtchn
)
1137 int irq
= get_evtchn_to_irq(evtchn
);
1138 struct irq_info
*info
;
1143 info
= irq_get_handler_data(irq
);
1148 WARN_ON(info
->refcnt
!= -1);
1154 EXPORT_SYMBOL_GPL(evtchn_make_refcounted
);
1156 int evtchn_get(unsigned int evtchn
)
1159 struct irq_info
*info
;
1162 if (evtchn
>= xen_evtchn_max_channels())
1165 mutex_lock(&irq_mapping_update_lock
);
1167 irq
= get_evtchn_to_irq(evtchn
);
1171 info
= irq_get_handler_data(irq
);
1177 if (info
->refcnt
<= 0)
1183 mutex_unlock(&irq_mapping_update_lock
);
1187 EXPORT_SYMBOL_GPL(evtchn_get
);
1189 void evtchn_put(unsigned int evtchn
)
1191 int irq
= get_evtchn_to_irq(evtchn
);
1192 if (WARN_ON(irq
== -1))
1194 unbind_from_irq(irq
);
1196 EXPORT_SYMBOL_GPL(evtchn_put
);
1198 void xen_send_IPI_one(unsigned int cpu
, enum ipi_vector vector
)
1203 if (unlikely(vector
== XEN_NMI_VECTOR
)) {
1204 int rc
= HYPERVISOR_vcpu_op(VCPUOP_send_nmi
, cpu
, NULL
);
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
);
1249 inc_irq_stat(irq_hv_callback_count
);
1252 __xen_evtchn_do_upcall();
1255 set_irq_regs(old_regs
);
1258 void xen_hvm_evtchn_do_upcall(void)
1260 __xen_evtchn_do_upcall();
1262 EXPORT_SYMBOL_GPL(xen_hvm_evtchn_do_upcall
);
1264 /* Rebind a new event channel to an existing irq. */
1265 void rebind_evtchn_irq(int evtchn
, int irq
)
1267 struct irq_info
*info
= info_for_irq(irq
);
1272 /* Make sure the irq is masked, since the new event channel
1273 will also be masked. */
1276 mutex_lock(&irq_mapping_update_lock
);
1278 /* After resume the irq<->evtchn mappings are all cleared out */
1279 BUG_ON(get_evtchn_to_irq(evtchn
) != -1);
1280 /* Expect irq to have been bound before,
1281 so there should be a proper type */
1282 BUG_ON(info
->type
== IRQT_UNBOUND
);
1284 (void)xen_irq_info_evtchn_setup(irq
, evtchn
);
1286 mutex_unlock(&irq_mapping_update_lock
);
1288 bind_evtchn_to_cpu(evtchn
, info
->cpu
);
1289 /* This will be deferred until interrupt is processed */
1290 irq_set_affinity(irq
, cpumask_of(info
->cpu
));
1292 /* Unmask the event channel. */
1296 /* Rebind an evtchn so that it gets delivered to a specific cpu */
1297 static int rebind_irq_to_cpu(unsigned irq
, unsigned tcpu
)
1299 struct evtchn_bind_vcpu bind_vcpu
;
1300 int evtchn
= evtchn_from_irq(irq
);
1303 if (!VALID_EVTCHN(evtchn
))
1307 * Events delivered via platform PCI interrupts are always
1308 * routed to vcpu 0 and hence cannot be rebound.
1310 if (xen_hvm_domain() && !xen_have_vector_callback
)
1313 /* Send future instances of this interrupt to other vcpu. */
1314 bind_vcpu
.port
= evtchn
;
1315 bind_vcpu
.vcpu
= tcpu
;
1318 * Mask the event while changing the VCPU binding to prevent
1319 * it being delivered on an unexpected VCPU.
1321 masked
= test_and_set_mask(evtchn
);
1324 * If this fails, it usually just indicates that we're dealing with a
1325 * virq or IPI channel, which don't actually need to be rebound. Ignore
1326 * it, but don't do the xenlinux-level rebind in that case.
1328 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_vcpu
, &bind_vcpu
) >= 0)
1329 bind_evtchn_to_cpu(evtchn
, tcpu
);
1332 unmask_evtchn(evtchn
);
1337 static int set_affinity_irq(struct irq_data
*data
, const struct cpumask
*dest
,
1340 unsigned tcpu
= cpumask_first_and(dest
, cpu_online_mask
);
1342 return rebind_irq_to_cpu(data
->irq
, tcpu
);
1345 static void enable_dynirq(struct irq_data
*data
)
1347 int evtchn
= evtchn_from_irq(data
->irq
);
1349 if (VALID_EVTCHN(evtchn
))
1350 unmask_evtchn(evtchn
);
1353 static void disable_dynirq(struct irq_data
*data
)
1355 int evtchn
= evtchn_from_irq(data
->irq
);
1357 if (VALID_EVTCHN(evtchn
))
1358 mask_evtchn(evtchn
);
1361 static void ack_dynirq(struct irq_data
*data
)
1363 int evtchn
= evtchn_from_irq(data
->irq
);
1367 if (VALID_EVTCHN(evtchn
))
1368 clear_evtchn(evtchn
);
1371 static void mask_ack_dynirq(struct irq_data
*data
)
1373 disable_dynirq(data
);
1377 static int retrigger_dynirq(struct irq_data
*data
)
1379 unsigned int evtchn
= evtchn_from_irq(data
->irq
);
1382 if (!VALID_EVTCHN(evtchn
))
1385 masked
= test_and_set_mask(evtchn
);
1388 unmask_evtchn(evtchn
);
1393 static void restore_pirqs(void)
1395 int pirq
, rc
, irq
, gsi
;
1396 struct physdev_map_pirq map_irq
;
1397 struct irq_info
*info
;
1399 list_for_each_entry(info
, &xen_irq_list_head
, list
) {
1400 if (info
->type
!= IRQT_PIRQ
)
1403 pirq
= info
->u
.pirq
.pirq
;
1404 gsi
= info
->u
.pirq
.gsi
;
1407 /* save/restore of PT devices doesn't work, so at this point the
1408 * only devices present are GSI based emulated devices */
1412 map_irq
.domid
= DOMID_SELF
;
1413 map_irq
.type
= MAP_PIRQ_TYPE_GSI
;
1414 map_irq
.index
= gsi
;
1415 map_irq
.pirq
= pirq
;
1417 rc
= HYPERVISOR_physdev_op(PHYSDEVOP_map_pirq
, &map_irq
);
1419 pr_warn("xen map irq failed gsi=%d irq=%d pirq=%d rc=%d\n",
1420 gsi
, irq
, pirq
, rc
);
1425 printk(KERN_DEBUG
"xen: --> irq=%d, pirq=%d\n", irq
, map_irq
.pirq
);
1427 __startup_pirq(irq
);
1431 static void restore_cpu_virqs(unsigned int cpu
)
1433 struct evtchn_bind_virq bind_virq
;
1434 int virq
, irq
, evtchn
;
1436 for (virq
= 0; virq
< NR_VIRQS
; virq
++) {
1437 if ((irq
= per_cpu(virq_to_irq
, cpu
)[virq
]) == -1)
1440 BUG_ON(virq_from_irq(irq
) != virq
);
1442 /* Get a new binding from Xen. */
1443 bind_virq
.virq
= virq
;
1444 bind_virq
.vcpu
= cpu
;
1445 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_virq
,
1448 evtchn
= bind_virq
.port
;
1450 /* Record the new mapping. */
1451 (void)xen_irq_info_virq_setup(cpu
, irq
, evtchn
, virq
);
1452 bind_evtchn_to_cpu(evtchn
, cpu
);
1456 static void restore_cpu_ipis(unsigned int cpu
)
1458 struct evtchn_bind_ipi bind_ipi
;
1459 int ipi
, irq
, evtchn
;
1461 for (ipi
= 0; ipi
< XEN_NR_IPIS
; ipi
++) {
1462 if ((irq
= per_cpu(ipi_to_irq
, cpu
)[ipi
]) == -1)
1465 BUG_ON(ipi_from_irq(irq
) != ipi
);
1467 /* Get a new binding from Xen. */
1468 bind_ipi
.vcpu
= cpu
;
1469 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_ipi
,
1472 evtchn
= bind_ipi
.port
;
1474 /* Record the new mapping. */
1475 (void)xen_irq_info_ipi_setup(cpu
, irq
, evtchn
, ipi
);
1476 bind_evtchn_to_cpu(evtchn
, cpu
);
1480 /* Clear an irq's pending state, in preparation for polling on it */
1481 void xen_clear_irq_pending(int irq
)
1483 int evtchn
= evtchn_from_irq(irq
);
1485 if (VALID_EVTCHN(evtchn
))
1486 clear_evtchn(evtchn
);
1488 EXPORT_SYMBOL(xen_clear_irq_pending
);
1489 void xen_set_irq_pending(int irq
)
1491 int evtchn
= evtchn_from_irq(irq
);
1493 if (VALID_EVTCHN(evtchn
))
1497 bool xen_test_irq_pending(int irq
)
1499 int evtchn
= evtchn_from_irq(irq
);
1502 if (VALID_EVTCHN(evtchn
))
1503 ret
= test_evtchn(evtchn
);
1508 /* Poll waiting for an irq to become pending with timeout. In the usual case,
1509 * the irq will be disabled so it won't deliver an interrupt. */
1510 void xen_poll_irq_timeout(int irq
, u64 timeout
)
1512 evtchn_port_t evtchn
= evtchn_from_irq(irq
);
1514 if (VALID_EVTCHN(evtchn
)) {
1515 struct sched_poll poll
;
1518 poll
.timeout
= timeout
;
1519 set_xen_guest_handle(poll
.ports
, &evtchn
);
1521 if (HYPERVISOR_sched_op(SCHEDOP_poll
, &poll
) != 0)
1525 EXPORT_SYMBOL(xen_poll_irq_timeout
);
1526 /* Poll waiting for an irq to become pending. In the usual case, the
1527 * irq will be disabled so it won't deliver an interrupt. */
1528 void xen_poll_irq(int irq
)
1530 xen_poll_irq_timeout(irq
, 0 /* no timeout */);
1533 /* Check whether the IRQ line is shared with other guests. */
1534 int xen_test_irq_shared(int irq
)
1536 struct irq_info
*info
= info_for_irq(irq
);
1537 struct physdev_irq_status_query irq_status
;
1542 irq_status
.irq
= info
->u
.pirq
.pirq
;
1544 if (HYPERVISOR_physdev_op(PHYSDEVOP_irq_status_query
, &irq_status
))
1546 return !(irq_status
.flags
& XENIRQSTAT_shared
);
1548 EXPORT_SYMBOL_GPL(xen_test_irq_shared
);
1550 void xen_irq_resume(void)
1553 struct irq_info
*info
;
1555 /* New event-channel space is not 'live' yet. */
1556 xen_evtchn_mask_all();
1557 xen_evtchn_resume();
1559 /* No IRQ <-> event-channel mappings. */
1560 list_for_each_entry(info
, &xen_irq_list_head
, list
)
1561 info
->evtchn
= 0; /* zap event-channel binding */
1563 clear_evtchn_to_irq_all();
1565 for_each_possible_cpu(cpu
) {
1566 restore_cpu_virqs(cpu
);
1567 restore_cpu_ipis(cpu
);
1573 static struct irq_chip xen_dynamic_chip __read_mostly
= {
1576 .irq_disable
= disable_dynirq
,
1577 .irq_mask
= disable_dynirq
,
1578 .irq_unmask
= enable_dynirq
,
1580 .irq_ack
= ack_dynirq
,
1581 .irq_mask_ack
= mask_ack_dynirq
,
1583 .irq_set_affinity
= set_affinity_irq
,
1584 .irq_retrigger
= retrigger_dynirq
,
1587 static struct irq_chip xen_pirq_chip __read_mostly
= {
1590 .irq_startup
= startup_pirq
,
1591 .irq_shutdown
= shutdown_pirq
,
1592 .irq_enable
= enable_pirq
,
1593 .irq_disable
= disable_pirq
,
1595 .irq_mask
= disable_dynirq
,
1596 .irq_unmask
= enable_dynirq
,
1598 .irq_ack
= eoi_pirq
,
1599 .irq_eoi
= eoi_pirq
,
1600 .irq_mask_ack
= mask_ack_pirq
,
1602 .irq_set_affinity
= set_affinity_irq
,
1604 .irq_retrigger
= retrigger_dynirq
,
1607 static struct irq_chip xen_percpu_chip __read_mostly
= {
1608 .name
= "xen-percpu",
1610 .irq_disable
= disable_dynirq
,
1611 .irq_mask
= disable_dynirq
,
1612 .irq_unmask
= enable_dynirq
,
1614 .irq_ack
= ack_dynirq
,
1617 int xen_set_callback_via(uint64_t via
)
1619 struct xen_hvm_param a
;
1620 a
.domid
= DOMID_SELF
;
1621 a
.index
= HVM_PARAM_CALLBACK_IRQ
;
1623 return HYPERVISOR_hvm_op(HVMOP_set_param
, &a
);
1625 EXPORT_SYMBOL_GPL(xen_set_callback_via
);
1627 #ifdef CONFIG_XEN_PVHVM
1628 /* Vector callbacks are better than PCI interrupts to receive event
1629 * channel notifications because we can receive vector callbacks on any
1630 * vcpu and we don't need PCI support or APIC interactions. */
1631 void xen_callback_vector(void)
1634 uint64_t callback_via
;
1635 if (xen_have_vector_callback
) {
1636 callback_via
= HVM_CALLBACK_VECTOR(HYPERVISOR_CALLBACK_VECTOR
);
1637 rc
= xen_set_callback_via(callback_via
);
1639 pr_err("Request for Xen HVM callback vector failed\n");
1640 xen_have_vector_callback
= 0;
1643 pr_info("Xen HVM callback vector for event delivery is enabled\n");
1644 /* in the restore case the vector has already been allocated */
1645 if (!test_bit(HYPERVISOR_CALLBACK_VECTOR
, used_vectors
))
1646 alloc_intr_gate(HYPERVISOR_CALLBACK_VECTOR
,
1647 xen_hvm_callback_vector
);
1651 void xen_callback_vector(void) {}
1654 #undef MODULE_PARAM_PREFIX
1655 #define MODULE_PARAM_PREFIX "xen."
1657 static bool fifo_events
= true;
1658 module_param(fifo_events
, bool, 0);
1660 void __init
xen_init_IRQ(void)
1665 ret
= xen_evtchn_fifo_init();
1667 xen_evtchn_2l_init();
1669 evtchn_to_irq
= kcalloc(EVTCHN_ROW(xen_evtchn_max_channels()),
1670 sizeof(*evtchn_to_irq
), GFP_KERNEL
);
1671 BUG_ON(!evtchn_to_irq
);
1673 /* No event channels are 'live' right now. */
1674 xen_evtchn_mask_all();
1676 pirq_needs_eoi
= pirq_needs_eoi_flag
;
1679 if (xen_pv_domain()) {
1680 irq_ctx_init(smp_processor_id());
1681 if (xen_initial_domain())
1682 pci_xen_initial_domain();
1684 if (xen_feature(XENFEAT_hvm_callback_vector
))
1685 xen_callback_vector();
1687 if (xen_hvm_domain()) {
1689 /* pci_xen_hvm_init must be called after native_init_IRQ so that
1690 * __acpi_register_gsi can point at the right function */
1694 struct physdev_pirq_eoi_gmfn eoi_gmfn
;
1696 pirq_eoi_map
= (void *)__get_free_page(GFP_KERNEL
|__GFP_ZERO
);
1697 eoi_gmfn
.gmfn
= virt_to_mfn(pirq_eoi_map
);
1698 rc
= HYPERVISOR_physdev_op(PHYSDEVOP_pirq_eoi_gmfn_v2
, &eoi_gmfn
);
1699 /* TODO: No PVH support for PIRQ EOI */
1701 free_page((unsigned long) pirq_eoi_map
);
1702 pirq_eoi_map
= NULL
;
1704 pirq_needs_eoi
= pirq_check_eoi_map
;