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/page.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>
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
< 0 || 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_irq_dynamic(void)
396 #ifdef CONFIG_X86_IO_APIC
398 * For an HVM guest or domain 0 which see "real" (emulated or
399 * actual respectively) GSIs we allocate dynamic IRQs
400 * e.g. those corresponding to event channels or MSIs
401 * etc. from the range above those "real" GSIs to avoid
404 if (xen_initial_domain() || xen_hvm_domain())
405 first
= get_nr_irqs_gsi();
408 irq
= irq_alloc_desc_from(first
, -1);
416 static int __must_check
xen_allocate_irq_gsi(unsigned gsi
)
421 * A PV guest has no concept of a GSI (since it has no ACPI
422 * nor access to/knowledge of the physical APICs). Therefore
423 * all IRQs are dynamically allocated from the entire IRQ
426 if (xen_pv_domain() && !xen_initial_domain())
427 return xen_allocate_irq_dynamic();
429 /* Legacy IRQ descriptors are already allocated by the arch. */
430 if (gsi
< NR_IRQS_LEGACY
)
433 irq
= irq_alloc_desc_at(gsi
, -1);
440 static void xen_free_irq(unsigned irq
)
442 struct irq_info
*info
= irq_get_handler_data(irq
);
447 list_del(&info
->list
);
449 irq_set_handler_data(irq
, NULL
);
451 WARN_ON(info
->refcnt
> 0);
455 /* Legacy IRQ descriptors are managed by the arch. */
456 if (irq
< NR_IRQS_LEGACY
)
462 static void xen_evtchn_close(unsigned int port
)
464 struct evtchn_close close
;
467 if (HYPERVISOR_event_channel_op(EVTCHNOP_close
, &close
) != 0)
470 /* Closed ports are implicitly re-bound to VCPU0. */
471 bind_evtchn_to_cpu(port
, 0);
474 static void pirq_query_unmask(int irq
)
476 struct physdev_irq_status_query irq_status
;
477 struct irq_info
*info
= info_for_irq(irq
);
479 BUG_ON(info
->type
!= IRQT_PIRQ
);
481 irq_status
.irq
= pirq_from_irq(irq
);
482 if (HYPERVISOR_physdev_op(PHYSDEVOP_irq_status_query
, &irq_status
))
483 irq_status
.flags
= 0;
485 info
->u
.pirq
.flags
&= ~PIRQ_NEEDS_EOI
;
486 if (irq_status
.flags
& XENIRQSTAT_needs_eoi
)
487 info
->u
.pirq
.flags
|= PIRQ_NEEDS_EOI
;
490 static void eoi_pirq(struct irq_data
*data
)
492 int evtchn
= evtchn_from_irq(data
->irq
);
493 struct physdev_eoi eoi
= { .irq
= pirq_from_irq(data
->irq
) };
498 if (VALID_EVTCHN(evtchn
))
499 clear_evtchn(evtchn
);
501 if (pirq_needs_eoi(data
->irq
)) {
502 rc
= HYPERVISOR_physdev_op(PHYSDEVOP_eoi
, &eoi
);
507 static void mask_ack_pirq(struct irq_data
*data
)
509 disable_dynirq(data
);
513 static unsigned int __startup_pirq(unsigned int irq
)
515 struct evtchn_bind_pirq bind_pirq
;
516 struct irq_info
*info
= info_for_irq(irq
);
517 int evtchn
= evtchn_from_irq(irq
);
520 BUG_ON(info
->type
!= IRQT_PIRQ
);
522 if (VALID_EVTCHN(evtchn
))
525 bind_pirq
.pirq
= pirq_from_irq(irq
);
526 /* NB. We are happy to share unless we are probing. */
527 bind_pirq
.flags
= info
->u
.pirq
.flags
& PIRQ_SHAREABLE
?
528 BIND_PIRQ__WILL_SHARE
: 0;
529 rc
= HYPERVISOR_event_channel_op(EVTCHNOP_bind_pirq
, &bind_pirq
);
531 pr_warn("Failed to obtain physical IRQ %d\n", irq
);
534 evtchn
= bind_pirq
.port
;
536 pirq_query_unmask(irq
);
538 rc
= set_evtchn_to_irq(evtchn
, irq
);
540 pr_err("irq%d: Failed to set port to irq mapping (%d)\n",
542 xen_evtchn_close(evtchn
);
545 bind_evtchn_to_cpu(evtchn
, 0);
546 info
->evtchn
= evtchn
;
549 unmask_evtchn(evtchn
);
550 eoi_pirq(irq_get_irq_data(irq
));
555 static unsigned int startup_pirq(struct irq_data
*data
)
557 return __startup_pirq(data
->irq
);
560 static void shutdown_pirq(struct irq_data
*data
)
562 unsigned int irq
= data
->irq
;
563 struct irq_info
*info
= info_for_irq(irq
);
564 unsigned evtchn
= evtchn_from_irq(irq
);
566 BUG_ON(info
->type
!= IRQT_PIRQ
);
568 if (!VALID_EVTCHN(evtchn
))
572 xen_evtchn_close(evtchn
);
573 xen_irq_info_cleanup(info
);
576 static void enable_pirq(struct irq_data
*data
)
581 static void disable_pirq(struct irq_data
*data
)
583 disable_dynirq(data
);
586 int xen_irq_from_gsi(unsigned gsi
)
588 struct irq_info
*info
;
590 list_for_each_entry(info
, &xen_irq_list_head
, list
) {
591 if (info
->type
!= IRQT_PIRQ
)
594 if (info
->u
.pirq
.gsi
== gsi
)
600 EXPORT_SYMBOL_GPL(xen_irq_from_gsi
);
602 static void __unbind_from_irq(unsigned int irq
)
604 int evtchn
= evtchn_from_irq(irq
);
605 struct irq_info
*info
= irq_get_handler_data(irq
);
607 if (info
->refcnt
> 0) {
609 if (info
->refcnt
!= 0)
613 if (VALID_EVTCHN(evtchn
)) {
614 unsigned int cpu
= cpu_from_irq(irq
);
616 xen_evtchn_close(evtchn
);
618 switch (type_from_irq(irq
)) {
620 per_cpu(virq_to_irq
, cpu
)[virq_from_irq(irq
)] = -1;
623 per_cpu(ipi_to_irq
, cpu
)[ipi_from_irq(irq
)] = -1;
629 xen_irq_info_cleanup(info
);
632 BUG_ON(info_for_irq(irq
)->type
== IRQT_UNBOUND
);
638 * Do not make any assumptions regarding the relationship between the
639 * IRQ number returned here and the Xen pirq argument.
641 * Note: We don't assign an event channel until the irq actually started
642 * up. Return an existing irq if we've already got one for the gsi.
644 * Shareable implies level triggered, not shareable implies edge
647 int xen_bind_pirq_gsi_to_irq(unsigned gsi
,
648 unsigned pirq
, int shareable
, char *name
)
651 struct physdev_irq irq_op
;
654 mutex_lock(&irq_mapping_update_lock
);
656 irq
= xen_irq_from_gsi(gsi
);
658 pr_info("%s: returning irq %d for gsi %u\n",
663 irq
= xen_allocate_irq_gsi(gsi
);
670 /* Only the privileged domain can do this. For non-priv, the pcifront
671 * driver provides a PCI bus that does the call to do exactly
672 * this in the priv domain. */
673 if (xen_initial_domain() &&
674 HYPERVISOR_physdev_op(PHYSDEVOP_alloc_irq_vector
, &irq_op
)) {
680 ret
= xen_irq_info_pirq_setup(irq
, 0, pirq
, gsi
, DOMID_SELF
,
681 shareable
? PIRQ_SHAREABLE
: 0);
683 __unbind_from_irq(irq
);
688 pirq_query_unmask(irq
);
689 /* We try to use the handler with the appropriate semantic for the
690 * type of interrupt: if the interrupt is an edge triggered
691 * interrupt we use handle_edge_irq.
693 * On the other hand if the interrupt is level triggered we use
694 * handle_fasteoi_irq like the native code does for this kind of
697 * Depending on the Xen version, pirq_needs_eoi might return true
698 * not only for level triggered interrupts but for edge triggered
699 * interrupts too. In any case Xen always honors the eoi mechanism,
700 * not injecting any more pirqs of the same kind if the first one
701 * hasn't received an eoi yet. Therefore using the fasteoi handler
702 * is the right choice either way.
705 irq_set_chip_and_handler_name(irq
, &xen_pirq_chip
,
706 handle_fasteoi_irq
, name
);
708 irq_set_chip_and_handler_name(irq
, &xen_pirq_chip
,
709 handle_edge_irq
, name
);
712 mutex_unlock(&irq_mapping_update_lock
);
717 #ifdef CONFIG_PCI_MSI
718 int xen_allocate_pirq_msi(struct pci_dev
*dev
, struct msi_desc
*msidesc
)
721 struct physdev_get_free_pirq op_get_free_pirq
;
723 op_get_free_pirq
.type
= MAP_PIRQ_TYPE_MSI
;
724 rc
= HYPERVISOR_physdev_op(PHYSDEVOP_get_free_pirq
, &op_get_free_pirq
);
726 WARN_ONCE(rc
== -ENOSYS
,
727 "hypervisor does not support the PHYSDEVOP_get_free_pirq interface\n");
729 return rc
? -1 : op_get_free_pirq
.pirq
;
732 int xen_bind_pirq_msi_to_irq(struct pci_dev
*dev
, struct msi_desc
*msidesc
,
733 int pirq
, const char *name
, domid_t domid
)
737 mutex_lock(&irq_mapping_update_lock
);
739 irq
= xen_allocate_irq_dynamic();
743 irq_set_chip_and_handler_name(irq
, &xen_pirq_chip
, handle_edge_irq
,
746 ret
= xen_irq_info_pirq_setup(irq
, 0, pirq
, 0, domid
, 0);
749 ret
= irq_set_msi_desc(irq
, msidesc
);
753 mutex_unlock(&irq_mapping_update_lock
);
756 __unbind_from_irq(irq
);
757 mutex_unlock(&irq_mapping_update_lock
);
762 int xen_destroy_irq(int irq
)
764 struct physdev_unmap_pirq unmap_irq
;
765 struct irq_info
*info
= info_for_irq(irq
);
768 mutex_lock(&irq_mapping_update_lock
);
770 if (xen_initial_domain()) {
771 unmap_irq
.pirq
= info
->u
.pirq
.pirq
;
772 unmap_irq
.domid
= info
->u
.pirq
.domid
;
773 rc
= HYPERVISOR_physdev_op(PHYSDEVOP_unmap_pirq
, &unmap_irq
);
774 /* If another domain quits without making the pci_disable_msix
775 * call, the Xen hypervisor takes care of freeing the PIRQs
776 * (free_domain_pirqs).
778 if ((rc
== -ESRCH
&& info
->u
.pirq
.domid
!= DOMID_SELF
))
779 pr_info("domain %d does not have %d anymore\n",
780 info
->u
.pirq
.domid
, info
->u
.pirq
.pirq
);
782 pr_warn("unmap irq failed %d\n", rc
);
790 mutex_unlock(&irq_mapping_update_lock
);
794 int xen_irq_from_pirq(unsigned pirq
)
798 struct irq_info
*info
;
800 mutex_lock(&irq_mapping_update_lock
);
802 list_for_each_entry(info
, &xen_irq_list_head
, list
) {
803 if (info
->type
!= IRQT_PIRQ
)
806 if (info
->u
.pirq
.pirq
== pirq
)
811 mutex_unlock(&irq_mapping_update_lock
);
817 int xen_pirq_from_irq(unsigned irq
)
819 return pirq_from_irq(irq
);
821 EXPORT_SYMBOL_GPL(xen_pirq_from_irq
);
823 int bind_evtchn_to_irq(unsigned int evtchn
)
828 if (evtchn
>= xen_evtchn_max_channels())
831 mutex_lock(&irq_mapping_update_lock
);
833 irq
= get_evtchn_to_irq(evtchn
);
836 irq
= xen_allocate_irq_dynamic();
840 irq_set_chip_and_handler_name(irq
, &xen_dynamic_chip
,
841 handle_edge_irq
, "event");
843 ret
= xen_irq_info_evtchn_setup(irq
, evtchn
);
845 __unbind_from_irq(irq
);
849 /* New interdomain events are bound to VCPU 0. */
850 bind_evtchn_to_cpu(evtchn
, 0);
852 struct irq_info
*info
= info_for_irq(irq
);
853 WARN_ON(info
== NULL
|| info
->type
!= IRQT_EVTCHN
);
857 mutex_unlock(&irq_mapping_update_lock
);
861 EXPORT_SYMBOL_GPL(bind_evtchn_to_irq
);
863 static int bind_ipi_to_irq(unsigned int ipi
, unsigned int cpu
)
865 struct evtchn_bind_ipi bind_ipi
;
869 mutex_lock(&irq_mapping_update_lock
);
871 irq
= per_cpu(ipi_to_irq
, cpu
)[ipi
];
874 irq
= xen_allocate_irq_dynamic();
878 irq_set_chip_and_handler_name(irq
, &xen_percpu_chip
,
879 handle_percpu_irq
, "ipi");
882 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_ipi
,
885 evtchn
= bind_ipi
.port
;
887 ret
= xen_irq_info_ipi_setup(cpu
, irq
, evtchn
, ipi
);
889 __unbind_from_irq(irq
);
893 bind_evtchn_to_cpu(evtchn
, cpu
);
895 struct irq_info
*info
= info_for_irq(irq
);
896 WARN_ON(info
== NULL
|| info
->type
!= IRQT_IPI
);
900 mutex_unlock(&irq_mapping_update_lock
);
904 static int bind_interdomain_evtchn_to_irq(unsigned int remote_domain
,
905 unsigned int remote_port
)
907 struct evtchn_bind_interdomain bind_interdomain
;
910 bind_interdomain
.remote_dom
= remote_domain
;
911 bind_interdomain
.remote_port
= remote_port
;
913 err
= HYPERVISOR_event_channel_op(EVTCHNOP_bind_interdomain
,
916 return err
? : bind_evtchn_to_irq(bind_interdomain
.local_port
);
919 static int find_virq(unsigned int virq
, unsigned int cpu
)
921 struct evtchn_status status
;
922 int port
, rc
= -ENOENT
;
924 memset(&status
, 0, sizeof(status
));
925 for (port
= 0; port
< xen_evtchn_max_channels(); port
++) {
926 status
.dom
= DOMID_SELF
;
928 rc
= HYPERVISOR_event_channel_op(EVTCHNOP_status
, &status
);
931 if (status
.status
!= EVTCHNSTAT_virq
)
933 if (status
.u
.virq
== virq
&& status
.vcpu
== cpu
) {
942 * xen_evtchn_nr_channels - number of usable event channel ports
944 * This may be less than the maximum supported by the current
945 * hypervisor ABI. Use xen_evtchn_max_channels() for the maximum
948 unsigned xen_evtchn_nr_channels(void)
950 return evtchn_ops
->nr_channels();
952 EXPORT_SYMBOL_GPL(xen_evtchn_nr_channels
);
954 int bind_virq_to_irq(unsigned int virq
, unsigned int cpu
)
956 struct evtchn_bind_virq bind_virq
;
957 int evtchn
, irq
, ret
;
959 mutex_lock(&irq_mapping_update_lock
);
961 irq
= per_cpu(virq_to_irq
, cpu
)[virq
];
964 irq
= xen_allocate_irq_dynamic();
968 irq_set_chip_and_handler_name(irq
, &xen_percpu_chip
,
969 handle_percpu_irq
, "virq");
971 bind_virq
.virq
= virq
;
972 bind_virq
.vcpu
= cpu
;
973 ret
= HYPERVISOR_event_channel_op(EVTCHNOP_bind_virq
,
976 evtchn
= bind_virq
.port
;
979 ret
= find_virq(virq
, cpu
);
984 ret
= xen_irq_info_virq_setup(cpu
, irq
, evtchn
, virq
);
986 __unbind_from_irq(irq
);
991 bind_evtchn_to_cpu(evtchn
, cpu
);
993 struct irq_info
*info
= info_for_irq(irq
);
994 WARN_ON(info
== NULL
|| info
->type
!= IRQT_VIRQ
);
998 mutex_unlock(&irq_mapping_update_lock
);
1003 static void unbind_from_irq(unsigned int irq
)
1005 mutex_lock(&irq_mapping_update_lock
);
1006 __unbind_from_irq(irq
);
1007 mutex_unlock(&irq_mapping_update_lock
);
1010 int bind_evtchn_to_irqhandler(unsigned int evtchn
,
1011 irq_handler_t handler
,
1012 unsigned long irqflags
,
1013 const char *devname
, void *dev_id
)
1017 irq
= bind_evtchn_to_irq(evtchn
);
1020 retval
= request_irq(irq
, handler
, irqflags
, devname
, dev_id
);
1022 unbind_from_irq(irq
);
1028 EXPORT_SYMBOL_GPL(bind_evtchn_to_irqhandler
);
1030 int bind_interdomain_evtchn_to_irqhandler(unsigned int remote_domain
,
1031 unsigned int remote_port
,
1032 irq_handler_t handler
,
1033 unsigned long irqflags
,
1034 const char *devname
,
1039 irq
= bind_interdomain_evtchn_to_irq(remote_domain
, remote_port
);
1043 retval
= request_irq(irq
, handler
, irqflags
, devname
, dev_id
);
1045 unbind_from_irq(irq
);
1051 EXPORT_SYMBOL_GPL(bind_interdomain_evtchn_to_irqhandler
);
1053 int bind_virq_to_irqhandler(unsigned int virq
, unsigned int cpu
,
1054 irq_handler_t handler
,
1055 unsigned long irqflags
, const char *devname
, void *dev_id
)
1059 irq
= bind_virq_to_irq(virq
, cpu
);
1062 retval
= request_irq(irq
, handler
, irqflags
, devname
, dev_id
);
1064 unbind_from_irq(irq
);
1070 EXPORT_SYMBOL_GPL(bind_virq_to_irqhandler
);
1072 int bind_ipi_to_irqhandler(enum ipi_vector ipi
,
1074 irq_handler_t handler
,
1075 unsigned long irqflags
,
1076 const char *devname
,
1081 irq
= bind_ipi_to_irq(ipi
, cpu
);
1085 irqflags
|= IRQF_NO_SUSPEND
| IRQF_FORCE_RESUME
| IRQF_EARLY_RESUME
;
1086 retval
= request_irq(irq
, handler
, irqflags
, devname
, dev_id
);
1088 unbind_from_irq(irq
);
1095 void unbind_from_irqhandler(unsigned int irq
, void *dev_id
)
1097 struct irq_info
*info
= irq_get_handler_data(irq
);
1101 free_irq(irq
, dev_id
);
1102 unbind_from_irq(irq
);
1104 EXPORT_SYMBOL_GPL(unbind_from_irqhandler
);
1107 * xen_set_irq_priority() - set an event channel priority.
1108 * @irq:irq bound to an event channel.
1109 * @priority: priority between XEN_IRQ_PRIORITY_MAX and XEN_IRQ_PRIORITY_MIN.
1111 int xen_set_irq_priority(unsigned irq
, unsigned priority
)
1113 struct evtchn_set_priority set_priority
;
1115 set_priority
.port
= evtchn_from_irq(irq
);
1116 set_priority
.priority
= priority
;
1118 return HYPERVISOR_event_channel_op(EVTCHNOP_set_priority
,
1121 EXPORT_SYMBOL_GPL(xen_set_irq_priority
);
1123 int evtchn_make_refcounted(unsigned int evtchn
)
1125 int irq
= get_evtchn_to_irq(evtchn
);
1126 struct irq_info
*info
;
1131 info
= irq_get_handler_data(irq
);
1136 WARN_ON(info
->refcnt
!= -1);
1142 EXPORT_SYMBOL_GPL(evtchn_make_refcounted
);
1144 int evtchn_get(unsigned int evtchn
)
1147 struct irq_info
*info
;
1150 if (evtchn
>= xen_evtchn_max_channels())
1153 mutex_lock(&irq_mapping_update_lock
);
1155 irq
= get_evtchn_to_irq(evtchn
);
1159 info
= irq_get_handler_data(irq
);
1165 if (info
->refcnt
<= 0)
1171 mutex_unlock(&irq_mapping_update_lock
);
1175 EXPORT_SYMBOL_GPL(evtchn_get
);
1177 void evtchn_put(unsigned int evtchn
)
1179 int irq
= get_evtchn_to_irq(evtchn
);
1180 if (WARN_ON(irq
== -1))
1182 unbind_from_irq(irq
);
1184 EXPORT_SYMBOL_GPL(evtchn_put
);
1186 void xen_send_IPI_one(unsigned int cpu
, enum ipi_vector vector
)
1191 if (unlikely(vector
== XEN_NMI_VECTOR
)) {
1192 int rc
= HYPERVISOR_vcpu_op(VCPUOP_send_nmi
, cpu
, NULL
);
1194 printk(KERN_WARNING
"Sending nmi to CPU%d failed (rc:%d)\n", cpu
, rc
);
1198 irq
= per_cpu(ipi_to_irq
, cpu
)[vector
];
1200 notify_remote_via_irq(irq
);
1203 static DEFINE_PER_CPU(unsigned, xed_nesting_count
);
1205 static void __xen_evtchn_do_upcall(void)
1207 struct vcpu_info
*vcpu_info
= __this_cpu_read(xen_vcpu
);
1208 int cpu
= get_cpu();
1212 vcpu_info
->evtchn_upcall_pending
= 0;
1214 if (__this_cpu_inc_return(xed_nesting_count
) - 1)
1217 xen_evtchn_handle_events(cpu
);
1219 BUG_ON(!irqs_disabled());
1221 count
= __this_cpu_read(xed_nesting_count
);
1222 __this_cpu_write(xed_nesting_count
, 0);
1223 } while (count
!= 1 || vcpu_info
->evtchn_upcall_pending
);
1230 void xen_evtchn_do_upcall(struct pt_regs
*regs
)
1232 struct pt_regs
*old_regs
= set_irq_regs(regs
);
1238 inc_irq_stat(irq_hv_callback_count
);
1240 __xen_evtchn_do_upcall();
1243 set_irq_regs(old_regs
);
1246 void xen_hvm_evtchn_do_upcall(void)
1248 __xen_evtchn_do_upcall();
1250 EXPORT_SYMBOL_GPL(xen_hvm_evtchn_do_upcall
);
1252 /* Rebind a new event channel to an existing irq. */
1253 void rebind_evtchn_irq(int evtchn
, int irq
)
1255 struct irq_info
*info
= info_for_irq(irq
);
1260 /* Make sure the irq is masked, since the new event channel
1261 will also be masked. */
1264 mutex_lock(&irq_mapping_update_lock
);
1266 /* After resume the irq<->evtchn mappings are all cleared out */
1267 BUG_ON(get_evtchn_to_irq(evtchn
) != -1);
1268 /* Expect irq to have been bound before,
1269 so there should be a proper type */
1270 BUG_ON(info
->type
== IRQT_UNBOUND
);
1272 (void)xen_irq_info_evtchn_setup(irq
, evtchn
);
1274 mutex_unlock(&irq_mapping_update_lock
);
1276 /* new event channels are always bound to cpu 0 */
1277 irq_set_affinity(irq
, cpumask_of(0));
1279 /* Unmask the event channel. */
1283 /* Rebind an evtchn so that it gets delivered to a specific cpu */
1284 static int rebind_irq_to_cpu(unsigned irq
, unsigned tcpu
)
1286 struct evtchn_bind_vcpu bind_vcpu
;
1287 int evtchn
= evtchn_from_irq(irq
);
1290 if (!VALID_EVTCHN(evtchn
))
1294 * Events delivered via platform PCI interrupts are always
1295 * routed to vcpu 0 and hence cannot be rebound.
1297 if (xen_hvm_domain() && !xen_have_vector_callback
)
1300 /* Send future instances of this interrupt to other vcpu. */
1301 bind_vcpu
.port
= evtchn
;
1302 bind_vcpu
.vcpu
= tcpu
;
1305 * Mask the event while changing the VCPU binding to prevent
1306 * it being delivered on an unexpected VCPU.
1308 masked
= test_and_set_mask(evtchn
);
1311 * If this fails, it usually just indicates that we're dealing with a
1312 * virq or IPI channel, which don't actually need to be rebound. Ignore
1313 * it, but don't do the xenlinux-level rebind in that case.
1315 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_vcpu
, &bind_vcpu
) >= 0)
1316 bind_evtchn_to_cpu(evtchn
, tcpu
);
1319 unmask_evtchn(evtchn
);
1324 static int set_affinity_irq(struct irq_data
*data
, const struct cpumask
*dest
,
1327 unsigned tcpu
= cpumask_first_and(dest
, cpu_online_mask
);
1329 return rebind_irq_to_cpu(data
->irq
, tcpu
);
1332 static int retrigger_evtchn(int evtchn
)
1336 if (!VALID_EVTCHN(evtchn
))
1339 masked
= test_and_set_mask(evtchn
);
1342 unmask_evtchn(evtchn
);
1347 int resend_irq_on_evtchn(unsigned int irq
)
1349 return retrigger_evtchn(evtchn_from_irq(irq
));
1352 static void enable_dynirq(struct irq_data
*data
)
1354 int evtchn
= evtchn_from_irq(data
->irq
);
1356 if (VALID_EVTCHN(evtchn
))
1357 unmask_evtchn(evtchn
);
1360 static void disable_dynirq(struct irq_data
*data
)
1362 int evtchn
= evtchn_from_irq(data
->irq
);
1364 if (VALID_EVTCHN(evtchn
))
1365 mask_evtchn(evtchn
);
1368 static void ack_dynirq(struct irq_data
*data
)
1370 int evtchn
= evtchn_from_irq(data
->irq
);
1374 if (VALID_EVTCHN(evtchn
))
1375 clear_evtchn(evtchn
);
1378 static void mask_ack_dynirq(struct irq_data
*data
)
1380 disable_dynirq(data
);
1384 static int retrigger_dynirq(struct irq_data
*data
)
1386 return retrigger_evtchn(evtchn_from_irq(data
->irq
));
1389 static void restore_pirqs(void)
1391 int pirq
, rc
, irq
, gsi
;
1392 struct physdev_map_pirq map_irq
;
1393 struct irq_info
*info
;
1395 list_for_each_entry(info
, &xen_irq_list_head
, list
) {
1396 if (info
->type
!= IRQT_PIRQ
)
1399 pirq
= info
->u
.pirq
.pirq
;
1400 gsi
= info
->u
.pirq
.gsi
;
1403 /* save/restore of PT devices doesn't work, so at this point the
1404 * only devices present are GSI based emulated devices */
1408 map_irq
.domid
= DOMID_SELF
;
1409 map_irq
.type
= MAP_PIRQ_TYPE_GSI
;
1410 map_irq
.index
= gsi
;
1411 map_irq
.pirq
= pirq
;
1413 rc
= HYPERVISOR_physdev_op(PHYSDEVOP_map_pirq
, &map_irq
);
1415 pr_warn("xen map irq failed gsi=%d irq=%d pirq=%d rc=%d\n",
1416 gsi
, irq
, pirq
, rc
);
1421 printk(KERN_DEBUG
"xen: --> irq=%d, pirq=%d\n", irq
, map_irq
.pirq
);
1423 __startup_pirq(irq
);
1427 static void restore_cpu_virqs(unsigned int cpu
)
1429 struct evtchn_bind_virq bind_virq
;
1430 int virq
, irq
, evtchn
;
1432 for (virq
= 0; virq
< NR_VIRQS
; virq
++) {
1433 if ((irq
= per_cpu(virq_to_irq
, cpu
)[virq
]) == -1)
1436 BUG_ON(virq_from_irq(irq
) != virq
);
1438 /* Get a new binding from Xen. */
1439 bind_virq
.virq
= virq
;
1440 bind_virq
.vcpu
= cpu
;
1441 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_virq
,
1444 evtchn
= bind_virq
.port
;
1446 /* Record the new mapping. */
1447 (void)xen_irq_info_virq_setup(cpu
, irq
, evtchn
, virq
);
1448 bind_evtchn_to_cpu(evtchn
, cpu
);
1452 static void restore_cpu_ipis(unsigned int cpu
)
1454 struct evtchn_bind_ipi bind_ipi
;
1455 int ipi
, irq
, evtchn
;
1457 for (ipi
= 0; ipi
< XEN_NR_IPIS
; ipi
++) {
1458 if ((irq
= per_cpu(ipi_to_irq
, cpu
)[ipi
]) == -1)
1461 BUG_ON(ipi_from_irq(irq
) != ipi
);
1463 /* Get a new binding from Xen. */
1464 bind_ipi
.vcpu
= cpu
;
1465 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_ipi
,
1468 evtchn
= bind_ipi
.port
;
1470 /* Record the new mapping. */
1471 (void)xen_irq_info_ipi_setup(cpu
, irq
, evtchn
, ipi
);
1472 bind_evtchn_to_cpu(evtchn
, cpu
);
1476 /* Clear an irq's pending state, in preparation for polling on it */
1477 void xen_clear_irq_pending(int irq
)
1479 int evtchn
= evtchn_from_irq(irq
);
1481 if (VALID_EVTCHN(evtchn
))
1482 clear_evtchn(evtchn
);
1484 EXPORT_SYMBOL(xen_clear_irq_pending
);
1485 void xen_set_irq_pending(int irq
)
1487 int evtchn
= evtchn_from_irq(irq
);
1489 if (VALID_EVTCHN(evtchn
))
1493 bool xen_test_irq_pending(int irq
)
1495 int evtchn
= evtchn_from_irq(irq
);
1498 if (VALID_EVTCHN(evtchn
))
1499 ret
= test_evtchn(evtchn
);
1504 /* Poll waiting for an irq to become pending with timeout. In the usual case,
1505 * the irq will be disabled so it won't deliver an interrupt. */
1506 void xen_poll_irq_timeout(int irq
, u64 timeout
)
1508 evtchn_port_t evtchn
= evtchn_from_irq(irq
);
1510 if (VALID_EVTCHN(evtchn
)) {
1511 struct sched_poll poll
;
1514 poll
.timeout
= timeout
;
1515 set_xen_guest_handle(poll
.ports
, &evtchn
);
1517 if (HYPERVISOR_sched_op(SCHEDOP_poll
, &poll
) != 0)
1521 EXPORT_SYMBOL(xen_poll_irq_timeout
);
1522 /* Poll waiting for an irq to become pending. In the usual case, the
1523 * irq will be disabled so it won't deliver an interrupt. */
1524 void xen_poll_irq(int irq
)
1526 xen_poll_irq_timeout(irq
, 0 /* no timeout */);
1529 /* Check whether the IRQ line is shared with other guests. */
1530 int xen_test_irq_shared(int irq
)
1532 struct irq_info
*info
= info_for_irq(irq
);
1533 struct physdev_irq_status_query irq_status
;
1538 irq_status
.irq
= info
->u
.pirq
.pirq
;
1540 if (HYPERVISOR_physdev_op(PHYSDEVOP_irq_status_query
, &irq_status
))
1542 return !(irq_status
.flags
& XENIRQSTAT_shared
);
1544 EXPORT_SYMBOL_GPL(xen_test_irq_shared
);
1546 void xen_irq_resume(void)
1549 struct irq_info
*info
;
1551 /* New event-channel space is not 'live' yet. */
1552 xen_evtchn_mask_all();
1553 xen_evtchn_resume();
1555 /* No IRQ <-> event-channel mappings. */
1556 list_for_each_entry(info
, &xen_irq_list_head
, list
)
1557 info
->evtchn
= 0; /* zap event-channel binding */
1559 clear_evtchn_to_irq_all();
1561 for_each_possible_cpu(cpu
) {
1562 restore_cpu_virqs(cpu
);
1563 restore_cpu_ipis(cpu
);
1569 static struct irq_chip xen_dynamic_chip __read_mostly
= {
1572 .irq_disable
= disable_dynirq
,
1573 .irq_mask
= disable_dynirq
,
1574 .irq_unmask
= enable_dynirq
,
1576 .irq_ack
= ack_dynirq
,
1577 .irq_mask_ack
= mask_ack_dynirq
,
1579 .irq_set_affinity
= set_affinity_irq
,
1580 .irq_retrigger
= retrigger_dynirq
,
1583 static struct irq_chip xen_pirq_chip __read_mostly
= {
1586 .irq_startup
= startup_pirq
,
1587 .irq_shutdown
= shutdown_pirq
,
1588 .irq_enable
= enable_pirq
,
1589 .irq_disable
= disable_pirq
,
1591 .irq_mask
= disable_dynirq
,
1592 .irq_unmask
= enable_dynirq
,
1594 .irq_ack
= eoi_pirq
,
1595 .irq_eoi
= eoi_pirq
,
1596 .irq_mask_ack
= mask_ack_pirq
,
1598 .irq_set_affinity
= set_affinity_irq
,
1600 .irq_retrigger
= retrigger_dynirq
,
1603 static struct irq_chip xen_percpu_chip __read_mostly
= {
1604 .name
= "xen-percpu",
1606 .irq_disable
= disable_dynirq
,
1607 .irq_mask
= disable_dynirq
,
1608 .irq_unmask
= enable_dynirq
,
1610 .irq_ack
= ack_dynirq
,
1613 int xen_set_callback_via(uint64_t via
)
1615 struct xen_hvm_param a
;
1616 a
.domid
= DOMID_SELF
;
1617 a
.index
= HVM_PARAM_CALLBACK_IRQ
;
1619 return HYPERVISOR_hvm_op(HVMOP_set_param
, &a
);
1621 EXPORT_SYMBOL_GPL(xen_set_callback_via
);
1623 #ifdef CONFIG_XEN_PVHVM
1624 /* Vector callbacks are better than PCI interrupts to receive event
1625 * channel notifications because we can receive vector callbacks on any
1626 * vcpu and we don't need PCI support or APIC interactions. */
1627 void xen_callback_vector(void)
1630 uint64_t callback_via
;
1631 if (xen_have_vector_callback
) {
1632 callback_via
= HVM_CALLBACK_VECTOR(HYPERVISOR_CALLBACK_VECTOR
);
1633 rc
= xen_set_callback_via(callback_via
);
1635 pr_err("Request for Xen HVM callback vector failed\n");
1636 xen_have_vector_callback
= 0;
1639 pr_info("Xen HVM callback vector for event delivery is enabled\n");
1640 /* in the restore case the vector has already been allocated */
1641 if (!test_bit(HYPERVISOR_CALLBACK_VECTOR
, used_vectors
))
1642 alloc_intr_gate(HYPERVISOR_CALLBACK_VECTOR
,
1643 xen_hvm_callback_vector
);
1647 void xen_callback_vector(void) {}
1650 #undef MODULE_PARAM_PREFIX
1651 #define MODULE_PARAM_PREFIX "xen."
1653 static bool fifo_events
= true;
1654 module_param(fifo_events
, bool, 0);
1656 void __init
xen_init_IRQ(void)
1661 ret
= xen_evtchn_fifo_init();
1663 xen_evtchn_2l_init();
1665 evtchn_to_irq
= kcalloc(EVTCHN_ROW(xen_evtchn_max_channels()),
1666 sizeof(*evtchn_to_irq
), GFP_KERNEL
);
1667 BUG_ON(!evtchn_to_irq
);
1669 /* No event channels are 'live' right now. */
1670 xen_evtchn_mask_all();
1672 pirq_needs_eoi
= pirq_needs_eoi_flag
;
1675 if (xen_pv_domain()) {
1676 irq_ctx_init(smp_processor_id());
1677 if (xen_initial_domain())
1678 pci_xen_initial_domain();
1680 if (xen_feature(XENFEAT_hvm_callback_vector
))
1681 xen_callback_vector();
1683 if (xen_hvm_domain()) {
1685 /* pci_xen_hvm_init must be called after native_init_IRQ so that
1686 * __acpi_register_gsi can point at the right function */
1690 struct physdev_pirq_eoi_gmfn eoi_gmfn
;
1692 pirq_eoi_map
= (void *)__get_free_page(GFP_KERNEL
|__GFP_ZERO
);
1693 eoi_gmfn
.gmfn
= virt_to_mfn(pirq_eoi_map
);
1694 rc
= HYPERVISOR_physdev_op(PHYSDEVOP_pirq_eoi_gmfn_v2
, &eoi_gmfn
);
1695 /* TODO: No PVH support for PIRQ EOI */
1697 free_page((unsigned long) pirq_eoi_map
);
1698 pirq_eoi_map
= NULL
;
1700 pirq_needs_eoi
= pirq_check_eoi_map
;