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_to_desc(irq
)->irq_data
.affinity
, cpumask_of(cpu
));
342 xen_evtchn_port_bind_to_cpu(info
, cpu
);
347 static void xen_evtchn_mask_all(void)
351 for (evtchn
= 0; evtchn
< xen_evtchn_nr_channels(); evtchn
++)
356 * notify_remote_via_irq - send event to remote end of event channel via irq
357 * @irq: irq of event channel to send event to
359 * Unlike notify_remote_via_evtchn(), this is safe to use across
360 * save/restore. Notifications on a broken connection are silently
363 void notify_remote_via_irq(int irq
)
365 int evtchn
= evtchn_from_irq(irq
);
367 if (VALID_EVTCHN(evtchn
))
368 notify_remote_via_evtchn(evtchn
);
370 EXPORT_SYMBOL_GPL(notify_remote_via_irq
);
372 static void xen_irq_init(unsigned irq
)
374 struct irq_info
*info
;
376 struct irq_desc
*desc
= irq_to_desc(irq
);
378 /* By default all event channels notify CPU#0. */
379 cpumask_copy(desc
->irq_data
.affinity
, cpumask_of(0));
382 info
= kzalloc(sizeof(*info
), GFP_KERNEL
);
384 panic("Unable to allocate metadata for IRQ%d\n", irq
);
386 info
->type
= IRQT_UNBOUND
;
389 irq_set_handler_data(irq
, info
);
391 list_add_tail(&info
->list
, &xen_irq_list_head
);
394 static int __must_check
xen_allocate_irq_dynamic(void)
399 #ifdef CONFIG_X86_IO_APIC
401 * For an HVM guest or domain 0 which see "real" (emulated or
402 * actual respectively) GSIs we allocate dynamic IRQs
403 * e.g. those corresponding to event channels or MSIs
404 * etc. from the range above those "real" GSIs to avoid
407 if (xen_initial_domain() || xen_hvm_domain())
408 first
= get_nr_irqs_gsi();
411 irq
= irq_alloc_desc_from(first
, -1);
419 static int __must_check
xen_allocate_irq_gsi(unsigned gsi
)
424 * A PV guest has no concept of a GSI (since it has no ACPI
425 * nor access to/knowledge of the physical APICs). Therefore
426 * all IRQs are dynamically allocated from the entire IRQ
429 if (xen_pv_domain() && !xen_initial_domain())
430 return xen_allocate_irq_dynamic();
432 /* Legacy IRQ descriptors are already allocated by the arch. */
433 if (gsi
< NR_IRQS_LEGACY
)
436 irq
= irq_alloc_desc_at(gsi
, -1);
443 static void xen_free_irq(unsigned irq
)
445 struct irq_info
*info
= irq_get_handler_data(irq
);
450 list_del(&info
->list
);
452 irq_set_handler_data(irq
, NULL
);
454 WARN_ON(info
->refcnt
> 0);
458 /* Legacy IRQ descriptors are managed by the arch. */
459 if (irq
< NR_IRQS_LEGACY
)
465 static void xen_evtchn_close(unsigned int port
)
467 struct evtchn_close close
;
470 if (HYPERVISOR_event_channel_op(EVTCHNOP_close
, &close
) != 0)
473 /* Closed ports are implicitly re-bound to VCPU0. */
474 bind_evtchn_to_cpu(port
, 0);
477 static void pirq_query_unmask(int irq
)
479 struct physdev_irq_status_query irq_status
;
480 struct irq_info
*info
= info_for_irq(irq
);
482 BUG_ON(info
->type
!= IRQT_PIRQ
);
484 irq_status
.irq
= pirq_from_irq(irq
);
485 if (HYPERVISOR_physdev_op(PHYSDEVOP_irq_status_query
, &irq_status
))
486 irq_status
.flags
= 0;
488 info
->u
.pirq
.flags
&= ~PIRQ_NEEDS_EOI
;
489 if (irq_status
.flags
& XENIRQSTAT_needs_eoi
)
490 info
->u
.pirq
.flags
|= PIRQ_NEEDS_EOI
;
493 static bool probing_irq(int irq
)
495 struct irq_desc
*desc
= irq_to_desc(irq
);
497 return desc
&& desc
->action
== NULL
;
500 static void eoi_pirq(struct irq_data
*data
)
502 int evtchn
= evtchn_from_irq(data
->irq
);
503 struct physdev_eoi eoi
= { .irq
= pirq_from_irq(data
->irq
) };
508 if (VALID_EVTCHN(evtchn
))
509 clear_evtchn(evtchn
);
511 if (pirq_needs_eoi(data
->irq
)) {
512 rc
= HYPERVISOR_physdev_op(PHYSDEVOP_eoi
, &eoi
);
517 static void mask_ack_pirq(struct irq_data
*data
)
519 disable_dynirq(data
);
523 static unsigned int __startup_pirq(unsigned int irq
)
525 struct evtchn_bind_pirq bind_pirq
;
526 struct irq_info
*info
= info_for_irq(irq
);
527 int evtchn
= evtchn_from_irq(irq
);
530 BUG_ON(info
->type
!= IRQT_PIRQ
);
532 if (VALID_EVTCHN(evtchn
))
535 bind_pirq
.pirq
= pirq_from_irq(irq
);
536 /* NB. We are happy to share unless we are probing. */
537 bind_pirq
.flags
= info
->u
.pirq
.flags
& PIRQ_SHAREABLE
?
538 BIND_PIRQ__WILL_SHARE
: 0;
539 rc
= HYPERVISOR_event_channel_op(EVTCHNOP_bind_pirq
, &bind_pirq
);
541 if (!probing_irq(irq
))
542 pr_info("Failed to obtain physical IRQ %d\n", irq
);
545 evtchn
= bind_pirq
.port
;
547 pirq_query_unmask(irq
);
549 rc
= set_evtchn_to_irq(evtchn
, irq
);
553 info
->evtchn
= evtchn
;
554 bind_evtchn_to_cpu(evtchn
, 0);
556 rc
= xen_evtchn_port_setup(info
);
561 unmask_evtchn(evtchn
);
562 eoi_pirq(irq_get_irq_data(irq
));
567 pr_err("irq%d: Failed to set port to irq mapping (%d)\n", irq
, rc
);
568 xen_evtchn_close(evtchn
);
572 static unsigned int startup_pirq(struct irq_data
*data
)
574 return __startup_pirq(data
->irq
);
577 static void shutdown_pirq(struct irq_data
*data
)
579 unsigned int irq
= data
->irq
;
580 struct irq_info
*info
= info_for_irq(irq
);
581 unsigned evtchn
= evtchn_from_irq(irq
);
583 BUG_ON(info
->type
!= IRQT_PIRQ
);
585 if (!VALID_EVTCHN(evtchn
))
589 xen_evtchn_close(evtchn
);
590 xen_irq_info_cleanup(info
);
593 static void enable_pirq(struct irq_data
*data
)
598 static void disable_pirq(struct irq_data
*data
)
600 disable_dynirq(data
);
603 int xen_irq_from_gsi(unsigned gsi
)
605 struct irq_info
*info
;
607 list_for_each_entry(info
, &xen_irq_list_head
, list
) {
608 if (info
->type
!= IRQT_PIRQ
)
611 if (info
->u
.pirq
.gsi
== gsi
)
617 EXPORT_SYMBOL_GPL(xen_irq_from_gsi
);
619 static void __unbind_from_irq(unsigned int irq
)
621 int evtchn
= evtchn_from_irq(irq
);
622 struct irq_info
*info
= irq_get_handler_data(irq
);
624 if (info
->refcnt
> 0) {
626 if (info
->refcnt
!= 0)
630 if (VALID_EVTCHN(evtchn
)) {
631 unsigned int cpu
= cpu_from_irq(irq
);
633 xen_evtchn_close(evtchn
);
635 switch (type_from_irq(irq
)) {
637 per_cpu(virq_to_irq
, cpu
)[virq_from_irq(irq
)] = -1;
640 per_cpu(ipi_to_irq
, cpu
)[ipi_from_irq(irq
)] = -1;
646 xen_irq_info_cleanup(info
);
649 BUG_ON(info_for_irq(irq
)->type
== IRQT_UNBOUND
);
655 * Do not make any assumptions regarding the relationship between the
656 * IRQ number returned here and the Xen pirq argument.
658 * Note: We don't assign an event channel until the irq actually started
659 * up. Return an existing irq if we've already got one for the gsi.
661 * Shareable implies level triggered, not shareable implies edge
664 int xen_bind_pirq_gsi_to_irq(unsigned gsi
,
665 unsigned pirq
, int shareable
, char *name
)
668 struct physdev_irq irq_op
;
671 mutex_lock(&irq_mapping_update_lock
);
673 irq
= xen_irq_from_gsi(gsi
);
675 pr_info("%s: returning irq %d for gsi %u\n",
680 irq
= xen_allocate_irq_gsi(gsi
);
687 /* Only the privileged domain can do this. For non-priv, the pcifront
688 * driver provides a PCI bus that does the call to do exactly
689 * this in the priv domain. */
690 if (xen_initial_domain() &&
691 HYPERVISOR_physdev_op(PHYSDEVOP_alloc_irq_vector
, &irq_op
)) {
697 ret
= xen_irq_info_pirq_setup(irq
, 0, pirq
, gsi
, DOMID_SELF
,
698 shareable
? PIRQ_SHAREABLE
: 0);
700 __unbind_from_irq(irq
);
705 pirq_query_unmask(irq
);
706 /* We try to use the handler with the appropriate semantic for the
707 * type of interrupt: if the interrupt is an edge triggered
708 * interrupt we use handle_edge_irq.
710 * On the other hand if the interrupt is level triggered we use
711 * handle_fasteoi_irq like the native code does for this kind of
714 * Depending on the Xen version, pirq_needs_eoi might return true
715 * not only for level triggered interrupts but for edge triggered
716 * interrupts too. In any case Xen always honors the eoi mechanism,
717 * not injecting any more pirqs of the same kind if the first one
718 * hasn't received an eoi yet. Therefore using the fasteoi handler
719 * is the right choice either way.
722 irq_set_chip_and_handler_name(irq
, &xen_pirq_chip
,
723 handle_fasteoi_irq
, name
);
725 irq_set_chip_and_handler_name(irq
, &xen_pirq_chip
,
726 handle_edge_irq
, name
);
729 mutex_unlock(&irq_mapping_update_lock
);
734 #ifdef CONFIG_PCI_MSI
735 int xen_allocate_pirq_msi(struct pci_dev
*dev
, struct msi_desc
*msidesc
)
738 struct physdev_get_free_pirq op_get_free_pirq
;
740 op_get_free_pirq
.type
= MAP_PIRQ_TYPE_MSI
;
741 rc
= HYPERVISOR_physdev_op(PHYSDEVOP_get_free_pirq
, &op_get_free_pirq
);
743 WARN_ONCE(rc
== -ENOSYS
,
744 "hypervisor does not support the PHYSDEVOP_get_free_pirq interface\n");
746 return rc
? -1 : op_get_free_pirq
.pirq
;
749 int xen_bind_pirq_msi_to_irq(struct pci_dev
*dev
, struct msi_desc
*msidesc
,
750 int pirq
, const char *name
, domid_t domid
)
754 mutex_lock(&irq_mapping_update_lock
);
756 irq
= xen_allocate_irq_dynamic();
760 irq_set_chip_and_handler_name(irq
, &xen_pirq_chip
, handle_edge_irq
,
763 ret
= xen_irq_info_pirq_setup(irq
, 0, pirq
, 0, domid
, 0);
766 ret
= irq_set_msi_desc(irq
, msidesc
);
770 mutex_unlock(&irq_mapping_update_lock
);
773 __unbind_from_irq(irq
);
774 mutex_unlock(&irq_mapping_update_lock
);
779 int xen_destroy_irq(int irq
)
781 struct irq_desc
*desc
;
782 struct physdev_unmap_pirq unmap_irq
;
783 struct irq_info
*info
= info_for_irq(irq
);
786 mutex_lock(&irq_mapping_update_lock
);
788 desc
= irq_to_desc(irq
);
792 if (xen_initial_domain()) {
793 unmap_irq
.pirq
= info
->u
.pirq
.pirq
;
794 unmap_irq
.domid
= info
->u
.pirq
.domid
;
795 rc
= HYPERVISOR_physdev_op(PHYSDEVOP_unmap_pirq
, &unmap_irq
);
796 /* If another domain quits without making the pci_disable_msix
797 * call, the Xen hypervisor takes care of freeing the PIRQs
798 * (free_domain_pirqs).
800 if ((rc
== -ESRCH
&& info
->u
.pirq
.domid
!= DOMID_SELF
))
801 pr_info("domain %d does not have %d anymore\n",
802 info
->u
.pirq
.domid
, info
->u
.pirq
.pirq
);
804 pr_warn("unmap irq failed %d\n", rc
);
812 mutex_unlock(&irq_mapping_update_lock
);
816 int xen_irq_from_pirq(unsigned pirq
)
820 struct irq_info
*info
;
822 mutex_lock(&irq_mapping_update_lock
);
824 list_for_each_entry(info
, &xen_irq_list_head
, list
) {
825 if (info
->type
!= IRQT_PIRQ
)
828 if (info
->u
.pirq
.pirq
== pirq
)
833 mutex_unlock(&irq_mapping_update_lock
);
839 int xen_pirq_from_irq(unsigned irq
)
841 return pirq_from_irq(irq
);
843 EXPORT_SYMBOL_GPL(xen_pirq_from_irq
);
845 int bind_evtchn_to_irq(unsigned int evtchn
)
850 if (evtchn
>= xen_evtchn_max_channels())
853 mutex_lock(&irq_mapping_update_lock
);
855 irq
= get_evtchn_to_irq(evtchn
);
858 irq
= xen_allocate_irq_dynamic();
862 irq_set_chip_and_handler_name(irq
, &xen_dynamic_chip
,
863 handle_edge_irq
, "event");
865 ret
= xen_irq_info_evtchn_setup(irq
, evtchn
);
867 __unbind_from_irq(irq
);
871 /* New interdomain events are bound to VCPU 0. */
872 bind_evtchn_to_cpu(evtchn
, 0);
874 struct irq_info
*info
= info_for_irq(irq
);
875 WARN_ON(info
== NULL
|| info
->type
!= IRQT_EVTCHN
);
879 mutex_unlock(&irq_mapping_update_lock
);
883 EXPORT_SYMBOL_GPL(bind_evtchn_to_irq
);
885 static int bind_ipi_to_irq(unsigned int ipi
, unsigned int cpu
)
887 struct evtchn_bind_ipi bind_ipi
;
891 mutex_lock(&irq_mapping_update_lock
);
893 irq
= per_cpu(ipi_to_irq
, cpu
)[ipi
];
896 irq
= xen_allocate_irq_dynamic();
900 irq_set_chip_and_handler_name(irq
, &xen_percpu_chip
,
901 handle_percpu_irq
, "ipi");
904 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_ipi
,
907 evtchn
= bind_ipi
.port
;
909 ret
= xen_irq_info_ipi_setup(cpu
, irq
, evtchn
, ipi
);
911 __unbind_from_irq(irq
);
915 bind_evtchn_to_cpu(evtchn
, cpu
);
917 struct irq_info
*info
= info_for_irq(irq
);
918 WARN_ON(info
== NULL
|| info
->type
!= IRQT_IPI
);
922 mutex_unlock(&irq_mapping_update_lock
);
926 static int bind_interdomain_evtchn_to_irq(unsigned int remote_domain
,
927 unsigned int remote_port
)
929 struct evtchn_bind_interdomain bind_interdomain
;
932 bind_interdomain
.remote_dom
= remote_domain
;
933 bind_interdomain
.remote_port
= remote_port
;
935 err
= HYPERVISOR_event_channel_op(EVTCHNOP_bind_interdomain
,
938 return err
? : bind_evtchn_to_irq(bind_interdomain
.local_port
);
941 static int find_virq(unsigned int virq
, unsigned int cpu
)
943 struct evtchn_status status
;
944 int port
, rc
= -ENOENT
;
946 memset(&status
, 0, sizeof(status
));
947 for (port
= 0; port
< xen_evtchn_max_channels(); port
++) {
948 status
.dom
= DOMID_SELF
;
950 rc
= HYPERVISOR_event_channel_op(EVTCHNOP_status
, &status
);
953 if (status
.status
!= EVTCHNSTAT_virq
)
955 if (status
.u
.virq
== virq
&& status
.vcpu
== cpu
) {
964 * xen_evtchn_nr_channels - number of usable event channel ports
966 * This may be less than the maximum supported by the current
967 * hypervisor ABI. Use xen_evtchn_max_channels() for the maximum
970 unsigned xen_evtchn_nr_channels(void)
972 return evtchn_ops
->nr_channels();
974 EXPORT_SYMBOL_GPL(xen_evtchn_nr_channels
);
976 int bind_virq_to_irq(unsigned int virq
, unsigned int cpu
, bool percpu
)
978 struct evtchn_bind_virq bind_virq
;
979 int evtchn
, irq
, ret
;
981 mutex_lock(&irq_mapping_update_lock
);
983 irq
= per_cpu(virq_to_irq
, cpu
)[virq
];
986 irq
= xen_allocate_irq_dynamic();
991 irq_set_chip_and_handler_name(irq
, &xen_percpu_chip
,
992 handle_percpu_irq
, "virq");
994 irq_set_chip_and_handler_name(irq
, &xen_dynamic_chip
,
995 handle_edge_irq
, "virq");
997 bind_virq
.virq
= virq
;
998 bind_virq
.vcpu
= cpu
;
999 ret
= HYPERVISOR_event_channel_op(EVTCHNOP_bind_virq
,
1002 evtchn
= bind_virq
.port
;
1005 ret
= find_virq(virq
, cpu
);
1010 ret
= xen_irq_info_virq_setup(cpu
, irq
, evtchn
, virq
);
1012 __unbind_from_irq(irq
);
1017 bind_evtchn_to_cpu(evtchn
, cpu
);
1019 struct irq_info
*info
= info_for_irq(irq
);
1020 WARN_ON(info
== NULL
|| info
->type
!= IRQT_VIRQ
);
1024 mutex_unlock(&irq_mapping_update_lock
);
1029 static void unbind_from_irq(unsigned int irq
)
1031 mutex_lock(&irq_mapping_update_lock
);
1032 __unbind_from_irq(irq
);
1033 mutex_unlock(&irq_mapping_update_lock
);
1036 int bind_evtchn_to_irqhandler(unsigned int evtchn
,
1037 irq_handler_t handler
,
1038 unsigned long irqflags
,
1039 const char *devname
, void *dev_id
)
1043 irq
= bind_evtchn_to_irq(evtchn
);
1046 retval
= request_irq(irq
, handler
, irqflags
, devname
, dev_id
);
1048 unbind_from_irq(irq
);
1054 EXPORT_SYMBOL_GPL(bind_evtchn_to_irqhandler
);
1056 int bind_interdomain_evtchn_to_irqhandler(unsigned int remote_domain
,
1057 unsigned int remote_port
,
1058 irq_handler_t handler
,
1059 unsigned long irqflags
,
1060 const char *devname
,
1065 irq
= bind_interdomain_evtchn_to_irq(remote_domain
, remote_port
);
1069 retval
= request_irq(irq
, handler
, irqflags
, devname
, dev_id
);
1071 unbind_from_irq(irq
);
1077 EXPORT_SYMBOL_GPL(bind_interdomain_evtchn_to_irqhandler
);
1079 int bind_virq_to_irqhandler(unsigned int virq
, unsigned int cpu
,
1080 irq_handler_t handler
,
1081 unsigned long irqflags
, const char *devname
, void *dev_id
)
1085 irq
= bind_virq_to_irq(virq
, cpu
, irqflags
& IRQF_PERCPU
);
1088 retval
= request_irq(irq
, handler
, irqflags
, devname
, dev_id
);
1090 unbind_from_irq(irq
);
1096 EXPORT_SYMBOL_GPL(bind_virq_to_irqhandler
);
1098 int bind_ipi_to_irqhandler(enum ipi_vector ipi
,
1100 irq_handler_t handler
,
1101 unsigned long irqflags
,
1102 const char *devname
,
1107 irq
= bind_ipi_to_irq(ipi
, cpu
);
1111 irqflags
|= IRQF_NO_SUSPEND
| IRQF_FORCE_RESUME
| IRQF_EARLY_RESUME
;
1112 retval
= request_irq(irq
, handler
, irqflags
, devname
, dev_id
);
1114 unbind_from_irq(irq
);
1121 void unbind_from_irqhandler(unsigned int irq
, void *dev_id
)
1123 struct irq_info
*info
= irq_get_handler_data(irq
);
1127 free_irq(irq
, dev_id
);
1128 unbind_from_irq(irq
);
1130 EXPORT_SYMBOL_GPL(unbind_from_irqhandler
);
1133 * xen_set_irq_priority() - set an event channel priority.
1134 * @irq:irq bound to an event channel.
1135 * @priority: priority between XEN_IRQ_PRIORITY_MAX and XEN_IRQ_PRIORITY_MIN.
1137 int xen_set_irq_priority(unsigned irq
, unsigned priority
)
1139 struct evtchn_set_priority set_priority
;
1141 set_priority
.port
= evtchn_from_irq(irq
);
1142 set_priority
.priority
= priority
;
1144 return HYPERVISOR_event_channel_op(EVTCHNOP_set_priority
,
1147 EXPORT_SYMBOL_GPL(xen_set_irq_priority
);
1149 int evtchn_make_refcounted(unsigned int evtchn
)
1151 int irq
= get_evtchn_to_irq(evtchn
);
1152 struct irq_info
*info
;
1157 info
= irq_get_handler_data(irq
);
1162 WARN_ON(info
->refcnt
!= -1);
1168 EXPORT_SYMBOL_GPL(evtchn_make_refcounted
);
1170 int evtchn_get(unsigned int evtchn
)
1173 struct irq_info
*info
;
1176 if (evtchn
>= xen_evtchn_max_channels())
1179 mutex_lock(&irq_mapping_update_lock
);
1181 irq
= get_evtchn_to_irq(evtchn
);
1185 info
= irq_get_handler_data(irq
);
1191 if (info
->refcnt
<= 0)
1197 mutex_unlock(&irq_mapping_update_lock
);
1201 EXPORT_SYMBOL_GPL(evtchn_get
);
1203 void evtchn_put(unsigned int evtchn
)
1205 int irq
= get_evtchn_to_irq(evtchn
);
1206 if (WARN_ON(irq
== -1))
1208 unbind_from_irq(irq
);
1210 EXPORT_SYMBOL_GPL(evtchn_put
);
1212 void xen_send_IPI_one(unsigned int cpu
, enum ipi_vector vector
)
1217 if (unlikely(vector
== XEN_NMI_VECTOR
)) {
1218 int rc
= HYPERVISOR_vcpu_op(VCPUOP_send_nmi
, cpu
, NULL
);
1220 printk(KERN_WARNING
"Sending nmi to CPU%d failed (rc:%d)\n", cpu
, rc
);
1224 irq
= per_cpu(ipi_to_irq
, cpu
)[vector
];
1226 notify_remote_via_irq(irq
);
1229 static DEFINE_PER_CPU(unsigned, xed_nesting_count
);
1231 static void __xen_evtchn_do_upcall(void)
1233 struct vcpu_info
*vcpu_info
= __this_cpu_read(xen_vcpu
);
1234 int cpu
= get_cpu();
1238 vcpu_info
->evtchn_upcall_pending
= 0;
1240 if (__this_cpu_inc_return(xed_nesting_count
) - 1)
1243 xen_evtchn_handle_events(cpu
);
1245 BUG_ON(!irqs_disabled());
1247 count
= __this_cpu_read(xed_nesting_count
);
1248 __this_cpu_write(xed_nesting_count
, 0);
1249 } while (count
!= 1 || vcpu_info
->evtchn_upcall_pending
);
1256 void xen_evtchn_do_upcall(struct pt_regs
*regs
)
1258 struct pt_regs
*old_regs
= set_irq_regs(regs
);
1265 __xen_evtchn_do_upcall();
1268 set_irq_regs(old_regs
);
1271 void xen_hvm_evtchn_do_upcall(void)
1273 __xen_evtchn_do_upcall();
1275 EXPORT_SYMBOL_GPL(xen_hvm_evtchn_do_upcall
);
1277 /* Rebind a new event channel to an existing irq. */
1278 void rebind_evtchn_irq(int evtchn
, int irq
)
1280 struct irq_info
*info
= info_for_irq(irq
);
1285 /* Make sure the irq is masked, since the new event channel
1286 will also be masked. */
1289 mutex_lock(&irq_mapping_update_lock
);
1291 /* After resume the irq<->evtchn mappings are all cleared out */
1292 BUG_ON(get_evtchn_to_irq(evtchn
) != -1);
1293 /* Expect irq to have been bound before,
1294 so there should be a proper type */
1295 BUG_ON(info
->type
== IRQT_UNBOUND
);
1297 (void)xen_irq_info_evtchn_setup(irq
, evtchn
);
1299 mutex_unlock(&irq_mapping_update_lock
);
1301 bind_evtchn_to_cpu(evtchn
, info
->cpu
);
1302 /* This will be deferred until interrupt is processed */
1303 irq_set_affinity(irq
, cpumask_of(info
->cpu
));
1305 /* Unmask the event channel. */
1309 /* Rebind an evtchn so that it gets delivered to a specific cpu */
1310 static int rebind_irq_to_cpu(unsigned irq
, unsigned tcpu
)
1312 struct evtchn_bind_vcpu bind_vcpu
;
1313 int evtchn
= evtchn_from_irq(irq
);
1316 if (!VALID_EVTCHN(evtchn
))
1320 * Events delivered via platform PCI interrupts are always
1321 * routed to vcpu 0 and hence cannot be rebound.
1323 if (xen_hvm_domain() && !xen_have_vector_callback
)
1326 /* Send future instances of this interrupt to other vcpu. */
1327 bind_vcpu
.port
= evtchn
;
1328 bind_vcpu
.vcpu
= tcpu
;
1331 * Mask the event while changing the VCPU binding to prevent
1332 * it being delivered on an unexpected VCPU.
1334 masked
= test_and_set_mask(evtchn
);
1337 * If this fails, it usually just indicates that we're dealing with a
1338 * virq or IPI channel, which don't actually need to be rebound. Ignore
1339 * it, but don't do the xenlinux-level rebind in that case.
1341 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_vcpu
, &bind_vcpu
) >= 0)
1342 bind_evtchn_to_cpu(evtchn
, tcpu
);
1345 unmask_evtchn(evtchn
);
1350 static int set_affinity_irq(struct irq_data
*data
, const struct cpumask
*dest
,
1353 unsigned tcpu
= cpumask_first(dest
);
1355 return rebind_irq_to_cpu(data
->irq
, tcpu
);
1358 static int retrigger_evtchn(int evtchn
)
1362 if (!VALID_EVTCHN(evtchn
))
1365 masked
= test_and_set_mask(evtchn
);
1368 unmask_evtchn(evtchn
);
1373 int resend_irq_on_evtchn(unsigned int irq
)
1375 return retrigger_evtchn(evtchn_from_irq(irq
));
1378 static void enable_dynirq(struct irq_data
*data
)
1380 int evtchn
= evtchn_from_irq(data
->irq
);
1382 if (VALID_EVTCHN(evtchn
))
1383 unmask_evtchn(evtchn
);
1386 static void disable_dynirq(struct irq_data
*data
)
1388 int evtchn
= evtchn_from_irq(data
->irq
);
1390 if (VALID_EVTCHN(evtchn
))
1391 mask_evtchn(evtchn
);
1394 static void ack_dynirq(struct irq_data
*data
)
1396 int evtchn
= evtchn_from_irq(data
->irq
);
1400 if (VALID_EVTCHN(evtchn
))
1401 clear_evtchn(evtchn
);
1404 static void mask_ack_dynirq(struct irq_data
*data
)
1406 disable_dynirq(data
);
1410 static int retrigger_dynirq(struct irq_data
*data
)
1412 return retrigger_evtchn(evtchn_from_irq(data
->irq
));
1415 static void restore_pirqs(void)
1417 int pirq
, rc
, irq
, gsi
;
1418 struct physdev_map_pirq map_irq
;
1419 struct irq_info
*info
;
1421 list_for_each_entry(info
, &xen_irq_list_head
, list
) {
1422 if (info
->type
!= IRQT_PIRQ
)
1425 pirq
= info
->u
.pirq
.pirq
;
1426 gsi
= info
->u
.pirq
.gsi
;
1429 /* save/restore of PT devices doesn't work, so at this point the
1430 * only devices present are GSI based emulated devices */
1434 map_irq
.domid
= DOMID_SELF
;
1435 map_irq
.type
= MAP_PIRQ_TYPE_GSI
;
1436 map_irq
.index
= gsi
;
1437 map_irq
.pirq
= pirq
;
1439 rc
= HYPERVISOR_physdev_op(PHYSDEVOP_map_pirq
, &map_irq
);
1441 pr_warn("xen map irq failed gsi=%d irq=%d pirq=%d rc=%d\n",
1442 gsi
, irq
, pirq
, rc
);
1447 printk(KERN_DEBUG
"xen: --> irq=%d, pirq=%d\n", irq
, map_irq
.pirq
);
1449 __startup_pirq(irq
);
1453 static void restore_cpu_virqs(unsigned int cpu
)
1455 struct evtchn_bind_virq bind_virq
;
1456 int virq
, irq
, evtchn
;
1458 for (virq
= 0; virq
< NR_VIRQS
; virq
++) {
1459 if ((irq
= per_cpu(virq_to_irq
, cpu
)[virq
]) == -1)
1462 BUG_ON(virq_from_irq(irq
) != virq
);
1464 /* Get a new binding from Xen. */
1465 bind_virq
.virq
= virq
;
1466 bind_virq
.vcpu
= cpu
;
1467 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_virq
,
1470 evtchn
= bind_virq
.port
;
1472 /* Record the new mapping. */
1473 (void)xen_irq_info_virq_setup(cpu
, irq
, evtchn
, virq
);
1474 bind_evtchn_to_cpu(evtchn
, cpu
);
1478 static void restore_cpu_ipis(unsigned int cpu
)
1480 struct evtchn_bind_ipi bind_ipi
;
1481 int ipi
, irq
, evtchn
;
1483 for (ipi
= 0; ipi
< XEN_NR_IPIS
; ipi
++) {
1484 if ((irq
= per_cpu(ipi_to_irq
, cpu
)[ipi
]) == -1)
1487 BUG_ON(ipi_from_irq(irq
) != ipi
);
1489 /* Get a new binding from Xen. */
1490 bind_ipi
.vcpu
= cpu
;
1491 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_ipi
,
1494 evtchn
= bind_ipi
.port
;
1496 /* Record the new mapping. */
1497 (void)xen_irq_info_ipi_setup(cpu
, irq
, evtchn
, ipi
);
1498 bind_evtchn_to_cpu(evtchn
, cpu
);
1502 /* Clear an irq's pending state, in preparation for polling on it */
1503 void xen_clear_irq_pending(int irq
)
1505 int evtchn
= evtchn_from_irq(irq
);
1507 if (VALID_EVTCHN(evtchn
))
1508 clear_evtchn(evtchn
);
1510 EXPORT_SYMBOL(xen_clear_irq_pending
);
1511 void xen_set_irq_pending(int irq
)
1513 int evtchn
= evtchn_from_irq(irq
);
1515 if (VALID_EVTCHN(evtchn
))
1519 bool xen_test_irq_pending(int irq
)
1521 int evtchn
= evtchn_from_irq(irq
);
1524 if (VALID_EVTCHN(evtchn
))
1525 ret
= test_evtchn(evtchn
);
1530 /* Poll waiting for an irq to become pending with timeout. In the usual case,
1531 * the irq will be disabled so it won't deliver an interrupt. */
1532 void xen_poll_irq_timeout(int irq
, u64 timeout
)
1534 evtchn_port_t evtchn
= evtchn_from_irq(irq
);
1536 if (VALID_EVTCHN(evtchn
)) {
1537 struct sched_poll poll
;
1540 poll
.timeout
= timeout
;
1541 set_xen_guest_handle(poll
.ports
, &evtchn
);
1543 if (HYPERVISOR_sched_op(SCHEDOP_poll
, &poll
) != 0)
1547 EXPORT_SYMBOL(xen_poll_irq_timeout
);
1548 /* Poll waiting for an irq to become pending. In the usual case, the
1549 * irq will be disabled so it won't deliver an interrupt. */
1550 void xen_poll_irq(int irq
)
1552 xen_poll_irq_timeout(irq
, 0 /* no timeout */);
1555 /* Check whether the IRQ line is shared with other guests. */
1556 int xen_test_irq_shared(int irq
)
1558 struct irq_info
*info
= info_for_irq(irq
);
1559 struct physdev_irq_status_query irq_status
;
1564 irq_status
.irq
= info
->u
.pirq
.pirq
;
1566 if (HYPERVISOR_physdev_op(PHYSDEVOP_irq_status_query
, &irq_status
))
1568 return !(irq_status
.flags
& XENIRQSTAT_shared
);
1570 EXPORT_SYMBOL_GPL(xen_test_irq_shared
);
1572 void xen_irq_resume(void)
1575 struct irq_info
*info
;
1577 /* New event-channel space is not 'live' yet. */
1578 xen_evtchn_mask_all();
1579 xen_evtchn_resume();
1581 /* No IRQ <-> event-channel mappings. */
1582 list_for_each_entry(info
, &xen_irq_list_head
, list
)
1583 info
->evtchn
= 0; /* zap event-channel binding */
1585 clear_evtchn_to_irq_all();
1587 for_each_possible_cpu(cpu
) {
1588 restore_cpu_virqs(cpu
);
1589 restore_cpu_ipis(cpu
);
1595 static struct irq_chip xen_dynamic_chip __read_mostly
= {
1598 .irq_disable
= disable_dynirq
,
1599 .irq_mask
= disable_dynirq
,
1600 .irq_unmask
= enable_dynirq
,
1602 .irq_ack
= ack_dynirq
,
1603 .irq_mask_ack
= mask_ack_dynirq
,
1605 .irq_set_affinity
= set_affinity_irq
,
1606 .irq_retrigger
= retrigger_dynirq
,
1609 static struct irq_chip xen_pirq_chip __read_mostly
= {
1612 .irq_startup
= startup_pirq
,
1613 .irq_shutdown
= shutdown_pirq
,
1614 .irq_enable
= enable_pirq
,
1615 .irq_disable
= disable_pirq
,
1617 .irq_mask
= disable_dynirq
,
1618 .irq_unmask
= enable_dynirq
,
1620 .irq_ack
= eoi_pirq
,
1621 .irq_eoi
= eoi_pirq
,
1622 .irq_mask_ack
= mask_ack_pirq
,
1624 .irq_set_affinity
= set_affinity_irq
,
1626 .irq_retrigger
= retrigger_dynirq
,
1629 static struct irq_chip xen_percpu_chip __read_mostly
= {
1630 .name
= "xen-percpu",
1632 .irq_disable
= disable_dynirq
,
1633 .irq_mask
= disable_dynirq
,
1634 .irq_unmask
= enable_dynirq
,
1636 .irq_ack
= ack_dynirq
,
1639 int xen_set_callback_via(uint64_t via
)
1641 struct xen_hvm_param a
;
1642 a
.domid
= DOMID_SELF
;
1643 a
.index
= HVM_PARAM_CALLBACK_IRQ
;
1645 return HYPERVISOR_hvm_op(HVMOP_set_param
, &a
);
1647 EXPORT_SYMBOL_GPL(xen_set_callback_via
);
1649 #ifdef CONFIG_XEN_PVHVM
1650 /* Vector callbacks are better than PCI interrupts to receive event
1651 * channel notifications because we can receive vector callbacks on any
1652 * vcpu and we don't need PCI support or APIC interactions. */
1653 void xen_callback_vector(void)
1656 uint64_t callback_via
;
1657 if (xen_have_vector_callback
) {
1658 callback_via
= HVM_CALLBACK_VECTOR(HYPERVISOR_CALLBACK_VECTOR
);
1659 rc
= xen_set_callback_via(callback_via
);
1661 pr_err("Request for Xen HVM callback vector failed\n");
1662 xen_have_vector_callback
= 0;
1665 pr_info("Xen HVM callback vector for event delivery is enabled\n");
1666 /* in the restore case the vector has already been allocated */
1667 if (!test_bit(HYPERVISOR_CALLBACK_VECTOR
, used_vectors
))
1668 alloc_intr_gate(HYPERVISOR_CALLBACK_VECTOR
,
1669 xen_hvm_callback_vector
);
1673 void xen_callback_vector(void) {}
1676 #undef MODULE_PARAM_PREFIX
1677 #define MODULE_PARAM_PREFIX "xen."
1679 static bool fifo_events
= true;
1680 module_param(fifo_events
, bool, 0);
1682 void __init
xen_init_IRQ(void)
1687 ret
= xen_evtchn_fifo_init();
1689 xen_evtchn_2l_init();
1691 evtchn_to_irq
= kcalloc(EVTCHN_ROW(xen_evtchn_max_channels()),
1692 sizeof(*evtchn_to_irq
), GFP_KERNEL
);
1693 BUG_ON(!evtchn_to_irq
);
1695 /* No event channels are 'live' right now. */
1696 xen_evtchn_mask_all();
1698 pirq_needs_eoi
= pirq_needs_eoi_flag
;
1701 if (xen_pv_domain()) {
1702 irq_ctx_init(smp_processor_id());
1703 if (xen_initial_domain())
1704 pci_xen_initial_domain();
1706 if (xen_feature(XENFEAT_hvm_callback_vector
))
1707 xen_callback_vector();
1709 if (xen_hvm_domain()) {
1711 /* pci_xen_hvm_init must be called after native_init_IRQ so that
1712 * __acpi_register_gsi can point at the right function */
1716 struct physdev_pirq_eoi_gmfn eoi_gmfn
;
1718 pirq_eoi_map
= (void *)__get_free_page(GFP_KERNEL
|__GFP_ZERO
);
1719 eoi_gmfn
.gmfn
= virt_to_mfn(pirq_eoi_map
);
1720 rc
= HYPERVISOR_physdev_op(PHYSDEVOP_pirq_eoi_gmfn_v2
, &eoi_gmfn
);
1721 /* TODO: No PVH support for PIRQ EOI */
1723 free_page((unsigned long) pirq_eoi_map
);
1724 pirq_eoi_map
= NULL
;
1726 pirq_needs_eoi
= pirq_check_eoi_map
;