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 recieved, 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 #include <linux/linkage.h>
25 #include <linux/interrupt.h>
26 #include <linux/irq.h>
27 #include <linux/module.h>
28 #include <linux/string.h>
29 #include <linux/bootmem.h>
30 #include <linux/slab.h>
31 #include <linux/irqnr.h>
32 #include <linux/pci.h>
35 #include <asm/ptrace.h>
38 #include <asm/io_apic.h>
39 #include <asm/sync_bitops.h>
40 #include <asm/xen/pci.h>
41 #include <asm/xen/hypercall.h>
42 #include <asm/xen/hypervisor.h>
46 #include <xen/xen-ops.h>
47 #include <xen/events.h>
48 #include <xen/interface/xen.h>
49 #include <xen/interface/event_channel.h>
50 #include <xen/interface/hvm/hvm_op.h>
51 #include <xen/interface/hvm/params.h>
54 * This lock protects updates to the following mapping and reference-count
55 * arrays. The lock does not need to be acquired to read the mapping tables.
57 static DEFINE_SPINLOCK(irq_mapping_update_lock
);
59 /* IRQ <-> VIRQ mapping. */
60 static DEFINE_PER_CPU(int [NR_VIRQS
], virq_to_irq
) = {[0 ... NR_VIRQS
-1] = -1};
62 /* IRQ <-> IPI mapping */
63 static DEFINE_PER_CPU(int [XEN_NR_IPIS
], ipi_to_irq
) = {[0 ... XEN_NR_IPIS
-1] = -1};
65 /* Interrupt types. */
75 * Packed IRQ information:
76 * type - enum xen_irq_type
77 * event channel - irq->event channel mapping
78 * cpu - cpu this event channel is bound to
79 * index - type-specific information:
80 * PIRQ - vector, with MSB being "needs EIO", or physical IRQ of the HVM
81 * guest, or GSI (real passthrough IRQ) of the device.
88 enum xen_irq_type type
; /* type */
89 unsigned short evtchn
; /* event channel */
90 unsigned short cpu
; /* cpu bound */
103 #define PIRQ_NEEDS_EOI (1 << 0)
104 #define PIRQ_SHAREABLE (1 << 1)
106 static struct irq_info
*irq_info
;
107 static int *pirq_to_irq
;
109 static int *evtchn_to_irq
;
110 struct cpu_evtchn_s
{
111 unsigned long bits
[NR_EVENT_CHANNELS
/BITS_PER_LONG
];
114 static __initdata
struct cpu_evtchn_s init_evtchn_mask
= {
115 .bits
[0 ... (NR_EVENT_CHANNELS
/BITS_PER_LONG
)-1] = ~0ul,
117 static struct cpu_evtchn_s
*cpu_evtchn_mask_p
= &init_evtchn_mask
;
119 static inline unsigned long *cpu_evtchn_mask(int cpu
)
121 return cpu_evtchn_mask_p
[cpu
].bits
;
124 /* Xen will never allocate port zero for any purpose. */
125 #define VALID_EVTCHN(chn) ((chn) != 0)
127 static struct irq_chip xen_dynamic_chip
;
128 static struct irq_chip xen_percpu_chip
;
129 static struct irq_chip xen_pirq_chip
;
131 /* Constructor for packed IRQ information. */
132 static struct irq_info
mk_unbound_info(void)
134 return (struct irq_info
) { .type
= IRQT_UNBOUND
};
137 static struct irq_info
mk_evtchn_info(unsigned short evtchn
)
139 return (struct irq_info
) { .type
= IRQT_EVTCHN
, .evtchn
= evtchn
,
143 static struct irq_info
mk_ipi_info(unsigned short evtchn
, enum ipi_vector ipi
)
145 return (struct irq_info
) { .type
= IRQT_IPI
, .evtchn
= evtchn
,
146 .cpu
= 0, .u
.ipi
= ipi
};
149 static struct irq_info
mk_virq_info(unsigned short evtchn
, unsigned short virq
)
151 return (struct irq_info
) { .type
= IRQT_VIRQ
, .evtchn
= evtchn
,
152 .cpu
= 0, .u
.virq
= virq
};
155 static struct irq_info
mk_pirq_info(unsigned short evtchn
, unsigned short pirq
,
156 unsigned short gsi
, unsigned short vector
)
158 return (struct irq_info
) { .type
= IRQT_PIRQ
, .evtchn
= evtchn
,
160 .u
.pirq
= { .pirq
= pirq
, .gsi
= gsi
, .vector
= vector
} };
164 * Accessors for packed IRQ information.
166 static struct irq_info
*info_for_irq(unsigned irq
)
168 return &irq_info
[irq
];
171 static unsigned int evtchn_from_irq(unsigned irq
)
173 if (unlikely(WARN(irq
< 0 || irq
>= nr_irqs
, "Invalid irq %d!\n", irq
)))
176 return info_for_irq(irq
)->evtchn
;
179 unsigned irq_from_evtchn(unsigned int evtchn
)
181 return evtchn_to_irq
[evtchn
];
183 EXPORT_SYMBOL_GPL(irq_from_evtchn
);
185 static enum ipi_vector
ipi_from_irq(unsigned irq
)
187 struct irq_info
*info
= info_for_irq(irq
);
189 BUG_ON(info
== NULL
);
190 BUG_ON(info
->type
!= IRQT_IPI
);
195 static unsigned virq_from_irq(unsigned irq
)
197 struct irq_info
*info
= info_for_irq(irq
);
199 BUG_ON(info
== NULL
);
200 BUG_ON(info
->type
!= IRQT_VIRQ
);
205 static unsigned pirq_from_irq(unsigned irq
)
207 struct irq_info
*info
= info_for_irq(irq
);
209 BUG_ON(info
== NULL
);
210 BUG_ON(info
->type
!= IRQT_PIRQ
);
212 return info
->u
.pirq
.pirq
;
215 static unsigned gsi_from_irq(unsigned irq
)
217 struct irq_info
*info
= info_for_irq(irq
);
219 BUG_ON(info
== NULL
);
220 BUG_ON(info
->type
!= IRQT_PIRQ
);
222 return info
->u
.pirq
.gsi
;
225 static unsigned vector_from_irq(unsigned irq
)
227 struct irq_info
*info
= info_for_irq(irq
);
229 BUG_ON(info
== NULL
);
230 BUG_ON(info
->type
!= IRQT_PIRQ
);
232 return info
->u
.pirq
.vector
;
235 static enum xen_irq_type
type_from_irq(unsigned irq
)
237 return info_for_irq(irq
)->type
;
240 static unsigned cpu_from_irq(unsigned irq
)
242 return info_for_irq(irq
)->cpu
;
245 static unsigned int cpu_from_evtchn(unsigned int evtchn
)
247 int irq
= evtchn_to_irq
[evtchn
];
251 ret
= cpu_from_irq(irq
);
256 static bool pirq_needs_eoi(unsigned irq
)
258 struct irq_info
*info
= info_for_irq(irq
);
260 BUG_ON(info
->type
!= IRQT_PIRQ
);
262 return info
->u
.pirq
.flags
& PIRQ_NEEDS_EOI
;
265 static inline unsigned long active_evtchns(unsigned int cpu
,
266 struct shared_info
*sh
,
269 return (sh
->evtchn_pending
[idx
] &
270 cpu_evtchn_mask(cpu
)[idx
] &
271 ~sh
->evtchn_mask
[idx
]);
274 static void bind_evtchn_to_cpu(unsigned int chn
, unsigned int cpu
)
276 int irq
= evtchn_to_irq
[chn
];
280 cpumask_copy(irq_to_desc(irq
)->irq_data
.affinity
, cpumask_of(cpu
));
283 clear_bit(chn
, cpu_evtchn_mask(cpu_from_irq(irq
)));
284 set_bit(chn
, cpu_evtchn_mask(cpu
));
286 irq_info
[irq
].cpu
= cpu
;
289 static void init_evtchn_cpu_bindings(void)
293 struct irq_desc
*desc
;
295 /* By default all event channels notify CPU#0. */
296 for_each_irq_desc(i
, desc
) {
297 cpumask_copy(desc
->irq_data
.affinity
, cpumask_of(0));
301 for_each_possible_cpu(i
)
302 memset(cpu_evtchn_mask(i
),
303 (i
== 0) ? ~0 : 0, sizeof(struct cpu_evtchn_s
));
307 static inline void clear_evtchn(int port
)
309 struct shared_info
*s
= HYPERVISOR_shared_info
;
310 sync_clear_bit(port
, &s
->evtchn_pending
[0]);
313 static inline void set_evtchn(int port
)
315 struct shared_info
*s
= HYPERVISOR_shared_info
;
316 sync_set_bit(port
, &s
->evtchn_pending
[0]);
319 static inline int test_evtchn(int port
)
321 struct shared_info
*s
= HYPERVISOR_shared_info
;
322 return sync_test_bit(port
, &s
->evtchn_pending
[0]);
327 * notify_remote_via_irq - send event to remote end of event channel via irq
328 * @irq: irq of event channel to send event to
330 * Unlike notify_remote_via_evtchn(), this is safe to use across
331 * save/restore. Notifications on a broken connection are silently
334 void notify_remote_via_irq(int irq
)
336 int evtchn
= evtchn_from_irq(irq
);
338 if (VALID_EVTCHN(evtchn
))
339 notify_remote_via_evtchn(evtchn
);
341 EXPORT_SYMBOL_GPL(notify_remote_via_irq
);
343 static void mask_evtchn(int port
)
345 struct shared_info
*s
= HYPERVISOR_shared_info
;
346 sync_set_bit(port
, &s
->evtchn_mask
[0]);
349 static void unmask_evtchn(int port
)
351 struct shared_info
*s
= HYPERVISOR_shared_info
;
352 unsigned int cpu
= get_cpu();
354 BUG_ON(!irqs_disabled());
356 /* Slow path (hypercall) if this is a non-local port. */
357 if (unlikely(cpu
!= cpu_from_evtchn(port
))) {
358 struct evtchn_unmask unmask
= { .port
= port
};
359 (void)HYPERVISOR_event_channel_op(EVTCHNOP_unmask
, &unmask
);
361 struct vcpu_info
*vcpu_info
= __this_cpu_read(xen_vcpu
);
363 sync_clear_bit(port
, &s
->evtchn_mask
[0]);
366 * The following is basically the equivalent of
367 * 'hw_resend_irq'. Just like a real IO-APIC we 'lose
368 * the interrupt edge' if the channel is masked.
370 if (sync_test_bit(port
, &s
->evtchn_pending
[0]) &&
371 !sync_test_and_set_bit(port
/ BITS_PER_LONG
,
372 &vcpu_info
->evtchn_pending_sel
))
373 vcpu_info
->evtchn_upcall_pending
= 1;
379 static int xen_allocate_irq_dynamic(void)
384 #ifdef CONFIG_X86_IO_APIC
386 * For an HVM guest or domain 0 which see "real" (emulated or
387 * actual repectively) GSIs we allocate dynamic IRQs
388 * e.g. those corresponding to event channels or MSIs
389 * etc. from the range above those "real" GSIs to avoid
392 if (xen_initial_domain() || xen_hvm_domain())
393 first
= get_nr_irqs_gsi();
397 irq
= irq_alloc_desc_from(first
, -1);
399 if (irq
== -ENOMEM
&& first
> NR_IRQS_LEGACY
) {
400 printk(KERN_ERR
"Out of dynamic IRQ space and eating into GSI space. You should increase nr_irqs\n");
401 first
= max(NR_IRQS_LEGACY
, first
- NR_IRQS_LEGACY
);
406 panic("No available IRQ to bind to: increase nr_irqs!\n");
411 static int xen_allocate_irq_gsi(unsigned gsi
)
416 * A PV guest has no concept of a GSI (since it has no ACPI
417 * nor access to/knowledge of the physical APICs). Therefore
418 * all IRQs are dynamically allocated from the entire IRQ
421 if (xen_pv_domain() && !xen_initial_domain())
422 return xen_allocate_irq_dynamic();
424 /* Legacy IRQ descriptors are already allocated by the arch. */
425 if (gsi
< NR_IRQS_LEGACY
)
428 irq
= irq_alloc_desc_at(gsi
, -1);
430 panic("Unable to allocate to IRQ%d (%d)\n", gsi
, irq
);
435 static void xen_free_irq(unsigned irq
)
437 /* Legacy IRQ descriptors are managed by the arch. */
438 if (irq
< NR_IRQS_LEGACY
)
444 static void pirq_unmask_notify(int irq
)
446 struct physdev_eoi eoi
= { .irq
= pirq_from_irq(irq
) };
448 if (unlikely(pirq_needs_eoi(irq
))) {
449 int rc
= HYPERVISOR_physdev_op(PHYSDEVOP_eoi
, &eoi
);
454 static void pirq_query_unmask(int irq
)
456 struct physdev_irq_status_query irq_status
;
457 struct irq_info
*info
= info_for_irq(irq
);
459 BUG_ON(info
->type
!= IRQT_PIRQ
);
461 irq_status
.irq
= pirq_from_irq(irq
);
462 if (HYPERVISOR_physdev_op(PHYSDEVOP_irq_status_query
, &irq_status
))
463 irq_status
.flags
= 0;
465 info
->u
.pirq
.flags
&= ~PIRQ_NEEDS_EOI
;
466 if (irq_status
.flags
& XENIRQSTAT_needs_eoi
)
467 info
->u
.pirq
.flags
|= PIRQ_NEEDS_EOI
;
470 static bool probing_irq(int irq
)
472 struct irq_desc
*desc
= irq_to_desc(irq
);
474 return desc
&& desc
->action
== NULL
;
477 static unsigned int __startup_pirq(unsigned int irq
)
479 struct evtchn_bind_pirq bind_pirq
;
480 struct irq_info
*info
= info_for_irq(irq
);
481 int evtchn
= evtchn_from_irq(irq
);
484 BUG_ON(info
->type
!= IRQT_PIRQ
);
486 if (VALID_EVTCHN(evtchn
))
489 bind_pirq
.pirq
= pirq_from_irq(irq
);
490 /* NB. We are happy to share unless we are probing. */
491 bind_pirq
.flags
= info
->u
.pirq
.flags
& PIRQ_SHAREABLE
?
492 BIND_PIRQ__WILL_SHARE
: 0;
493 rc
= HYPERVISOR_event_channel_op(EVTCHNOP_bind_pirq
, &bind_pirq
);
495 if (!probing_irq(irq
))
496 printk(KERN_INFO
"Failed to obtain physical IRQ %d\n",
500 evtchn
= bind_pirq
.port
;
502 pirq_query_unmask(irq
);
504 evtchn_to_irq
[evtchn
] = irq
;
505 bind_evtchn_to_cpu(evtchn
, 0);
506 info
->evtchn
= evtchn
;
509 unmask_evtchn(evtchn
);
510 pirq_unmask_notify(irq
);
515 static unsigned int startup_pirq(struct irq_data
*data
)
517 return __startup_pirq(data
->irq
);
520 static void shutdown_pirq(struct irq_data
*data
)
522 struct evtchn_close close
;
523 unsigned int irq
= data
->irq
;
524 struct irq_info
*info
= info_for_irq(irq
);
525 int evtchn
= evtchn_from_irq(irq
);
527 BUG_ON(info
->type
!= IRQT_PIRQ
);
529 if (!VALID_EVTCHN(evtchn
))
535 if (HYPERVISOR_event_channel_op(EVTCHNOP_close
, &close
) != 0)
538 bind_evtchn_to_cpu(evtchn
, 0);
539 evtchn_to_irq
[evtchn
] = -1;
543 static void enable_pirq(struct irq_data
*data
)
548 static void disable_pirq(struct irq_data
*data
)
552 static void ack_pirq(struct irq_data
*data
)
554 int evtchn
= evtchn_from_irq(data
->irq
);
556 move_native_irq(data
->irq
);
558 if (VALID_EVTCHN(evtchn
)) {
560 clear_evtchn(evtchn
);
564 static int find_irq_by_gsi(unsigned gsi
)
568 for (irq
= 0; irq
< nr_irqs
; irq
++) {
569 struct irq_info
*info
= info_for_irq(irq
);
571 if (info
== NULL
|| info
->type
!= IRQT_PIRQ
)
574 if (gsi_from_irq(irq
) == gsi
)
581 int xen_allocate_pirq(unsigned gsi
, int shareable
, char *name
)
583 return xen_map_pirq_gsi(gsi
, gsi
, shareable
, name
);
586 /* xen_map_pirq_gsi might allocate irqs from the top down, as a
587 * consequence don't assume that the irq number returned has a low value
588 * or can be used as a pirq number unless you know otherwise.
590 * One notable exception is when xen_map_pirq_gsi is called passing an
591 * hardware gsi as argument, in that case the irq number returned
592 * matches the gsi number passed as second argument.
594 * Note: We don't assign an event channel until the irq actually started
595 * up. Return an existing irq if we've already got one for the gsi.
597 int xen_map_pirq_gsi(unsigned pirq
, unsigned gsi
, int shareable
, char *name
)
600 struct physdev_irq irq_op
;
602 spin_lock(&irq_mapping_update_lock
);
604 if ((pirq
> nr_irqs
) || (gsi
> nr_irqs
)) {
605 printk(KERN_WARNING
"xen_map_pirq_gsi: %s %s is incorrect!\n",
606 pirq
> nr_irqs
? "pirq" :"",
607 gsi
> nr_irqs
? "gsi" : "");
611 irq
= find_irq_by_gsi(gsi
);
613 printk(KERN_INFO
"xen_map_pirq_gsi: returning irq %d for gsi %u\n",
615 goto out
; /* XXX need refcount? */
618 irq
= xen_allocate_irq_gsi(gsi
);
620 set_irq_chip_and_handler_name(irq
, &xen_pirq_chip
,
621 handle_level_irq
, name
);
626 /* Only the privileged domain can do this. For non-priv, the pcifront
627 * driver provides a PCI bus that does the call to do exactly
628 * this in the priv domain. */
629 if (xen_initial_domain() &&
630 HYPERVISOR_physdev_op(PHYSDEVOP_alloc_irq_vector
, &irq_op
)) {
636 irq_info
[irq
] = mk_pirq_info(0, pirq
, gsi
, irq_op
.vector
);
637 irq_info
[irq
].u
.pirq
.flags
|= shareable
? PIRQ_SHAREABLE
: 0;
638 pirq_to_irq
[pirq
] = irq
;
641 spin_unlock(&irq_mapping_update_lock
);
646 #ifdef CONFIG_PCI_MSI
647 #include <linux/msi.h>
648 #include "../pci/msi.h"
650 int xen_allocate_pirq_msi(struct pci_dev
*dev
, struct msi_desc
*msidesc
)
653 struct physdev_get_free_pirq op_get_free_pirq
;
655 op_get_free_pirq
.type
= MAP_PIRQ_TYPE_MSI
;
656 rc
= HYPERVISOR_physdev_op(PHYSDEVOP_get_free_pirq
, &op_get_free_pirq
);
658 WARN_ONCE(rc
== -ENOSYS
,
659 "hypervisor does not support the PHYSDEVOP_get_free_pirq interface\n");
661 return rc
? -1 : op_get_free_pirq
.pirq
;
664 int xen_bind_pirq_msi_to_irq(struct pci_dev
*dev
, struct msi_desc
*msidesc
,
665 int pirq
, const char *name
)
669 spin_lock(&irq_mapping_update_lock
);
671 irq
= xen_allocate_irq_dynamic();
675 set_irq_chip_and_handler_name(irq
, &xen_pirq_chip
,
676 handle_level_irq
, name
);
678 irq_info
[irq
] = mk_pirq_info(0, pirq
, 0, 0);
679 pirq_to_irq
[pirq
] = irq
;
680 ret
= set_irq_msi(irq
, msidesc
);
684 spin_unlock(&irq_mapping_update_lock
);
687 spin_unlock(&irq_mapping_update_lock
);
692 int xen_create_msi_irq(struct pci_dev
*dev
, struct msi_desc
*msidesc
, int type
)
695 struct physdev_map_pirq map_irq
;
698 u32 table_offset
, bir
;
700 memset(&map_irq
, 0, sizeof(map_irq
));
701 map_irq
.domid
= DOMID_SELF
;
702 map_irq
.type
= MAP_PIRQ_TYPE_MSI
;
705 map_irq
.bus
= dev
->bus
->number
;
706 map_irq
.devfn
= dev
->devfn
;
708 if (type
== PCI_CAP_ID_MSIX
) {
709 pos
= pci_find_capability(dev
, PCI_CAP_ID_MSIX
);
711 pci_read_config_dword(dev
, msix_table_offset_reg(pos
),
713 bir
= (u8
)(table_offset
& PCI_MSIX_FLAGS_BIRMASK
);
715 map_irq
.table_base
= pci_resource_start(dev
, bir
);
716 map_irq
.entry_nr
= msidesc
->msi_attrib
.entry_nr
;
719 rc
= HYPERVISOR_physdev_op(PHYSDEVOP_map_pirq
, &map_irq
);
721 dev_warn(&dev
->dev
, "xen map irq failed %d\n", rc
);
725 spin_lock(&irq_mapping_update_lock
);
727 irq
= xen_allocate_irq_dynamic();
732 irq_info
[irq
] = mk_pirq_info(0, map_irq
.pirq
, 0, map_irq
.index
);
734 set_irq_chip_and_handler_name(irq
, &xen_pirq_chip
,
736 (type
== PCI_CAP_ID_MSIX
) ? "msi-x":"msi");
739 spin_unlock(&irq_mapping_update_lock
);
744 int xen_destroy_irq(int irq
)
746 struct irq_desc
*desc
;
747 struct physdev_unmap_pirq unmap_irq
;
748 struct irq_info
*info
= info_for_irq(irq
);
751 spin_lock(&irq_mapping_update_lock
);
753 desc
= irq_to_desc(irq
);
757 if (xen_initial_domain()) {
758 unmap_irq
.pirq
= info
->u
.pirq
.pirq
;
759 unmap_irq
.domid
= DOMID_SELF
;
760 rc
= HYPERVISOR_physdev_op(PHYSDEVOP_unmap_pirq
, &unmap_irq
);
762 printk(KERN_WARNING
"unmap irq failed %d\n", rc
);
766 pirq_to_irq
[info
->u
.pirq
.pirq
] = -1;
768 irq_info
[irq
] = mk_unbound_info();
773 spin_unlock(&irq_mapping_update_lock
);
777 int xen_vector_from_irq(unsigned irq
)
779 return vector_from_irq(irq
);
782 int xen_gsi_from_irq(unsigned irq
)
784 return gsi_from_irq(irq
);
787 int xen_irq_from_pirq(unsigned pirq
)
789 return pirq_to_irq
[pirq
];
792 int bind_evtchn_to_irq(unsigned int evtchn
)
796 spin_lock(&irq_mapping_update_lock
);
798 irq
= evtchn_to_irq
[evtchn
];
801 irq
= xen_allocate_irq_dynamic();
803 set_irq_chip_and_handler_name(irq
, &xen_dynamic_chip
,
804 handle_fasteoi_irq
, "event");
806 evtchn_to_irq
[evtchn
] = irq
;
807 irq_info
[irq
] = mk_evtchn_info(evtchn
);
810 spin_unlock(&irq_mapping_update_lock
);
814 EXPORT_SYMBOL_GPL(bind_evtchn_to_irq
);
816 static int bind_ipi_to_irq(unsigned int ipi
, unsigned int cpu
)
818 struct evtchn_bind_ipi bind_ipi
;
821 spin_lock(&irq_mapping_update_lock
);
823 irq
= per_cpu(ipi_to_irq
, cpu
)[ipi
];
826 irq
= xen_allocate_irq_dynamic();
830 set_irq_chip_and_handler_name(irq
, &xen_percpu_chip
,
831 handle_percpu_irq
, "ipi");
834 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_ipi
,
837 evtchn
= bind_ipi
.port
;
839 evtchn_to_irq
[evtchn
] = irq
;
840 irq_info
[irq
] = mk_ipi_info(evtchn
, ipi
);
841 per_cpu(ipi_to_irq
, cpu
)[ipi
] = irq
;
843 bind_evtchn_to_cpu(evtchn
, cpu
);
847 spin_unlock(&irq_mapping_update_lock
);
852 int bind_virq_to_irq(unsigned int virq
, unsigned int cpu
)
854 struct evtchn_bind_virq bind_virq
;
857 spin_lock(&irq_mapping_update_lock
);
859 irq
= per_cpu(virq_to_irq
, cpu
)[virq
];
862 irq
= xen_allocate_irq_dynamic();
864 set_irq_chip_and_handler_name(irq
, &xen_percpu_chip
,
865 handle_percpu_irq
, "virq");
867 bind_virq
.virq
= virq
;
868 bind_virq
.vcpu
= cpu
;
869 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_virq
,
872 evtchn
= bind_virq
.port
;
874 evtchn_to_irq
[evtchn
] = irq
;
875 irq_info
[irq
] = mk_virq_info(evtchn
, virq
);
877 per_cpu(virq_to_irq
, cpu
)[virq
] = irq
;
879 bind_evtchn_to_cpu(evtchn
, cpu
);
882 spin_unlock(&irq_mapping_update_lock
);
887 static void unbind_from_irq(unsigned int irq
)
889 struct evtchn_close close
;
890 int evtchn
= evtchn_from_irq(irq
);
892 spin_lock(&irq_mapping_update_lock
);
894 if (VALID_EVTCHN(evtchn
)) {
896 if (HYPERVISOR_event_channel_op(EVTCHNOP_close
, &close
) != 0)
899 switch (type_from_irq(irq
)) {
901 per_cpu(virq_to_irq
, cpu_from_evtchn(evtchn
))
902 [virq_from_irq(irq
)] = -1;
905 per_cpu(ipi_to_irq
, cpu_from_evtchn(evtchn
))
906 [ipi_from_irq(irq
)] = -1;
912 /* Closed ports are implicitly re-bound to VCPU0. */
913 bind_evtchn_to_cpu(evtchn
, 0);
915 evtchn_to_irq
[evtchn
] = -1;
918 if (irq_info
[irq
].type
!= IRQT_UNBOUND
) {
919 irq_info
[irq
] = mk_unbound_info();
924 spin_unlock(&irq_mapping_update_lock
);
927 int bind_evtchn_to_irqhandler(unsigned int evtchn
,
928 irq_handler_t handler
,
929 unsigned long irqflags
,
930 const char *devname
, void *dev_id
)
935 irq
= bind_evtchn_to_irq(evtchn
);
936 retval
= request_irq(irq
, handler
, irqflags
, devname
, dev_id
);
938 unbind_from_irq(irq
);
944 EXPORT_SYMBOL_GPL(bind_evtchn_to_irqhandler
);
946 int bind_virq_to_irqhandler(unsigned int virq
, unsigned int cpu
,
947 irq_handler_t handler
,
948 unsigned long irqflags
, const char *devname
, void *dev_id
)
953 irq
= bind_virq_to_irq(virq
, cpu
);
954 retval
= request_irq(irq
, handler
, irqflags
, devname
, dev_id
);
956 unbind_from_irq(irq
);
962 EXPORT_SYMBOL_GPL(bind_virq_to_irqhandler
);
964 int bind_ipi_to_irqhandler(enum ipi_vector ipi
,
966 irq_handler_t handler
,
967 unsigned long irqflags
,
973 irq
= bind_ipi_to_irq(ipi
, cpu
);
977 irqflags
|= IRQF_NO_SUSPEND
| IRQF_FORCE_RESUME
;
978 retval
= request_irq(irq
, handler
, irqflags
, devname
, dev_id
);
980 unbind_from_irq(irq
);
987 void unbind_from_irqhandler(unsigned int irq
, void *dev_id
)
989 free_irq(irq
, dev_id
);
990 unbind_from_irq(irq
);
992 EXPORT_SYMBOL_GPL(unbind_from_irqhandler
);
994 void xen_send_IPI_one(unsigned int cpu
, enum ipi_vector vector
)
996 int irq
= per_cpu(ipi_to_irq
, cpu
)[vector
];
998 notify_remote_via_irq(irq
);
1001 irqreturn_t
xen_debug_interrupt(int irq
, void *dev_id
)
1003 struct shared_info
*sh
= HYPERVISOR_shared_info
;
1004 int cpu
= smp_processor_id();
1005 unsigned long *cpu_evtchn
= cpu_evtchn_mask(cpu
);
1007 unsigned long flags
;
1008 static DEFINE_SPINLOCK(debug_lock
);
1009 struct vcpu_info
*v
;
1011 spin_lock_irqsave(&debug_lock
, flags
);
1013 printk("\nvcpu %d\n ", cpu
);
1015 for_each_online_cpu(i
) {
1017 v
= per_cpu(xen_vcpu
, i
);
1018 pending
= (get_irq_regs() && i
== cpu
)
1019 ? xen_irqs_disabled(get_irq_regs())
1020 : v
->evtchn_upcall_mask
;
1021 printk("%d: masked=%d pending=%d event_sel %0*lx\n ", i
,
1022 pending
, v
->evtchn_upcall_pending
,
1023 (int)(sizeof(v
->evtchn_pending_sel
)*2),
1024 v
->evtchn_pending_sel
);
1026 v
= per_cpu(xen_vcpu
, cpu
);
1028 printk("\npending:\n ");
1029 for (i
= ARRAY_SIZE(sh
->evtchn_pending
)-1; i
>= 0; i
--)
1030 printk("%0*lx%s", (int)sizeof(sh
->evtchn_pending
[0])*2,
1031 sh
->evtchn_pending
[i
],
1032 i
% 8 == 0 ? "\n " : " ");
1033 printk("\nglobal mask:\n ");
1034 for (i
= ARRAY_SIZE(sh
->evtchn_mask
)-1; i
>= 0; i
--)
1036 (int)(sizeof(sh
->evtchn_mask
[0])*2),
1038 i
% 8 == 0 ? "\n " : " ");
1040 printk("\nglobally unmasked:\n ");
1041 for (i
= ARRAY_SIZE(sh
->evtchn_mask
)-1; i
>= 0; i
--)
1042 printk("%0*lx%s", (int)(sizeof(sh
->evtchn_mask
[0])*2),
1043 sh
->evtchn_pending
[i
] & ~sh
->evtchn_mask
[i
],
1044 i
% 8 == 0 ? "\n " : " ");
1046 printk("\nlocal cpu%d mask:\n ", cpu
);
1047 for (i
= (NR_EVENT_CHANNELS
/BITS_PER_LONG
)-1; i
>= 0; i
--)
1048 printk("%0*lx%s", (int)(sizeof(cpu_evtchn
[0])*2),
1050 i
% 8 == 0 ? "\n " : " ");
1052 printk("\nlocally unmasked:\n ");
1053 for (i
= ARRAY_SIZE(sh
->evtchn_mask
)-1; i
>= 0; i
--) {
1054 unsigned long pending
= sh
->evtchn_pending
[i
]
1055 & ~sh
->evtchn_mask
[i
]
1057 printk("%0*lx%s", (int)(sizeof(sh
->evtchn_mask
[0])*2),
1058 pending
, i
% 8 == 0 ? "\n " : " ");
1061 printk("\npending list:\n");
1062 for (i
= 0; i
< NR_EVENT_CHANNELS
; i
++) {
1063 if (sync_test_bit(i
, sh
->evtchn_pending
)) {
1064 int word_idx
= i
/ BITS_PER_LONG
;
1065 printk(" %d: event %d -> irq %d%s%s%s\n",
1066 cpu_from_evtchn(i
), i
,
1068 sync_test_bit(word_idx
, &v
->evtchn_pending_sel
)
1070 !sync_test_bit(i
, sh
->evtchn_mask
)
1071 ? "" : " globally-masked",
1072 sync_test_bit(i
, cpu_evtchn
)
1073 ? "" : " locally-masked");
1077 spin_unlock_irqrestore(&debug_lock
, flags
);
1082 static DEFINE_PER_CPU(unsigned, xed_nesting_count
);
1085 * Search the CPUs pending events bitmasks. For each one found, map
1086 * the event number to an irq, and feed it into do_IRQ() for
1089 * Xen uses a two-level bitmap to speed searching. The first level is
1090 * a bitset of words which contain pending event bits. The second
1091 * level is a bitset of pending events themselves.
1093 static void __xen_evtchn_do_upcall(void)
1095 int cpu
= get_cpu();
1096 struct shared_info
*s
= HYPERVISOR_shared_info
;
1097 struct vcpu_info
*vcpu_info
= __this_cpu_read(xen_vcpu
);
1101 unsigned long pending_words
;
1103 vcpu_info
->evtchn_upcall_pending
= 0;
1105 if (__this_cpu_inc_return(xed_nesting_count
) - 1)
1108 #ifndef CONFIG_X86 /* No need for a barrier -- XCHG is a barrier on x86. */
1109 /* Clear master flag /before/ clearing selector flag. */
1112 pending_words
= xchg(&vcpu_info
->evtchn_pending_sel
, 0);
1113 while (pending_words
!= 0) {
1114 unsigned long pending_bits
;
1115 int word_idx
= __ffs(pending_words
);
1116 pending_words
&= ~(1UL << word_idx
);
1118 while ((pending_bits
= active_evtchns(cpu
, s
, word_idx
)) != 0) {
1119 int bit_idx
= __ffs(pending_bits
);
1120 int port
= (word_idx
* BITS_PER_LONG
) + bit_idx
;
1121 int irq
= evtchn_to_irq
[port
];
1122 struct irq_desc
*desc
;
1128 desc
= irq_to_desc(irq
);
1130 generic_handle_irq_desc(irq
, desc
);
1135 BUG_ON(!irqs_disabled());
1137 count
= __this_cpu_read(xed_nesting_count
);
1138 __this_cpu_write(xed_nesting_count
, 0);
1139 } while (count
!= 1 || vcpu_info
->evtchn_upcall_pending
);
1146 void xen_evtchn_do_upcall(struct pt_regs
*regs
)
1148 struct pt_regs
*old_regs
= set_irq_regs(regs
);
1153 __xen_evtchn_do_upcall();
1156 set_irq_regs(old_regs
);
1159 void xen_hvm_evtchn_do_upcall(void)
1161 __xen_evtchn_do_upcall();
1163 EXPORT_SYMBOL_GPL(xen_hvm_evtchn_do_upcall
);
1165 /* Rebind a new event channel to an existing irq. */
1166 void rebind_evtchn_irq(int evtchn
, int irq
)
1168 struct irq_info
*info
= info_for_irq(irq
);
1170 /* Make sure the irq is masked, since the new event channel
1171 will also be masked. */
1174 spin_lock(&irq_mapping_update_lock
);
1176 /* After resume the irq<->evtchn mappings are all cleared out */
1177 BUG_ON(evtchn_to_irq
[evtchn
] != -1);
1178 /* Expect irq to have been bound before,
1179 so there should be a proper type */
1180 BUG_ON(info
->type
== IRQT_UNBOUND
);
1182 evtchn_to_irq
[evtchn
] = irq
;
1183 irq_info
[irq
] = mk_evtchn_info(evtchn
);
1185 spin_unlock(&irq_mapping_update_lock
);
1187 /* new event channels are always bound to cpu 0 */
1188 irq_set_affinity(irq
, cpumask_of(0));
1190 /* Unmask the event channel. */
1194 /* Rebind an evtchn so that it gets delivered to a specific cpu */
1195 static int rebind_irq_to_cpu(unsigned irq
, unsigned tcpu
)
1197 struct evtchn_bind_vcpu bind_vcpu
;
1198 int evtchn
= evtchn_from_irq(irq
);
1200 /* events delivered via platform PCI interrupts are always
1201 * routed to vcpu 0 */
1202 if (!VALID_EVTCHN(evtchn
) ||
1203 (xen_hvm_domain() && !xen_have_vector_callback
))
1206 /* Send future instances of this interrupt to other vcpu. */
1207 bind_vcpu
.port
= evtchn
;
1208 bind_vcpu
.vcpu
= tcpu
;
1211 * If this fails, it usually just indicates that we're dealing with a
1212 * virq or IPI channel, which don't actually need to be rebound. Ignore
1213 * it, but don't do the xenlinux-level rebind in that case.
1215 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_vcpu
, &bind_vcpu
) >= 0)
1216 bind_evtchn_to_cpu(evtchn
, tcpu
);
1221 static int set_affinity_irq(struct irq_data
*data
, const struct cpumask
*dest
,
1224 unsigned tcpu
= cpumask_first(dest
);
1226 return rebind_irq_to_cpu(data
->irq
, tcpu
);
1229 int resend_irq_on_evtchn(unsigned int irq
)
1231 int masked
, evtchn
= evtchn_from_irq(irq
);
1232 struct shared_info
*s
= HYPERVISOR_shared_info
;
1234 if (!VALID_EVTCHN(evtchn
))
1237 masked
= sync_test_and_set_bit(evtchn
, s
->evtchn_mask
);
1238 sync_set_bit(evtchn
, s
->evtchn_pending
);
1240 unmask_evtchn(evtchn
);
1245 static void enable_dynirq(struct irq_data
*data
)
1247 int evtchn
= evtchn_from_irq(data
->irq
);
1249 if (VALID_EVTCHN(evtchn
))
1250 unmask_evtchn(evtchn
);
1253 static void disable_dynirq(struct irq_data
*data
)
1255 int evtchn
= evtchn_from_irq(data
->irq
);
1257 if (VALID_EVTCHN(evtchn
))
1258 mask_evtchn(evtchn
);
1261 static void ack_dynirq(struct irq_data
*data
)
1263 int evtchn
= evtchn_from_irq(data
->irq
);
1265 move_masked_irq(data
->irq
);
1267 if (VALID_EVTCHN(evtchn
))
1268 unmask_evtchn(evtchn
);
1271 static int retrigger_dynirq(struct irq_data
*data
)
1273 int evtchn
= evtchn_from_irq(data
->irq
);
1274 struct shared_info
*sh
= HYPERVISOR_shared_info
;
1277 if (VALID_EVTCHN(evtchn
)) {
1280 masked
= sync_test_and_set_bit(evtchn
, sh
->evtchn_mask
);
1281 sync_set_bit(evtchn
, sh
->evtchn_pending
);
1283 unmask_evtchn(evtchn
);
1290 static void restore_cpu_pirqs(void)
1292 int pirq
, rc
, irq
, gsi
;
1293 struct physdev_map_pirq map_irq
;
1295 for (pirq
= 0; pirq
< nr_irqs
; pirq
++) {
1296 irq
= pirq_to_irq
[pirq
];
1300 /* save/restore of PT devices doesn't work, so at this point the
1301 * only devices present are GSI based emulated devices */
1302 gsi
= gsi_from_irq(irq
);
1306 map_irq
.domid
= DOMID_SELF
;
1307 map_irq
.type
= MAP_PIRQ_TYPE_GSI
;
1308 map_irq
.index
= gsi
;
1309 map_irq
.pirq
= pirq
;
1311 rc
= HYPERVISOR_physdev_op(PHYSDEVOP_map_pirq
, &map_irq
);
1313 printk(KERN_WARNING
"xen map irq failed gsi=%d irq=%d pirq=%d rc=%d\n",
1314 gsi
, irq
, pirq
, rc
);
1315 irq_info
[irq
] = mk_unbound_info();
1316 pirq_to_irq
[pirq
] = -1;
1320 printk(KERN_DEBUG
"xen: --> irq=%d, pirq=%d\n", irq
, map_irq
.pirq
);
1322 __startup_pirq(irq
);
1326 static void restore_cpu_virqs(unsigned int cpu
)
1328 struct evtchn_bind_virq bind_virq
;
1329 int virq
, irq
, evtchn
;
1331 for (virq
= 0; virq
< NR_VIRQS
; virq
++) {
1332 if ((irq
= per_cpu(virq_to_irq
, cpu
)[virq
]) == -1)
1335 BUG_ON(virq_from_irq(irq
) != virq
);
1337 /* Get a new binding from Xen. */
1338 bind_virq
.virq
= virq
;
1339 bind_virq
.vcpu
= cpu
;
1340 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_virq
,
1343 evtchn
= bind_virq
.port
;
1345 /* Record the new mapping. */
1346 evtchn_to_irq
[evtchn
] = irq
;
1347 irq_info
[irq
] = mk_virq_info(evtchn
, virq
);
1348 bind_evtchn_to_cpu(evtchn
, cpu
);
1352 static void restore_cpu_ipis(unsigned int cpu
)
1354 struct evtchn_bind_ipi bind_ipi
;
1355 int ipi
, irq
, evtchn
;
1357 for (ipi
= 0; ipi
< XEN_NR_IPIS
; ipi
++) {
1358 if ((irq
= per_cpu(ipi_to_irq
, cpu
)[ipi
]) == -1)
1361 BUG_ON(ipi_from_irq(irq
) != ipi
);
1363 /* Get a new binding from Xen. */
1364 bind_ipi
.vcpu
= cpu
;
1365 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_ipi
,
1368 evtchn
= bind_ipi
.port
;
1370 /* Record the new mapping. */
1371 evtchn_to_irq
[evtchn
] = irq
;
1372 irq_info
[irq
] = mk_ipi_info(evtchn
, ipi
);
1373 bind_evtchn_to_cpu(evtchn
, cpu
);
1377 /* Clear an irq's pending state, in preparation for polling on it */
1378 void xen_clear_irq_pending(int irq
)
1380 int evtchn
= evtchn_from_irq(irq
);
1382 if (VALID_EVTCHN(evtchn
))
1383 clear_evtchn(evtchn
);
1385 EXPORT_SYMBOL(xen_clear_irq_pending
);
1386 void xen_set_irq_pending(int irq
)
1388 int evtchn
= evtchn_from_irq(irq
);
1390 if (VALID_EVTCHN(evtchn
))
1394 bool xen_test_irq_pending(int irq
)
1396 int evtchn
= evtchn_from_irq(irq
);
1399 if (VALID_EVTCHN(evtchn
))
1400 ret
= test_evtchn(evtchn
);
1405 /* Poll waiting for an irq to become pending with timeout. In the usual case,
1406 * the irq will be disabled so it won't deliver an interrupt. */
1407 void xen_poll_irq_timeout(int irq
, u64 timeout
)
1409 evtchn_port_t evtchn
= evtchn_from_irq(irq
);
1411 if (VALID_EVTCHN(evtchn
)) {
1412 struct sched_poll poll
;
1415 poll
.timeout
= timeout
;
1416 set_xen_guest_handle(poll
.ports
, &evtchn
);
1418 if (HYPERVISOR_sched_op(SCHEDOP_poll
, &poll
) != 0)
1422 EXPORT_SYMBOL(xen_poll_irq_timeout
);
1423 /* Poll waiting for an irq to become pending. In the usual case, the
1424 * irq will be disabled so it won't deliver an interrupt. */
1425 void xen_poll_irq(int irq
)
1427 xen_poll_irq_timeout(irq
, 0 /* no timeout */);
1430 void xen_irq_resume(void)
1432 unsigned int cpu
, irq
, evtchn
;
1434 init_evtchn_cpu_bindings();
1436 /* New event-channel space is not 'live' yet. */
1437 for (evtchn
= 0; evtchn
< NR_EVENT_CHANNELS
; evtchn
++)
1438 mask_evtchn(evtchn
);
1440 /* No IRQ <-> event-channel mappings. */
1441 for (irq
= 0; irq
< nr_irqs
; irq
++)
1442 irq_info
[irq
].evtchn
= 0; /* zap event-channel binding */
1444 for (evtchn
= 0; evtchn
< NR_EVENT_CHANNELS
; evtchn
++)
1445 evtchn_to_irq
[evtchn
] = -1;
1447 for_each_possible_cpu(cpu
) {
1448 restore_cpu_virqs(cpu
);
1449 restore_cpu_ipis(cpu
);
1452 restore_cpu_pirqs();
1455 static struct irq_chip xen_dynamic_chip __read_mostly
= {
1458 .irq_disable
= disable_dynirq
,
1459 .irq_mask
= disable_dynirq
,
1460 .irq_unmask
= enable_dynirq
,
1462 .irq_eoi
= ack_dynirq
,
1463 .irq_set_affinity
= set_affinity_irq
,
1464 .irq_retrigger
= retrigger_dynirq
,
1467 static struct irq_chip xen_pirq_chip __read_mostly
= {
1470 .irq_startup
= startup_pirq
,
1471 .irq_shutdown
= shutdown_pirq
,
1473 .irq_enable
= enable_pirq
,
1474 .irq_unmask
= enable_pirq
,
1476 .irq_disable
= disable_pirq
,
1477 .irq_mask
= disable_pirq
,
1479 .irq_ack
= ack_pirq
,
1481 .irq_set_affinity
= set_affinity_irq
,
1483 .irq_retrigger
= retrigger_dynirq
,
1486 static struct irq_chip xen_percpu_chip __read_mostly
= {
1487 .name
= "xen-percpu",
1489 .irq_disable
= disable_dynirq
,
1490 .irq_mask
= disable_dynirq
,
1491 .irq_unmask
= enable_dynirq
,
1493 .irq_ack
= ack_dynirq
,
1496 int xen_set_callback_via(uint64_t via
)
1498 struct xen_hvm_param a
;
1499 a
.domid
= DOMID_SELF
;
1500 a
.index
= HVM_PARAM_CALLBACK_IRQ
;
1502 return HYPERVISOR_hvm_op(HVMOP_set_param
, &a
);
1504 EXPORT_SYMBOL_GPL(xen_set_callback_via
);
1506 #ifdef CONFIG_XEN_PVHVM
1507 /* Vector callbacks are better than PCI interrupts to receive event
1508 * channel notifications because we can receive vector callbacks on any
1509 * vcpu and we don't need PCI support or APIC interactions. */
1510 void xen_callback_vector(void)
1513 uint64_t callback_via
;
1514 if (xen_have_vector_callback
) {
1515 callback_via
= HVM_CALLBACK_VECTOR(XEN_HVM_EVTCHN_CALLBACK
);
1516 rc
= xen_set_callback_via(callback_via
);
1518 printk(KERN_ERR
"Request for Xen HVM callback vector"
1520 xen_have_vector_callback
= 0;
1523 printk(KERN_INFO
"Xen HVM callback vector for event delivery is "
1525 /* in the restore case the vector has already been allocated */
1526 if (!test_bit(XEN_HVM_EVTCHN_CALLBACK
, used_vectors
))
1527 alloc_intr_gate(XEN_HVM_EVTCHN_CALLBACK
, xen_hvm_callback_vector
);
1531 void xen_callback_vector(void) {}
1534 void __init
xen_init_IRQ(void)
1538 cpu_evtchn_mask_p
= kcalloc(nr_cpu_ids
, sizeof(struct cpu_evtchn_s
),
1540 irq_info
= kcalloc(nr_irqs
, sizeof(*irq_info
), GFP_KERNEL
);
1542 /* We are using nr_irqs as the maximum number of pirq available but
1543 * that number is actually chosen by Xen and we don't know exactly
1544 * what it is. Be careful choosing high pirq numbers. */
1545 pirq_to_irq
= kcalloc(nr_irqs
, sizeof(*pirq_to_irq
), GFP_KERNEL
);
1546 for (i
= 0; i
< nr_irqs
; i
++)
1547 pirq_to_irq
[i
] = -1;
1549 evtchn_to_irq
= kcalloc(NR_EVENT_CHANNELS
, sizeof(*evtchn_to_irq
),
1551 for (i
= 0; i
< NR_EVENT_CHANNELS
; i
++)
1552 evtchn_to_irq
[i
] = -1;
1554 init_evtchn_cpu_bindings();
1556 /* No event channels are 'live' right now. */
1557 for (i
= 0; i
< NR_EVENT_CHANNELS
; i
++)
1560 if (xen_hvm_domain()) {
1561 xen_callback_vector();
1563 /* pci_xen_hvm_init must be called after native_init_IRQ so that
1564 * __acpi_register_gsi can point at the right function */
1567 irq_ctx_init(smp_processor_id());
1568 if (xen_initial_domain())