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 static LIST_HEAD(xen_irq_list_head);
61 /* IRQ <-> VIRQ mapping. */
62 static DEFINE_PER_CPU(int [NR_VIRQS], virq_to_irq) = {[0 ... NR_VIRQS-1] = -1};
64 /* IRQ <-> IPI mapping */
65 static DEFINE_PER_CPU(int [XEN_NR_IPIS], ipi_to_irq) = {[0 ... XEN_NR_IPIS-1] = -1};
67 /* Interrupt types. */
77 * Packed IRQ information:
78 * type - enum xen_irq_type
79 * event channel - irq->event channel mapping
80 * cpu - cpu this event channel is bound to
81 * index - type-specific information:
82 * PIRQ - vector, with MSB being "needs EIO", or physical IRQ of the HVM
83 * guest, or GSI (real passthrough IRQ) of the device.
90 struct list_head list;
91 enum xen_irq_type type; /* type */
93 unsigned short evtchn; /* event channel */
94 unsigned short cpu; /* cpu bound */
102 unsigned char vector;
107 #define PIRQ_NEEDS_EOI (1 << 0)
108 #define PIRQ_SHAREABLE (1 << 1)
110 static struct irq_info *irq_info;
111 static int *pirq_to_irq;
113 static int *evtchn_to_irq;
115 static DEFINE_PER_CPU(unsigned long [NR_EVENT_CHANNELS/BITS_PER_LONG],
118 /* Xen will never allocate port zero for any purpose. */
119 #define VALID_EVTCHN(chn) ((chn) != 0)
121 static struct irq_chip xen_dynamic_chip;
122 static struct irq_chip xen_percpu_chip;
123 static struct irq_chip xen_pirq_chip;
125 /* Get info for IRQ */
126 static struct irq_info *info_for_irq(unsigned irq)
128 return &irq_info[irq];
131 /* Constructors for packed IRQ information. */
132 static void xen_irq_info_common_init(struct irq_info *info,
134 enum xen_irq_type type,
135 unsigned short evtchn,
139 BUG_ON(info->type != IRQT_UNBOUND && info->type != type);
143 info->evtchn = evtchn;
146 evtchn_to_irq[evtchn] = irq;
149 static void xen_irq_info_evtchn_init(unsigned irq,
150 unsigned short evtchn)
152 struct irq_info *info = info_for_irq(irq);
154 xen_irq_info_common_init(info, irq, IRQT_EVTCHN, evtchn, 0);
157 static void xen_irq_info_ipi_init(unsigned cpu,
159 unsigned short evtchn,
162 struct irq_info *info = info_for_irq(irq);
164 xen_irq_info_common_init(info, irq, IRQT_IPI, evtchn, 0);
168 per_cpu(ipi_to_irq, cpu)[ipi] = irq;
171 static void xen_irq_info_virq_init(unsigned cpu,
173 unsigned short evtchn,
176 struct irq_info *info = info_for_irq(irq);
178 xen_irq_info_common_init(info, irq, IRQT_VIRQ, evtchn, 0);
182 per_cpu(virq_to_irq, cpu)[virq] = irq;
185 static void xen_irq_info_pirq_init(unsigned irq,
186 unsigned short evtchn,
189 unsigned short vector,
192 struct irq_info *info = info_for_irq(irq);
194 xen_irq_info_common_init(info, irq, IRQT_PIRQ, evtchn, 0);
196 info->u.pirq.pirq = pirq;
197 info->u.pirq.gsi = gsi;
198 info->u.pirq.vector = vector;
199 info->u.pirq.flags = flags;
201 pirq_to_irq[pirq] = irq;
205 * Accessors for packed IRQ information.
207 static unsigned int evtchn_from_irq(unsigned irq)
209 if (unlikely(WARN(irq < 0 || irq >= nr_irqs, "Invalid irq %d!\n", irq)))
212 return info_for_irq(irq)->evtchn;
215 unsigned irq_from_evtchn(unsigned int evtchn)
217 return evtchn_to_irq[evtchn];
219 EXPORT_SYMBOL_GPL(irq_from_evtchn);
221 static enum ipi_vector ipi_from_irq(unsigned irq)
223 struct irq_info *info = info_for_irq(irq);
225 BUG_ON(info == NULL);
226 BUG_ON(info->type != IRQT_IPI);
231 static unsigned virq_from_irq(unsigned irq)
233 struct irq_info *info = info_for_irq(irq);
235 BUG_ON(info == NULL);
236 BUG_ON(info->type != IRQT_VIRQ);
241 static unsigned pirq_from_irq(unsigned irq)
243 struct irq_info *info = info_for_irq(irq);
245 BUG_ON(info == NULL);
246 BUG_ON(info->type != IRQT_PIRQ);
248 return info->u.pirq.pirq;
251 static unsigned gsi_from_irq(unsigned irq)
253 struct irq_info *info = info_for_irq(irq);
255 BUG_ON(info == NULL);
256 BUG_ON(info->type != IRQT_PIRQ);
258 return info->u.pirq.gsi;
261 static enum xen_irq_type type_from_irq(unsigned irq)
263 return info_for_irq(irq)->type;
266 static unsigned cpu_from_irq(unsigned irq)
268 return info_for_irq(irq)->cpu;
271 static unsigned int cpu_from_evtchn(unsigned int evtchn)
273 int irq = evtchn_to_irq[evtchn];
277 ret = cpu_from_irq(irq);
282 static bool pirq_needs_eoi(unsigned irq)
284 struct irq_info *info = info_for_irq(irq);
286 BUG_ON(info->type != IRQT_PIRQ);
288 return info->u.pirq.flags & PIRQ_NEEDS_EOI;
291 static inline unsigned long active_evtchns(unsigned int cpu,
292 struct shared_info *sh,
295 return (sh->evtchn_pending[idx] &
296 per_cpu(cpu_evtchn_mask, cpu)[idx] &
297 ~sh->evtchn_mask[idx]);
300 static void bind_evtchn_to_cpu(unsigned int chn, unsigned int cpu)
302 int irq = evtchn_to_irq[chn];
306 cpumask_copy(irq_to_desc(irq)->irq_data.affinity, cpumask_of(cpu));
309 clear_bit(chn, per_cpu(cpu_evtchn_mask, cpu_from_irq(irq)));
310 set_bit(chn, per_cpu(cpu_evtchn_mask, cpu));
312 irq_info[irq].cpu = cpu;
315 static void init_evtchn_cpu_bindings(void)
319 struct irq_info *info;
321 /* By default all event channels notify CPU#0. */
322 list_for_each_entry(info, &xen_irq_list_head, list) {
323 struct irq_desc *desc = irq_to_desc(info->irq);
324 cpumask_copy(desc->irq_data.affinity, cpumask_of(0));
328 for_each_possible_cpu(i)
329 memset(per_cpu(cpu_evtchn_mask, i),
330 (i == 0) ? ~0 : 0, sizeof(*per_cpu(cpu_evtchn_mask, i)));
334 static inline void clear_evtchn(int port)
336 struct shared_info *s = HYPERVISOR_shared_info;
337 sync_clear_bit(port, &s->evtchn_pending[0]);
340 static inline void set_evtchn(int port)
342 struct shared_info *s = HYPERVISOR_shared_info;
343 sync_set_bit(port, &s->evtchn_pending[0]);
346 static inline int test_evtchn(int port)
348 struct shared_info *s = HYPERVISOR_shared_info;
349 return sync_test_bit(port, &s->evtchn_pending[0]);
354 * notify_remote_via_irq - send event to remote end of event channel via irq
355 * @irq: irq of event channel to send event to
357 * Unlike notify_remote_via_evtchn(), this is safe to use across
358 * save/restore. Notifications on a broken connection are silently
361 void notify_remote_via_irq(int irq)
363 int evtchn = evtchn_from_irq(irq);
365 if (VALID_EVTCHN(evtchn))
366 notify_remote_via_evtchn(evtchn);
368 EXPORT_SYMBOL_GPL(notify_remote_via_irq);
370 static void mask_evtchn(int port)
372 struct shared_info *s = HYPERVISOR_shared_info;
373 sync_set_bit(port, &s->evtchn_mask[0]);
376 static void unmask_evtchn(int port)
378 struct shared_info *s = HYPERVISOR_shared_info;
379 unsigned int cpu = get_cpu();
381 BUG_ON(!irqs_disabled());
383 /* Slow path (hypercall) if this is a non-local port. */
384 if (unlikely(cpu != cpu_from_evtchn(port))) {
385 struct evtchn_unmask unmask = { .port = port };
386 (void)HYPERVISOR_event_channel_op(EVTCHNOP_unmask, &unmask);
388 struct vcpu_info *vcpu_info = __this_cpu_read(xen_vcpu);
390 sync_clear_bit(port, &s->evtchn_mask[0]);
393 * The following is basically the equivalent of
394 * 'hw_resend_irq'. Just like a real IO-APIC we 'lose
395 * the interrupt edge' if the channel is masked.
397 if (sync_test_bit(port, &s->evtchn_pending[0]) &&
398 !sync_test_and_set_bit(port / BITS_PER_LONG,
399 &vcpu_info->evtchn_pending_sel))
400 vcpu_info->evtchn_upcall_pending = 1;
406 static void xen_irq_init(unsigned irq)
408 struct irq_info *info;
409 struct irq_desc *desc = irq_to_desc(irq);
411 /* By default all event channels notify CPU#0. */
412 cpumask_copy(desc->irq_data.affinity, cpumask_of(0));
414 info = &irq_info[irq];
416 info->type = IRQT_UNBOUND;
418 list_add_tail(&info->list, &xen_irq_list_head);
421 static int xen_allocate_irq_dynamic(void)
426 #ifdef CONFIG_X86_IO_APIC
428 * For an HVM guest or domain 0 which see "real" (emulated or
429 * actual repectively) GSIs we allocate dynamic IRQs
430 * e.g. those corresponding to event channels or MSIs
431 * etc. from the range above those "real" GSIs to avoid
434 if (xen_initial_domain() || xen_hvm_domain())
435 first = get_nr_irqs_gsi();
439 irq = irq_alloc_desc_from(first, -1);
441 if (irq == -ENOMEM && first > NR_IRQS_LEGACY) {
442 printk(KERN_ERR "Out of dynamic IRQ space and eating into GSI space. You should increase nr_irqs\n");
443 first = max(NR_IRQS_LEGACY, first - NR_IRQS_LEGACY);
448 panic("No available IRQ to bind to: increase nr_irqs!\n");
455 static int xen_allocate_irq_gsi(unsigned gsi)
460 * A PV guest has no concept of a GSI (since it has no ACPI
461 * nor access to/knowledge of the physical APICs). Therefore
462 * all IRQs are dynamically allocated from the entire IRQ
465 if (xen_pv_domain() && !xen_initial_domain())
466 return xen_allocate_irq_dynamic();
468 /* Legacy IRQ descriptors are already allocated by the arch. */
469 if (gsi < NR_IRQS_LEGACY)
472 irq = irq_alloc_desc_at(gsi, -1);
475 panic("Unable to allocate to IRQ%d (%d)\n", gsi, irq);
482 static void xen_free_irq(unsigned irq)
484 struct irq_info *info = &irq_info[irq];
486 info->type = IRQT_UNBOUND;
488 list_del(&info->list);
490 /* Legacy IRQ descriptors are managed by the arch. */
491 if (irq < NR_IRQS_LEGACY)
497 static void pirq_unmask_notify(int irq)
499 struct physdev_eoi eoi = { .irq = pirq_from_irq(irq) };
501 if (unlikely(pirq_needs_eoi(irq))) {
502 int rc = HYPERVISOR_physdev_op(PHYSDEVOP_eoi, &eoi);
507 static void pirq_query_unmask(int irq)
509 struct physdev_irq_status_query irq_status;
510 struct irq_info *info = info_for_irq(irq);
512 BUG_ON(info->type != IRQT_PIRQ);
514 irq_status.irq = pirq_from_irq(irq);
515 if (HYPERVISOR_physdev_op(PHYSDEVOP_irq_status_query, &irq_status))
516 irq_status.flags = 0;
518 info->u.pirq.flags &= ~PIRQ_NEEDS_EOI;
519 if (irq_status.flags & XENIRQSTAT_needs_eoi)
520 info->u.pirq.flags |= PIRQ_NEEDS_EOI;
523 static bool probing_irq(int irq)
525 struct irq_desc *desc = irq_to_desc(irq);
527 return desc && desc->action == NULL;
530 static unsigned int __startup_pirq(unsigned int irq)
532 struct evtchn_bind_pirq bind_pirq;
533 struct irq_info *info = info_for_irq(irq);
534 int evtchn = evtchn_from_irq(irq);
537 BUG_ON(info->type != IRQT_PIRQ);
539 if (VALID_EVTCHN(evtchn))
542 bind_pirq.pirq = pirq_from_irq(irq);
543 /* NB. We are happy to share unless we are probing. */
544 bind_pirq.flags = info->u.pirq.flags & PIRQ_SHAREABLE ?
545 BIND_PIRQ__WILL_SHARE : 0;
546 rc = HYPERVISOR_event_channel_op(EVTCHNOP_bind_pirq, &bind_pirq);
548 if (!probing_irq(irq))
549 printk(KERN_INFO "Failed to obtain physical IRQ %d\n",
553 evtchn = bind_pirq.port;
555 pirq_query_unmask(irq);
557 evtchn_to_irq[evtchn] = irq;
558 bind_evtchn_to_cpu(evtchn, 0);
559 info->evtchn = evtchn;
562 unmask_evtchn(evtchn);
563 pirq_unmask_notify(irq);
568 static unsigned int startup_pirq(struct irq_data *data)
570 return __startup_pirq(data->irq);
573 static void shutdown_pirq(struct irq_data *data)
575 struct evtchn_close close;
576 unsigned int irq = data->irq;
577 struct irq_info *info = info_for_irq(irq);
578 int evtchn = evtchn_from_irq(irq);
580 BUG_ON(info->type != IRQT_PIRQ);
582 if (!VALID_EVTCHN(evtchn))
588 if (HYPERVISOR_event_channel_op(EVTCHNOP_close, &close) != 0)
591 bind_evtchn_to_cpu(evtchn, 0);
592 evtchn_to_irq[evtchn] = -1;
596 static void enable_pirq(struct irq_data *data)
601 static void disable_pirq(struct irq_data *data)
605 static void ack_pirq(struct irq_data *data)
607 int evtchn = evtchn_from_irq(data->irq);
609 move_native_irq(data->irq);
611 if (VALID_EVTCHN(evtchn)) {
613 clear_evtchn(evtchn);
617 static int find_irq_by_gsi(unsigned gsi)
619 struct irq_info *info;
621 list_for_each_entry(info, &xen_irq_list_head, list) {
622 if (info->type != IRQT_PIRQ)
625 if (info->u.pirq.gsi == gsi)
632 int xen_allocate_pirq_gsi(unsigned gsi)
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 int xen_bind_pirq_gsi_to_irq(unsigned gsi,
645 unsigned pirq, int shareable, char *name)
648 struct physdev_irq irq_op;
650 spin_lock(&irq_mapping_update_lock);
652 if ((pirq > nr_irqs) || (gsi > nr_irqs)) {
653 printk(KERN_WARNING "xen_map_pirq_gsi: %s %s is incorrect!\n",
654 pirq > nr_irqs ? "pirq" :"",
655 gsi > nr_irqs ? "gsi" : "");
659 irq = find_irq_by_gsi(gsi);
661 printk(KERN_INFO "xen_map_pirq_gsi: returning irq %d for gsi %u\n",
663 goto out; /* XXX need refcount? */
666 irq = xen_allocate_irq_gsi(gsi);
668 set_irq_chip_and_handler_name(irq, &xen_pirq_chip,
669 handle_level_irq, name);
674 /* Only the privileged domain can do this. For non-priv, the pcifront
675 * driver provides a PCI bus that does the call to do exactly
676 * this in the priv domain. */
677 if (xen_initial_domain() &&
678 HYPERVISOR_physdev_op(PHYSDEVOP_alloc_irq_vector, &irq_op)) {
684 xen_irq_info_pirq_init(irq, 0, pirq, gsi, irq_op.vector,
685 shareable ? PIRQ_SHAREABLE : 0);
688 spin_unlock(&irq_mapping_update_lock);
693 #ifdef CONFIG_PCI_MSI
694 int xen_allocate_pirq_msi(struct pci_dev *dev, struct msi_desc *msidesc)
697 struct physdev_get_free_pirq op_get_free_pirq;
699 op_get_free_pirq.type = MAP_PIRQ_TYPE_MSI;
700 rc = HYPERVISOR_physdev_op(PHYSDEVOP_get_free_pirq, &op_get_free_pirq);
702 WARN_ONCE(rc == -ENOSYS,
703 "hypervisor does not support the PHYSDEVOP_get_free_pirq interface\n");
705 return rc ? -1 : op_get_free_pirq.pirq;
708 int xen_bind_pirq_msi_to_irq(struct pci_dev *dev, struct msi_desc *msidesc,
709 int pirq, int vector, const char *name)
713 spin_lock(&irq_mapping_update_lock);
715 irq = xen_allocate_irq_dynamic();
719 set_irq_chip_and_handler_name(irq, &xen_pirq_chip,
720 handle_level_irq, name);
722 xen_irq_info_pirq_init(irq, 0, pirq, 0, vector, 0);
723 ret = set_irq_msi(irq, msidesc);
727 spin_unlock(&irq_mapping_update_lock);
730 spin_unlock(&irq_mapping_update_lock);
736 int xen_destroy_irq(int irq)
738 struct irq_desc *desc;
739 struct physdev_unmap_pirq unmap_irq;
740 struct irq_info *info = info_for_irq(irq);
743 spin_lock(&irq_mapping_update_lock);
745 desc = irq_to_desc(irq);
749 if (xen_initial_domain()) {
750 unmap_irq.pirq = info->u.pirq.pirq;
751 unmap_irq.domid = DOMID_SELF;
752 rc = HYPERVISOR_physdev_op(PHYSDEVOP_unmap_pirq, &unmap_irq);
754 printk(KERN_WARNING "unmap irq failed %d\n", rc);
758 pirq_to_irq[info->u.pirq.pirq] = -1;
763 spin_unlock(&irq_mapping_update_lock);
767 int xen_irq_from_pirq(unsigned pirq)
769 return pirq_to_irq[pirq];
772 int bind_evtchn_to_irq(unsigned int evtchn)
776 spin_lock(&irq_mapping_update_lock);
778 irq = evtchn_to_irq[evtchn];
781 irq = xen_allocate_irq_dynamic();
783 set_irq_chip_and_handler_name(irq, &xen_dynamic_chip,
784 handle_fasteoi_irq, "event");
786 xen_irq_info_evtchn_init(irq, evtchn);
789 spin_unlock(&irq_mapping_update_lock);
793 EXPORT_SYMBOL_GPL(bind_evtchn_to_irq);
795 static int bind_ipi_to_irq(unsigned int ipi, unsigned int cpu)
797 struct evtchn_bind_ipi bind_ipi;
800 spin_lock(&irq_mapping_update_lock);
802 irq = per_cpu(ipi_to_irq, cpu)[ipi];
805 irq = xen_allocate_irq_dynamic();
809 set_irq_chip_and_handler_name(irq, &xen_percpu_chip,
810 handle_percpu_irq, "ipi");
813 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_ipi,
816 evtchn = bind_ipi.port;
818 xen_irq_info_ipi_init(cpu, irq, evtchn, ipi);
820 bind_evtchn_to_cpu(evtchn, cpu);
824 spin_unlock(&irq_mapping_update_lock);
829 int bind_virq_to_irq(unsigned int virq, unsigned int cpu)
831 struct evtchn_bind_virq bind_virq;
834 spin_lock(&irq_mapping_update_lock);
836 irq = per_cpu(virq_to_irq, cpu)[virq];
839 irq = xen_allocate_irq_dynamic();
841 set_irq_chip_and_handler_name(irq, &xen_percpu_chip,
842 handle_percpu_irq, "virq");
844 bind_virq.virq = virq;
845 bind_virq.vcpu = cpu;
846 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_virq,
849 evtchn = bind_virq.port;
851 xen_irq_info_virq_init(cpu, irq, evtchn, virq);
853 bind_evtchn_to_cpu(evtchn, cpu);
856 spin_unlock(&irq_mapping_update_lock);
861 static void unbind_from_irq(unsigned int irq)
863 struct evtchn_close close;
864 int evtchn = evtchn_from_irq(irq);
866 spin_lock(&irq_mapping_update_lock);
868 if (VALID_EVTCHN(evtchn)) {
870 if (HYPERVISOR_event_channel_op(EVTCHNOP_close, &close) != 0)
873 switch (type_from_irq(irq)) {
875 per_cpu(virq_to_irq, cpu_from_evtchn(evtchn))
876 [virq_from_irq(irq)] = -1;
879 per_cpu(ipi_to_irq, cpu_from_evtchn(evtchn))
880 [ipi_from_irq(irq)] = -1;
886 /* Closed ports are implicitly re-bound to VCPU0. */
887 bind_evtchn_to_cpu(evtchn, 0);
889 evtchn_to_irq[evtchn] = -1;
892 BUG_ON(irq_info[irq].type == IRQT_UNBOUND);
896 spin_unlock(&irq_mapping_update_lock);
899 int bind_evtchn_to_irqhandler(unsigned int evtchn,
900 irq_handler_t handler,
901 unsigned long irqflags,
902 const char *devname, void *dev_id)
907 irq = bind_evtchn_to_irq(evtchn);
908 retval = request_irq(irq, handler, irqflags, devname, dev_id);
910 unbind_from_irq(irq);
916 EXPORT_SYMBOL_GPL(bind_evtchn_to_irqhandler);
918 int bind_virq_to_irqhandler(unsigned int virq, unsigned int cpu,
919 irq_handler_t handler,
920 unsigned long irqflags, const char *devname, void *dev_id)
925 irq = bind_virq_to_irq(virq, cpu);
926 retval = request_irq(irq, handler, irqflags, devname, dev_id);
928 unbind_from_irq(irq);
934 EXPORT_SYMBOL_GPL(bind_virq_to_irqhandler);
936 int bind_ipi_to_irqhandler(enum ipi_vector ipi,
938 irq_handler_t handler,
939 unsigned long irqflags,
945 irq = bind_ipi_to_irq(ipi, cpu);
949 irqflags |= IRQF_NO_SUSPEND | IRQF_FORCE_RESUME;
950 retval = request_irq(irq, handler, irqflags, devname, dev_id);
952 unbind_from_irq(irq);
959 void unbind_from_irqhandler(unsigned int irq, void *dev_id)
961 free_irq(irq, dev_id);
962 unbind_from_irq(irq);
964 EXPORT_SYMBOL_GPL(unbind_from_irqhandler);
966 void xen_send_IPI_one(unsigned int cpu, enum ipi_vector vector)
968 int irq = per_cpu(ipi_to_irq, cpu)[vector];
970 notify_remote_via_irq(irq);
973 irqreturn_t xen_debug_interrupt(int irq, void *dev_id)
975 struct shared_info *sh = HYPERVISOR_shared_info;
976 int cpu = smp_processor_id();
977 unsigned long *cpu_evtchn = per_cpu(cpu_evtchn_mask, cpu);
980 static DEFINE_SPINLOCK(debug_lock);
983 spin_lock_irqsave(&debug_lock, flags);
985 printk("\nvcpu %d\n ", cpu);
987 for_each_online_cpu(i) {
989 v = per_cpu(xen_vcpu, i);
990 pending = (get_irq_regs() && i == cpu)
991 ? xen_irqs_disabled(get_irq_regs())
992 : v->evtchn_upcall_mask;
993 printk("%d: masked=%d pending=%d event_sel %0*lx\n ", i,
994 pending, v->evtchn_upcall_pending,
995 (int)(sizeof(v->evtchn_pending_sel)*2),
996 v->evtchn_pending_sel);
998 v = per_cpu(xen_vcpu, cpu);
1000 printk("\npending:\n ");
1001 for (i = ARRAY_SIZE(sh->evtchn_pending)-1; i >= 0; i--)
1002 printk("%0*lx%s", (int)sizeof(sh->evtchn_pending[0])*2,
1003 sh->evtchn_pending[i],
1004 i % 8 == 0 ? "\n " : " ");
1005 printk("\nglobal mask:\n ");
1006 for (i = ARRAY_SIZE(sh->evtchn_mask)-1; i >= 0; i--)
1008 (int)(sizeof(sh->evtchn_mask[0])*2),
1010 i % 8 == 0 ? "\n " : " ");
1012 printk("\nglobally unmasked:\n ");
1013 for (i = ARRAY_SIZE(sh->evtchn_mask)-1; i >= 0; i--)
1014 printk("%0*lx%s", (int)(sizeof(sh->evtchn_mask[0])*2),
1015 sh->evtchn_pending[i] & ~sh->evtchn_mask[i],
1016 i % 8 == 0 ? "\n " : " ");
1018 printk("\nlocal cpu%d mask:\n ", cpu);
1019 for (i = (NR_EVENT_CHANNELS/BITS_PER_LONG)-1; i >= 0; i--)
1020 printk("%0*lx%s", (int)(sizeof(cpu_evtchn[0])*2),
1022 i % 8 == 0 ? "\n " : " ");
1024 printk("\nlocally unmasked:\n ");
1025 for (i = ARRAY_SIZE(sh->evtchn_mask)-1; i >= 0; i--) {
1026 unsigned long pending = sh->evtchn_pending[i]
1027 & ~sh->evtchn_mask[i]
1029 printk("%0*lx%s", (int)(sizeof(sh->evtchn_mask[0])*2),
1030 pending, i % 8 == 0 ? "\n " : " ");
1033 printk("\npending list:\n");
1034 for (i = 0; i < NR_EVENT_CHANNELS; i++) {
1035 if (sync_test_bit(i, sh->evtchn_pending)) {
1036 int word_idx = i / BITS_PER_LONG;
1037 printk(" %d: event %d -> irq %d%s%s%s\n",
1038 cpu_from_evtchn(i), i,
1040 sync_test_bit(word_idx, &v->evtchn_pending_sel)
1042 !sync_test_bit(i, sh->evtchn_mask)
1043 ? "" : " globally-masked",
1044 sync_test_bit(i, cpu_evtchn)
1045 ? "" : " locally-masked");
1049 spin_unlock_irqrestore(&debug_lock, flags);
1054 static DEFINE_PER_CPU(unsigned, xed_nesting_count);
1057 * Search the CPUs pending events bitmasks. For each one found, map
1058 * the event number to an irq, and feed it into do_IRQ() for
1061 * Xen uses a two-level bitmap to speed searching. The first level is
1062 * a bitset of words which contain pending event bits. The second
1063 * level is a bitset of pending events themselves.
1065 static void __xen_evtchn_do_upcall(void)
1067 int cpu = get_cpu();
1068 struct shared_info *s = HYPERVISOR_shared_info;
1069 struct vcpu_info *vcpu_info = __this_cpu_read(xen_vcpu);
1073 unsigned long pending_words;
1075 vcpu_info->evtchn_upcall_pending = 0;
1077 if (__this_cpu_inc_return(xed_nesting_count) - 1)
1080 #ifndef CONFIG_X86 /* No need for a barrier -- XCHG is a barrier on x86. */
1081 /* Clear master flag /before/ clearing selector flag. */
1084 pending_words = xchg(&vcpu_info->evtchn_pending_sel, 0);
1085 while (pending_words != 0) {
1086 unsigned long pending_bits;
1087 int word_idx = __ffs(pending_words);
1088 pending_words &= ~(1UL << word_idx);
1090 while ((pending_bits = active_evtchns(cpu, s, word_idx)) != 0) {
1091 int bit_idx = __ffs(pending_bits);
1092 int port = (word_idx * BITS_PER_LONG) + bit_idx;
1093 int irq = evtchn_to_irq[port];
1094 struct irq_desc *desc;
1100 desc = irq_to_desc(irq);
1102 generic_handle_irq_desc(irq, desc);
1107 BUG_ON(!irqs_disabled());
1109 count = __this_cpu_read(xed_nesting_count);
1110 __this_cpu_write(xed_nesting_count, 0);
1111 } while (count != 1 || vcpu_info->evtchn_upcall_pending);
1118 void xen_evtchn_do_upcall(struct pt_regs *regs)
1120 struct pt_regs *old_regs = set_irq_regs(regs);
1125 __xen_evtchn_do_upcall();
1128 set_irq_regs(old_regs);
1131 void xen_hvm_evtchn_do_upcall(void)
1133 __xen_evtchn_do_upcall();
1135 EXPORT_SYMBOL_GPL(xen_hvm_evtchn_do_upcall);
1137 /* Rebind a new event channel to an existing irq. */
1138 void rebind_evtchn_irq(int evtchn, int irq)
1140 struct irq_info *info = info_for_irq(irq);
1142 /* Make sure the irq is masked, since the new event channel
1143 will also be masked. */
1146 spin_lock(&irq_mapping_update_lock);
1148 /* After resume the irq<->evtchn mappings are all cleared out */
1149 BUG_ON(evtchn_to_irq[evtchn] != -1);
1150 /* Expect irq to have been bound before,
1151 so there should be a proper type */
1152 BUG_ON(info->type == IRQT_UNBOUND);
1154 xen_irq_info_evtchn_init(irq, evtchn);
1156 spin_unlock(&irq_mapping_update_lock);
1158 /* new event channels are always bound to cpu 0 */
1159 irq_set_affinity(irq, cpumask_of(0));
1161 /* Unmask the event channel. */
1165 /* Rebind an evtchn so that it gets delivered to a specific cpu */
1166 static int rebind_irq_to_cpu(unsigned irq, unsigned tcpu)
1168 struct evtchn_bind_vcpu bind_vcpu;
1169 int evtchn = evtchn_from_irq(irq);
1171 if (!VALID_EVTCHN(evtchn))
1175 * Events delivered via platform PCI interrupts are always
1176 * routed to vcpu 0 and hence cannot be rebound.
1178 if (xen_hvm_domain() && !xen_have_vector_callback)
1181 /* Send future instances of this interrupt to other vcpu. */
1182 bind_vcpu.port = evtchn;
1183 bind_vcpu.vcpu = tcpu;
1186 * If this fails, it usually just indicates that we're dealing with a
1187 * virq or IPI channel, which don't actually need to be rebound. Ignore
1188 * it, but don't do the xenlinux-level rebind in that case.
1190 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_vcpu, &bind_vcpu) >= 0)
1191 bind_evtchn_to_cpu(evtchn, tcpu);
1196 static int set_affinity_irq(struct irq_data *data, const struct cpumask *dest,
1199 unsigned tcpu = cpumask_first(dest);
1201 return rebind_irq_to_cpu(data->irq, tcpu);
1204 int resend_irq_on_evtchn(unsigned int irq)
1206 int masked, evtchn = evtchn_from_irq(irq);
1207 struct shared_info *s = HYPERVISOR_shared_info;
1209 if (!VALID_EVTCHN(evtchn))
1212 masked = sync_test_and_set_bit(evtchn, s->evtchn_mask);
1213 sync_set_bit(evtchn, s->evtchn_pending);
1215 unmask_evtchn(evtchn);
1220 static void enable_dynirq(struct irq_data *data)
1222 int evtchn = evtchn_from_irq(data->irq);
1224 if (VALID_EVTCHN(evtchn))
1225 unmask_evtchn(evtchn);
1228 static void disable_dynirq(struct irq_data *data)
1230 int evtchn = evtchn_from_irq(data->irq);
1232 if (VALID_EVTCHN(evtchn))
1233 mask_evtchn(evtchn);
1236 static void ack_dynirq(struct irq_data *data)
1238 int evtchn = evtchn_from_irq(data->irq);
1240 move_masked_irq(data->irq);
1242 if (VALID_EVTCHN(evtchn))
1243 unmask_evtchn(evtchn);
1246 static int retrigger_dynirq(struct irq_data *data)
1248 int evtchn = evtchn_from_irq(data->irq);
1249 struct shared_info *sh = HYPERVISOR_shared_info;
1252 if (VALID_EVTCHN(evtchn)) {
1255 masked = sync_test_and_set_bit(evtchn, sh->evtchn_mask);
1256 sync_set_bit(evtchn, sh->evtchn_pending);
1258 unmask_evtchn(evtchn);
1265 static void restore_pirqs(void)
1267 int pirq, rc, irq, gsi;
1268 struct physdev_map_pirq map_irq;
1270 for (pirq = 0; pirq < nr_irqs; pirq++) {
1271 irq = pirq_to_irq[pirq];
1275 /* save/restore of PT devices doesn't work, so at this point the
1276 * only devices present are GSI based emulated devices */
1277 gsi = gsi_from_irq(irq);
1281 map_irq.domid = DOMID_SELF;
1282 map_irq.type = MAP_PIRQ_TYPE_GSI;
1283 map_irq.index = gsi;
1284 map_irq.pirq = pirq;
1286 rc = HYPERVISOR_physdev_op(PHYSDEVOP_map_pirq, &map_irq);
1288 printk(KERN_WARNING "xen map irq failed gsi=%d irq=%d pirq=%d rc=%d\n",
1289 gsi, irq, pirq, rc);
1291 pirq_to_irq[pirq] = -1;
1295 printk(KERN_DEBUG "xen: --> irq=%d, pirq=%d\n", irq, map_irq.pirq);
1297 __startup_pirq(irq);
1301 static void restore_cpu_virqs(unsigned int cpu)
1303 struct evtchn_bind_virq bind_virq;
1304 int virq, irq, evtchn;
1306 for (virq = 0; virq < NR_VIRQS; virq++) {
1307 if ((irq = per_cpu(virq_to_irq, cpu)[virq]) == -1)
1310 BUG_ON(virq_from_irq(irq) != virq);
1312 /* Get a new binding from Xen. */
1313 bind_virq.virq = virq;
1314 bind_virq.vcpu = cpu;
1315 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_virq,
1318 evtchn = bind_virq.port;
1320 /* Record the new mapping. */
1321 xen_irq_info_virq_init(cpu, irq, evtchn, virq);
1322 bind_evtchn_to_cpu(evtchn, cpu);
1326 static void restore_cpu_ipis(unsigned int cpu)
1328 struct evtchn_bind_ipi bind_ipi;
1329 int ipi, irq, evtchn;
1331 for (ipi = 0; ipi < XEN_NR_IPIS; ipi++) {
1332 if ((irq = per_cpu(ipi_to_irq, cpu)[ipi]) == -1)
1335 BUG_ON(ipi_from_irq(irq) != ipi);
1337 /* Get a new binding from Xen. */
1338 bind_ipi.vcpu = cpu;
1339 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_ipi,
1342 evtchn = bind_ipi.port;
1344 /* Record the new mapping. */
1345 xen_irq_info_ipi_init(cpu, irq, evtchn, ipi);
1346 bind_evtchn_to_cpu(evtchn, cpu);
1350 /* Clear an irq's pending state, in preparation for polling on it */
1351 void xen_clear_irq_pending(int irq)
1353 int evtchn = evtchn_from_irq(irq);
1355 if (VALID_EVTCHN(evtchn))
1356 clear_evtchn(evtchn);
1358 EXPORT_SYMBOL(xen_clear_irq_pending);
1359 void xen_set_irq_pending(int irq)
1361 int evtchn = evtchn_from_irq(irq);
1363 if (VALID_EVTCHN(evtchn))
1367 bool xen_test_irq_pending(int irq)
1369 int evtchn = evtchn_from_irq(irq);
1372 if (VALID_EVTCHN(evtchn))
1373 ret = test_evtchn(evtchn);
1378 /* Poll waiting for an irq to become pending with timeout. In the usual case,
1379 * the irq will be disabled so it won't deliver an interrupt. */
1380 void xen_poll_irq_timeout(int irq, u64 timeout)
1382 evtchn_port_t evtchn = evtchn_from_irq(irq);
1384 if (VALID_EVTCHN(evtchn)) {
1385 struct sched_poll poll;
1388 poll.timeout = timeout;
1389 set_xen_guest_handle(poll.ports, &evtchn);
1391 if (HYPERVISOR_sched_op(SCHEDOP_poll, &poll) != 0)
1395 EXPORT_SYMBOL(xen_poll_irq_timeout);
1396 /* Poll waiting for an irq to become pending. In the usual case, the
1397 * irq will be disabled so it won't deliver an interrupt. */
1398 void xen_poll_irq(int irq)
1400 xen_poll_irq_timeout(irq, 0 /* no timeout */);
1403 void xen_irq_resume(void)
1405 unsigned int cpu, evtchn;
1406 struct irq_info *info;
1408 init_evtchn_cpu_bindings();
1410 /* New event-channel space is not 'live' yet. */
1411 for (evtchn = 0; evtchn < NR_EVENT_CHANNELS; evtchn++)
1412 mask_evtchn(evtchn);
1414 /* No IRQ <-> event-channel mappings. */
1415 list_for_each_entry(info, &xen_irq_list_head, list)
1416 info->evtchn = 0; /* zap event-channel binding */
1418 for (evtchn = 0; evtchn < NR_EVENT_CHANNELS; evtchn++)
1419 evtchn_to_irq[evtchn] = -1;
1421 for_each_possible_cpu(cpu) {
1422 restore_cpu_virqs(cpu);
1423 restore_cpu_ipis(cpu);
1429 static struct irq_chip xen_dynamic_chip __read_mostly = {
1432 .irq_disable = disable_dynirq,
1433 .irq_mask = disable_dynirq,
1434 .irq_unmask = enable_dynirq,
1436 .irq_eoi = ack_dynirq,
1437 .irq_set_affinity = set_affinity_irq,
1438 .irq_retrigger = retrigger_dynirq,
1441 static struct irq_chip xen_pirq_chip __read_mostly = {
1444 .irq_startup = startup_pirq,
1445 .irq_shutdown = shutdown_pirq,
1447 .irq_enable = enable_pirq,
1448 .irq_unmask = enable_pirq,
1450 .irq_disable = disable_pirq,
1451 .irq_mask = disable_pirq,
1453 .irq_ack = ack_pirq,
1455 .irq_set_affinity = set_affinity_irq,
1457 .irq_retrigger = retrigger_dynirq,
1460 static struct irq_chip xen_percpu_chip __read_mostly = {
1461 .name = "xen-percpu",
1463 .irq_disable = disable_dynirq,
1464 .irq_mask = disable_dynirq,
1465 .irq_unmask = enable_dynirq,
1467 .irq_ack = ack_dynirq,
1470 int xen_set_callback_via(uint64_t via)
1472 struct xen_hvm_param a;
1473 a.domid = DOMID_SELF;
1474 a.index = HVM_PARAM_CALLBACK_IRQ;
1476 return HYPERVISOR_hvm_op(HVMOP_set_param, &a);
1478 EXPORT_SYMBOL_GPL(xen_set_callback_via);
1480 #ifdef CONFIG_XEN_PVHVM
1481 /* Vector callbacks are better than PCI interrupts to receive event
1482 * channel notifications because we can receive vector callbacks on any
1483 * vcpu and we don't need PCI support or APIC interactions. */
1484 void xen_callback_vector(void)
1487 uint64_t callback_via;
1488 if (xen_have_vector_callback) {
1489 callback_via = HVM_CALLBACK_VECTOR(XEN_HVM_EVTCHN_CALLBACK);
1490 rc = xen_set_callback_via(callback_via);
1492 printk(KERN_ERR "Request for Xen HVM callback vector"
1494 xen_have_vector_callback = 0;
1497 printk(KERN_INFO "Xen HVM callback vector for event delivery is "
1499 /* in the restore case the vector has already been allocated */
1500 if (!test_bit(XEN_HVM_EVTCHN_CALLBACK, used_vectors))
1501 alloc_intr_gate(XEN_HVM_EVTCHN_CALLBACK, xen_hvm_callback_vector);
1505 void xen_callback_vector(void) {}
1508 void __init xen_init_IRQ(void)
1512 irq_info = kcalloc(nr_irqs, sizeof(*irq_info), GFP_KERNEL);
1514 /* We are using nr_irqs as the maximum number of pirq available but
1515 * that number is actually chosen by Xen and we don't know exactly
1516 * what it is. Be careful choosing high pirq numbers. */
1517 pirq_to_irq = kcalloc(nr_irqs, sizeof(*pirq_to_irq), GFP_KERNEL);
1518 for (i = 0; i < nr_irqs; i++)
1519 pirq_to_irq[i] = -1;
1521 evtchn_to_irq = kcalloc(NR_EVENT_CHANNELS, sizeof(*evtchn_to_irq),
1523 for (i = 0; i < NR_EVENT_CHANNELS; i++)
1524 evtchn_to_irq[i] = -1;
1526 init_evtchn_cpu_bindings();
1528 /* No event channels are 'live' right now. */
1529 for (i = 0; i < NR_EVENT_CHANNELS; i++)
1532 if (xen_hvm_domain()) {
1533 xen_callback_vector();
1535 /* pci_xen_hvm_init must be called after native_init_IRQ so that
1536 * __acpi_register_gsi can point at the right function */
1539 irq_ctx_init(smp_processor_id());
1540 if (xen_initial_domain())