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 int *evtchn_to_irq;
112 static DEFINE_PER_CPU(unsigned long [NR_EVENT_CHANNELS/BITS_PER_LONG],
115 /* Xen will never allocate port zero for any purpose. */
116 #define VALID_EVTCHN(chn) ((chn) != 0)
118 static struct irq_chip xen_dynamic_chip;
119 static struct irq_chip xen_percpu_chip;
120 static struct irq_chip xen_pirq_chip;
122 /* Get info for IRQ */
123 static struct irq_info *info_for_irq(unsigned irq)
125 return get_irq_data(irq);
128 /* Constructors for packed IRQ information. */
129 static void xen_irq_info_common_init(struct irq_info *info,
131 enum xen_irq_type type,
132 unsigned short evtchn,
136 BUG_ON(info->type != IRQT_UNBOUND && info->type != type);
140 info->evtchn = evtchn;
143 evtchn_to_irq[evtchn] = irq;
146 static void xen_irq_info_evtchn_init(unsigned irq,
147 unsigned short evtchn)
149 struct irq_info *info = info_for_irq(irq);
151 xen_irq_info_common_init(info, irq, IRQT_EVTCHN, evtchn, 0);
154 static void xen_irq_info_ipi_init(unsigned cpu,
156 unsigned short evtchn,
159 struct irq_info *info = info_for_irq(irq);
161 xen_irq_info_common_init(info, irq, IRQT_IPI, evtchn, 0);
165 per_cpu(ipi_to_irq, cpu)[ipi] = irq;
168 static void xen_irq_info_virq_init(unsigned cpu,
170 unsigned short evtchn,
173 struct irq_info *info = info_for_irq(irq);
175 xen_irq_info_common_init(info, irq, IRQT_VIRQ, evtchn, 0);
179 per_cpu(virq_to_irq, cpu)[virq] = irq;
182 static void xen_irq_info_pirq_init(unsigned irq,
183 unsigned short evtchn,
186 unsigned short vector,
189 struct irq_info *info = info_for_irq(irq);
191 xen_irq_info_common_init(info, irq, IRQT_PIRQ, evtchn, 0);
193 info->u.pirq.pirq = pirq;
194 info->u.pirq.gsi = gsi;
195 info->u.pirq.vector = vector;
196 info->u.pirq.flags = flags;
200 * Accessors for packed IRQ information.
202 static unsigned int evtchn_from_irq(unsigned irq)
204 if (unlikely(WARN(irq < 0 || irq >= nr_irqs, "Invalid irq %d!\n", irq)))
207 return info_for_irq(irq)->evtchn;
210 unsigned irq_from_evtchn(unsigned int evtchn)
212 return evtchn_to_irq[evtchn];
214 EXPORT_SYMBOL_GPL(irq_from_evtchn);
216 static enum ipi_vector ipi_from_irq(unsigned irq)
218 struct irq_info *info = info_for_irq(irq);
220 BUG_ON(info == NULL);
221 BUG_ON(info->type != IRQT_IPI);
226 static unsigned virq_from_irq(unsigned irq)
228 struct irq_info *info = info_for_irq(irq);
230 BUG_ON(info == NULL);
231 BUG_ON(info->type != IRQT_VIRQ);
236 static unsigned pirq_from_irq(unsigned irq)
238 struct irq_info *info = info_for_irq(irq);
240 BUG_ON(info == NULL);
241 BUG_ON(info->type != IRQT_PIRQ);
243 return info->u.pirq.pirq;
246 static enum xen_irq_type type_from_irq(unsigned irq)
248 return info_for_irq(irq)->type;
251 static unsigned cpu_from_irq(unsigned irq)
253 return info_for_irq(irq)->cpu;
256 static unsigned int cpu_from_evtchn(unsigned int evtchn)
258 int irq = evtchn_to_irq[evtchn];
262 ret = cpu_from_irq(irq);
267 static bool pirq_needs_eoi(unsigned irq)
269 struct irq_info *info = info_for_irq(irq);
271 BUG_ON(info->type != IRQT_PIRQ);
273 return info->u.pirq.flags & PIRQ_NEEDS_EOI;
276 static inline unsigned long active_evtchns(unsigned int cpu,
277 struct shared_info *sh,
280 return (sh->evtchn_pending[idx] &
281 per_cpu(cpu_evtchn_mask, cpu)[idx] &
282 ~sh->evtchn_mask[idx]);
285 static void bind_evtchn_to_cpu(unsigned int chn, unsigned int cpu)
287 int irq = evtchn_to_irq[chn];
291 cpumask_copy(irq_to_desc(irq)->irq_data.affinity, cpumask_of(cpu));
294 clear_bit(chn, per_cpu(cpu_evtchn_mask, cpu_from_irq(irq)));
295 set_bit(chn, per_cpu(cpu_evtchn_mask, cpu));
297 info_for_irq(irq)->cpu = cpu;
300 static void init_evtchn_cpu_bindings(void)
304 struct irq_info *info;
306 /* By default all event channels notify CPU#0. */
307 list_for_each_entry(info, &xen_irq_list_head, list) {
308 struct irq_desc *desc = irq_to_desc(info->irq);
309 cpumask_copy(desc->irq_data.affinity, cpumask_of(0));
313 for_each_possible_cpu(i)
314 memset(per_cpu(cpu_evtchn_mask, i),
315 (i == 0) ? ~0 : 0, sizeof(*per_cpu(cpu_evtchn_mask, i)));
318 static inline void clear_evtchn(int port)
320 struct shared_info *s = HYPERVISOR_shared_info;
321 sync_clear_bit(port, &s->evtchn_pending[0]);
324 static inline void set_evtchn(int port)
326 struct shared_info *s = HYPERVISOR_shared_info;
327 sync_set_bit(port, &s->evtchn_pending[0]);
330 static inline int test_evtchn(int port)
332 struct shared_info *s = HYPERVISOR_shared_info;
333 return sync_test_bit(port, &s->evtchn_pending[0]);
338 * notify_remote_via_irq - send event to remote end of event channel via irq
339 * @irq: irq of event channel to send event to
341 * Unlike notify_remote_via_evtchn(), this is safe to use across
342 * save/restore. Notifications on a broken connection are silently
345 void notify_remote_via_irq(int irq)
347 int evtchn = evtchn_from_irq(irq);
349 if (VALID_EVTCHN(evtchn))
350 notify_remote_via_evtchn(evtchn);
352 EXPORT_SYMBOL_GPL(notify_remote_via_irq);
354 static void mask_evtchn(int port)
356 struct shared_info *s = HYPERVISOR_shared_info;
357 sync_set_bit(port, &s->evtchn_mask[0]);
360 static void unmask_evtchn(int port)
362 struct shared_info *s = HYPERVISOR_shared_info;
363 unsigned int cpu = get_cpu();
365 BUG_ON(!irqs_disabled());
367 /* Slow path (hypercall) if this is a non-local port. */
368 if (unlikely(cpu != cpu_from_evtchn(port))) {
369 struct evtchn_unmask unmask = { .port = port };
370 (void)HYPERVISOR_event_channel_op(EVTCHNOP_unmask, &unmask);
372 struct vcpu_info *vcpu_info = __this_cpu_read(xen_vcpu);
374 sync_clear_bit(port, &s->evtchn_mask[0]);
377 * The following is basically the equivalent of
378 * 'hw_resend_irq'. Just like a real IO-APIC we 'lose
379 * the interrupt edge' if the channel is masked.
381 if (sync_test_bit(port, &s->evtchn_pending[0]) &&
382 !sync_test_and_set_bit(port / BITS_PER_LONG,
383 &vcpu_info->evtchn_pending_sel))
384 vcpu_info->evtchn_upcall_pending = 1;
390 static void xen_irq_init(unsigned irq)
392 struct irq_info *info;
393 struct irq_desc *desc = irq_to_desc(irq);
396 /* By default all event channels notify CPU#0. */
397 cpumask_copy(desc->irq_data.affinity, cpumask_of(0));
400 info = kzalloc(sizeof(*info), GFP_KERNEL);
402 panic("Unable to allocate metadata for IRQ%d\n", irq);
404 info->type = IRQT_UNBOUND;
406 set_irq_data(irq, info);
408 list_add_tail(&info->list, &xen_irq_list_head);
411 static int __must_check xen_allocate_irq_dynamic(void)
416 #ifdef CONFIG_X86_IO_APIC
418 * For an HVM guest or domain 0 which see "real" (emulated or
419 * actual repectively) GSIs we allocate dynamic IRQs
420 * e.g. those corresponding to event channels or MSIs
421 * etc. from the range above those "real" GSIs to avoid
424 if (xen_initial_domain() || xen_hvm_domain())
425 first = get_nr_irqs_gsi();
428 irq = irq_alloc_desc_from(first, -1);
435 static int __must_check xen_allocate_irq_gsi(unsigned gsi)
440 * A PV guest has no concept of a GSI (since it has no ACPI
441 * nor access to/knowledge of the physical APICs). Therefore
442 * all IRQs are dynamically allocated from the entire IRQ
445 if (xen_pv_domain() && !xen_initial_domain())
446 return xen_allocate_irq_dynamic();
448 /* Legacy IRQ descriptors are already allocated by the arch. */
449 if (gsi < NR_IRQS_LEGACY)
452 irq = irq_alloc_desc_at(gsi, -1);
459 static void xen_free_irq(unsigned irq)
461 struct irq_info *info = get_irq_data(irq);
463 list_del(&info->list);
465 set_irq_data(irq, NULL);
469 /* Legacy IRQ descriptors are managed by the arch. */
470 if (irq < NR_IRQS_LEGACY)
476 static void pirq_unmask_notify(int irq)
478 struct physdev_eoi eoi = { .irq = pirq_from_irq(irq) };
480 if (unlikely(pirq_needs_eoi(irq))) {
481 int rc = HYPERVISOR_physdev_op(PHYSDEVOP_eoi, &eoi);
486 static void pirq_query_unmask(int irq)
488 struct physdev_irq_status_query irq_status;
489 struct irq_info *info = info_for_irq(irq);
491 BUG_ON(info->type != IRQT_PIRQ);
493 irq_status.irq = pirq_from_irq(irq);
494 if (HYPERVISOR_physdev_op(PHYSDEVOP_irq_status_query, &irq_status))
495 irq_status.flags = 0;
497 info->u.pirq.flags &= ~PIRQ_NEEDS_EOI;
498 if (irq_status.flags & XENIRQSTAT_needs_eoi)
499 info->u.pirq.flags |= PIRQ_NEEDS_EOI;
502 static bool probing_irq(int irq)
504 struct irq_desc *desc = irq_to_desc(irq);
506 return desc && desc->action == NULL;
509 static unsigned int __startup_pirq(unsigned int irq)
511 struct evtchn_bind_pirq bind_pirq;
512 struct irq_info *info = info_for_irq(irq);
513 int evtchn = evtchn_from_irq(irq);
516 BUG_ON(info->type != IRQT_PIRQ);
518 if (VALID_EVTCHN(evtchn))
521 bind_pirq.pirq = pirq_from_irq(irq);
522 /* NB. We are happy to share unless we are probing. */
523 bind_pirq.flags = info->u.pirq.flags & PIRQ_SHAREABLE ?
524 BIND_PIRQ__WILL_SHARE : 0;
525 rc = HYPERVISOR_event_channel_op(EVTCHNOP_bind_pirq, &bind_pirq);
527 if (!probing_irq(irq))
528 printk(KERN_INFO "Failed to obtain physical IRQ %d\n",
532 evtchn = bind_pirq.port;
534 pirq_query_unmask(irq);
536 evtchn_to_irq[evtchn] = irq;
537 bind_evtchn_to_cpu(evtchn, 0);
538 info->evtchn = evtchn;
541 unmask_evtchn(evtchn);
542 pirq_unmask_notify(irq);
547 static unsigned int startup_pirq(struct irq_data *data)
549 return __startup_pirq(data->irq);
552 static void shutdown_pirq(struct irq_data *data)
554 struct evtchn_close close;
555 unsigned int irq = data->irq;
556 struct irq_info *info = info_for_irq(irq);
557 int evtchn = evtchn_from_irq(irq);
559 BUG_ON(info->type != IRQT_PIRQ);
561 if (!VALID_EVTCHN(evtchn))
567 if (HYPERVISOR_event_channel_op(EVTCHNOP_close, &close) != 0)
570 bind_evtchn_to_cpu(evtchn, 0);
571 evtchn_to_irq[evtchn] = -1;
575 static void enable_pirq(struct irq_data *data)
580 static void disable_pirq(struct irq_data *data)
584 static void ack_pirq(struct irq_data *data)
586 int evtchn = evtchn_from_irq(data->irq);
588 move_native_irq(data->irq);
590 if (VALID_EVTCHN(evtchn)) {
592 clear_evtchn(evtchn);
596 static int find_irq_by_gsi(unsigned gsi)
598 struct irq_info *info;
600 list_for_each_entry(info, &xen_irq_list_head, list) {
601 if (info->type != IRQT_PIRQ)
604 if (info->u.pirq.gsi == gsi)
611 int xen_allocate_pirq_gsi(unsigned gsi)
617 * Do not make any assumptions regarding the relationship between the
618 * IRQ number returned here and the Xen pirq argument.
620 * Note: We don't assign an event channel until the irq actually started
621 * up. Return an existing irq if we've already got one for the gsi.
623 int xen_bind_pirq_gsi_to_irq(unsigned gsi,
624 unsigned pirq, int shareable, char *name)
627 struct physdev_irq irq_op;
629 spin_lock(&irq_mapping_update_lock);
631 irq = find_irq_by_gsi(gsi);
633 printk(KERN_INFO "xen_map_pirq_gsi: returning irq %d for gsi %u\n",
635 goto out; /* XXX need refcount? */
638 irq = xen_allocate_irq_gsi(gsi);
642 set_irq_chip_and_handler_name(irq, &xen_pirq_chip,
643 handle_level_irq, name);
648 /* Only the privileged domain can do this. For non-priv, the pcifront
649 * driver provides a PCI bus that does the call to do exactly
650 * this in the priv domain. */
651 if (xen_initial_domain() &&
652 HYPERVISOR_physdev_op(PHYSDEVOP_alloc_irq_vector, &irq_op)) {
658 xen_irq_info_pirq_init(irq, 0, pirq, gsi, irq_op.vector,
659 shareable ? PIRQ_SHAREABLE : 0);
662 spin_unlock(&irq_mapping_update_lock);
667 #ifdef CONFIG_PCI_MSI
668 int xen_allocate_pirq_msi(struct pci_dev *dev, struct msi_desc *msidesc)
671 struct physdev_get_free_pirq op_get_free_pirq;
673 op_get_free_pirq.type = MAP_PIRQ_TYPE_MSI;
674 rc = HYPERVISOR_physdev_op(PHYSDEVOP_get_free_pirq, &op_get_free_pirq);
676 WARN_ONCE(rc == -ENOSYS,
677 "hypervisor does not support the PHYSDEVOP_get_free_pirq interface\n");
679 return rc ? -1 : op_get_free_pirq.pirq;
682 int xen_bind_pirq_msi_to_irq(struct pci_dev *dev, struct msi_desc *msidesc,
683 int pirq, int vector, const char *name)
687 spin_lock(&irq_mapping_update_lock);
689 irq = xen_allocate_irq_dynamic();
693 set_irq_chip_and_handler_name(irq, &xen_pirq_chip,
694 handle_level_irq, name);
696 xen_irq_info_pirq_init(irq, 0, pirq, 0, vector, 0);
697 ret = set_irq_msi(irq, msidesc);
701 spin_unlock(&irq_mapping_update_lock);
704 spin_unlock(&irq_mapping_update_lock);
710 int xen_destroy_irq(int irq)
712 struct irq_desc *desc;
713 struct physdev_unmap_pirq unmap_irq;
714 struct irq_info *info = info_for_irq(irq);
717 spin_lock(&irq_mapping_update_lock);
719 desc = irq_to_desc(irq);
723 if (xen_initial_domain()) {
724 unmap_irq.pirq = info->u.pirq.pirq;
725 unmap_irq.domid = DOMID_SELF;
726 rc = HYPERVISOR_physdev_op(PHYSDEVOP_unmap_pirq, &unmap_irq);
728 printk(KERN_WARNING "unmap irq failed %d\n", rc);
736 spin_unlock(&irq_mapping_update_lock);
740 int xen_irq_from_pirq(unsigned pirq)
744 struct irq_info *info;
746 spin_lock(&irq_mapping_update_lock);
748 list_for_each_entry(info, &xen_irq_list_head, list) {
749 if (info == NULL || info->type != IRQT_PIRQ)
752 if (info->u.pirq.pirq == pirq)
757 spin_unlock(&irq_mapping_update_lock);
762 int bind_evtchn_to_irq(unsigned int evtchn)
766 spin_lock(&irq_mapping_update_lock);
768 irq = evtchn_to_irq[evtchn];
771 irq = xen_allocate_irq_dynamic();
775 set_irq_chip_and_handler_name(irq, &xen_dynamic_chip,
776 handle_fasteoi_irq, "event");
778 xen_irq_info_evtchn_init(irq, evtchn);
782 spin_unlock(&irq_mapping_update_lock);
786 EXPORT_SYMBOL_GPL(bind_evtchn_to_irq);
788 static int bind_ipi_to_irq(unsigned int ipi, unsigned int cpu)
790 struct evtchn_bind_ipi bind_ipi;
793 spin_lock(&irq_mapping_update_lock);
795 irq = per_cpu(ipi_to_irq, cpu)[ipi];
798 irq = xen_allocate_irq_dynamic();
802 set_irq_chip_and_handler_name(irq, &xen_percpu_chip,
803 handle_percpu_irq, "ipi");
806 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_ipi,
809 evtchn = bind_ipi.port;
811 xen_irq_info_ipi_init(cpu, irq, evtchn, ipi);
813 bind_evtchn_to_cpu(evtchn, cpu);
817 spin_unlock(&irq_mapping_update_lock);
822 int bind_virq_to_irq(unsigned int virq, unsigned int cpu)
824 struct evtchn_bind_virq bind_virq;
827 spin_lock(&irq_mapping_update_lock);
829 irq = per_cpu(virq_to_irq, cpu)[virq];
832 irq = xen_allocate_irq_dynamic();
836 set_irq_chip_and_handler_name(irq, &xen_percpu_chip,
837 handle_percpu_irq, "virq");
839 bind_virq.virq = virq;
840 bind_virq.vcpu = cpu;
841 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_virq,
844 evtchn = bind_virq.port;
846 xen_irq_info_virq_init(cpu, irq, evtchn, virq);
848 bind_evtchn_to_cpu(evtchn, cpu);
852 spin_unlock(&irq_mapping_update_lock);
857 static void unbind_from_irq(unsigned int irq)
859 struct evtchn_close close;
860 int evtchn = evtchn_from_irq(irq);
862 spin_lock(&irq_mapping_update_lock);
864 if (VALID_EVTCHN(evtchn)) {
866 if (HYPERVISOR_event_channel_op(EVTCHNOP_close, &close) != 0)
869 switch (type_from_irq(irq)) {
871 per_cpu(virq_to_irq, cpu_from_evtchn(evtchn))
872 [virq_from_irq(irq)] = -1;
875 per_cpu(ipi_to_irq, cpu_from_evtchn(evtchn))
876 [ipi_from_irq(irq)] = -1;
882 /* Closed ports are implicitly re-bound to VCPU0. */
883 bind_evtchn_to_cpu(evtchn, 0);
885 evtchn_to_irq[evtchn] = -1;
888 BUG_ON(info_for_irq(irq)->type == IRQT_UNBOUND);
892 spin_unlock(&irq_mapping_update_lock);
895 int bind_evtchn_to_irqhandler(unsigned int evtchn,
896 irq_handler_t handler,
897 unsigned long irqflags,
898 const char *devname, void *dev_id)
903 irq = bind_evtchn_to_irq(evtchn);
906 retval = request_irq(irq, handler, irqflags, devname, dev_id);
908 unbind_from_irq(irq);
914 EXPORT_SYMBOL_GPL(bind_evtchn_to_irqhandler);
916 int bind_virq_to_irqhandler(unsigned int virq, unsigned int cpu,
917 irq_handler_t handler,
918 unsigned long irqflags, const char *devname, void *dev_id)
923 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;
1269 struct irq_info *info;
1271 list_for_each_entry(info, &xen_irq_list_head, list) {
1272 if (info->type != IRQT_PIRQ)
1275 pirq = info->u.pirq.pirq;
1276 gsi = info->u.pirq.gsi;
1279 /* save/restore of PT devices doesn't work, so at this point the
1280 * only devices present are GSI based emulated devices */
1284 map_irq.domid = DOMID_SELF;
1285 map_irq.type = MAP_PIRQ_TYPE_GSI;
1286 map_irq.index = gsi;
1287 map_irq.pirq = pirq;
1289 rc = HYPERVISOR_physdev_op(PHYSDEVOP_map_pirq, &map_irq);
1291 printk(KERN_WARNING "xen map irq failed gsi=%d irq=%d pirq=%d rc=%d\n",
1292 gsi, irq, pirq, rc);
1297 printk(KERN_DEBUG "xen: --> irq=%d, pirq=%d\n", irq, map_irq.pirq);
1299 __startup_pirq(irq);
1303 static void restore_cpu_virqs(unsigned int cpu)
1305 struct evtchn_bind_virq bind_virq;
1306 int virq, irq, evtchn;
1308 for (virq = 0; virq < NR_VIRQS; virq++) {
1309 if ((irq = per_cpu(virq_to_irq, cpu)[virq]) == -1)
1312 BUG_ON(virq_from_irq(irq) != virq);
1314 /* Get a new binding from Xen. */
1315 bind_virq.virq = virq;
1316 bind_virq.vcpu = cpu;
1317 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_virq,
1320 evtchn = bind_virq.port;
1322 /* Record the new mapping. */
1323 xen_irq_info_virq_init(cpu, irq, evtchn, virq);
1324 bind_evtchn_to_cpu(evtchn, cpu);
1328 static void restore_cpu_ipis(unsigned int cpu)
1330 struct evtchn_bind_ipi bind_ipi;
1331 int ipi, irq, evtchn;
1333 for (ipi = 0; ipi < XEN_NR_IPIS; ipi++) {
1334 if ((irq = per_cpu(ipi_to_irq, cpu)[ipi]) == -1)
1337 BUG_ON(ipi_from_irq(irq) != ipi);
1339 /* Get a new binding from Xen. */
1340 bind_ipi.vcpu = cpu;
1341 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_ipi,
1344 evtchn = bind_ipi.port;
1346 /* Record the new mapping. */
1347 xen_irq_info_ipi_init(cpu, irq, evtchn, ipi);
1348 bind_evtchn_to_cpu(evtchn, cpu);
1352 /* Clear an irq's pending state, in preparation for polling on it */
1353 void xen_clear_irq_pending(int irq)
1355 int evtchn = evtchn_from_irq(irq);
1357 if (VALID_EVTCHN(evtchn))
1358 clear_evtchn(evtchn);
1360 EXPORT_SYMBOL(xen_clear_irq_pending);
1361 void xen_set_irq_pending(int irq)
1363 int evtchn = evtchn_from_irq(irq);
1365 if (VALID_EVTCHN(evtchn))
1369 bool xen_test_irq_pending(int irq)
1371 int evtchn = evtchn_from_irq(irq);
1374 if (VALID_EVTCHN(evtchn))
1375 ret = test_evtchn(evtchn);
1380 /* Poll waiting for an irq to become pending with timeout. In the usual case,
1381 * the irq will be disabled so it won't deliver an interrupt. */
1382 void xen_poll_irq_timeout(int irq, u64 timeout)
1384 evtchn_port_t evtchn = evtchn_from_irq(irq);
1386 if (VALID_EVTCHN(evtchn)) {
1387 struct sched_poll poll;
1390 poll.timeout = timeout;
1391 set_xen_guest_handle(poll.ports, &evtchn);
1393 if (HYPERVISOR_sched_op(SCHEDOP_poll, &poll) != 0)
1397 EXPORT_SYMBOL(xen_poll_irq_timeout);
1398 /* Poll waiting for an irq to become pending. In the usual case, the
1399 * irq will be disabled so it won't deliver an interrupt. */
1400 void xen_poll_irq(int irq)
1402 xen_poll_irq_timeout(irq, 0 /* no timeout */);
1405 void xen_irq_resume(void)
1407 unsigned int cpu, evtchn;
1408 struct irq_info *info;
1410 init_evtchn_cpu_bindings();
1412 /* New event-channel space is not 'live' yet. */
1413 for (evtchn = 0; evtchn < NR_EVENT_CHANNELS; evtchn++)
1414 mask_evtchn(evtchn);
1416 /* No IRQ <-> event-channel mappings. */
1417 list_for_each_entry(info, &xen_irq_list_head, list)
1418 info->evtchn = 0; /* zap event-channel binding */
1420 for (evtchn = 0; evtchn < NR_EVENT_CHANNELS; evtchn++)
1421 evtchn_to_irq[evtchn] = -1;
1423 for_each_possible_cpu(cpu) {
1424 restore_cpu_virqs(cpu);
1425 restore_cpu_ipis(cpu);
1431 static struct irq_chip xen_dynamic_chip __read_mostly = {
1434 .irq_disable = disable_dynirq,
1435 .irq_mask = disable_dynirq,
1436 .irq_unmask = enable_dynirq,
1438 .irq_eoi = ack_dynirq,
1439 .irq_set_affinity = set_affinity_irq,
1440 .irq_retrigger = retrigger_dynirq,
1443 static struct irq_chip xen_pirq_chip __read_mostly = {
1446 .irq_startup = startup_pirq,
1447 .irq_shutdown = shutdown_pirq,
1449 .irq_enable = enable_pirq,
1450 .irq_unmask = enable_pirq,
1452 .irq_disable = disable_pirq,
1453 .irq_mask = disable_pirq,
1455 .irq_ack = ack_pirq,
1457 .irq_set_affinity = set_affinity_irq,
1459 .irq_retrigger = retrigger_dynirq,
1462 static struct irq_chip xen_percpu_chip __read_mostly = {
1463 .name = "xen-percpu",
1465 .irq_disable = disable_dynirq,
1466 .irq_mask = disable_dynirq,
1467 .irq_unmask = enable_dynirq,
1469 .irq_ack = ack_dynirq,
1472 int xen_set_callback_via(uint64_t via)
1474 struct xen_hvm_param a;
1475 a.domid = DOMID_SELF;
1476 a.index = HVM_PARAM_CALLBACK_IRQ;
1478 return HYPERVISOR_hvm_op(HVMOP_set_param, &a);
1480 EXPORT_SYMBOL_GPL(xen_set_callback_via);
1482 #ifdef CONFIG_XEN_PVHVM
1483 /* Vector callbacks are better than PCI interrupts to receive event
1484 * channel notifications because we can receive vector callbacks on any
1485 * vcpu and we don't need PCI support or APIC interactions. */
1486 void xen_callback_vector(void)
1489 uint64_t callback_via;
1490 if (xen_have_vector_callback) {
1491 callback_via = HVM_CALLBACK_VECTOR(XEN_HVM_EVTCHN_CALLBACK);
1492 rc = xen_set_callback_via(callback_via);
1494 printk(KERN_ERR "Request for Xen HVM callback vector"
1496 xen_have_vector_callback = 0;
1499 printk(KERN_INFO "Xen HVM callback vector for event delivery is "
1501 /* in the restore case the vector has already been allocated */
1502 if (!test_bit(XEN_HVM_EVTCHN_CALLBACK, used_vectors))
1503 alloc_intr_gate(XEN_HVM_EVTCHN_CALLBACK, xen_hvm_callback_vector);
1507 void xen_callback_vector(void) {}
1510 void __init xen_init_IRQ(void)
1514 evtchn_to_irq = kcalloc(NR_EVENT_CHANNELS, sizeof(*evtchn_to_irq),
1516 for (i = 0; i < NR_EVENT_CHANNELS; i++)
1517 evtchn_to_irq[i] = -1;
1519 init_evtchn_cpu_bindings();
1521 /* No event channels are 'live' right now. */
1522 for (i = 0; i < NR_EVENT_CHANNELS; i++)
1525 if (xen_hvm_domain()) {
1526 xen_callback_vector();
1528 /* pci_xen_hvm_init must be called after native_init_IRQ so that
1529 * __acpi_register_gsi can point at the right function */
1532 irq_ctx_init(smp_processor_id());
1533 if (xen_initial_domain())