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);
395 /* By default all event channels notify CPU#0. */
396 cpumask_copy(desc->irq_data.affinity, cpumask_of(0));
398 info = kzalloc(sizeof(*info), GFP_KERNEL);
400 panic("Unable to allocate metadata for IRQ%d\n", irq);
402 info->type = IRQT_UNBOUND;
404 set_irq_data(irq, info);
406 list_add_tail(&info->list, &xen_irq_list_head);
409 static int __must_check xen_allocate_irq_dynamic(void)
414 #ifdef CONFIG_X86_IO_APIC
416 * For an HVM guest or domain 0 which see "real" (emulated or
417 * actual repectively) GSIs we allocate dynamic IRQs
418 * e.g. those corresponding to event channels or MSIs
419 * etc. from the range above those "real" GSIs to avoid
422 if (xen_initial_domain() || xen_hvm_domain())
423 first = get_nr_irqs_gsi();
426 irq = irq_alloc_desc_from(first, -1);
433 static int __must_check xen_allocate_irq_gsi(unsigned gsi)
438 * A PV guest has no concept of a GSI (since it has no ACPI
439 * nor access to/knowledge of the physical APICs). Therefore
440 * all IRQs are dynamically allocated from the entire IRQ
443 if (xen_pv_domain() && !xen_initial_domain())
444 return xen_allocate_irq_dynamic();
446 /* Legacy IRQ descriptors are already allocated by the arch. */
447 if (gsi < NR_IRQS_LEGACY)
450 irq = irq_alloc_desc_at(gsi, -1);
457 static void xen_free_irq(unsigned irq)
459 struct irq_info *info = get_irq_data(irq);
461 list_del(&info->list);
463 set_irq_data(irq, NULL);
467 /* Legacy IRQ descriptors are managed by the arch. */
468 if (irq < NR_IRQS_LEGACY)
474 static void pirq_unmask_notify(int irq)
476 struct physdev_eoi eoi = { .irq = pirq_from_irq(irq) };
478 if (unlikely(pirq_needs_eoi(irq))) {
479 int rc = HYPERVISOR_physdev_op(PHYSDEVOP_eoi, &eoi);
484 static void pirq_query_unmask(int irq)
486 struct physdev_irq_status_query irq_status;
487 struct irq_info *info = info_for_irq(irq);
489 BUG_ON(info->type != IRQT_PIRQ);
491 irq_status.irq = pirq_from_irq(irq);
492 if (HYPERVISOR_physdev_op(PHYSDEVOP_irq_status_query, &irq_status))
493 irq_status.flags = 0;
495 info->u.pirq.flags &= ~PIRQ_NEEDS_EOI;
496 if (irq_status.flags & XENIRQSTAT_needs_eoi)
497 info->u.pirq.flags |= PIRQ_NEEDS_EOI;
500 static bool probing_irq(int irq)
502 struct irq_desc *desc = irq_to_desc(irq);
504 return desc && desc->action == NULL;
507 static unsigned int __startup_pirq(unsigned int irq)
509 struct evtchn_bind_pirq bind_pirq;
510 struct irq_info *info = info_for_irq(irq);
511 int evtchn = evtchn_from_irq(irq);
514 BUG_ON(info->type != IRQT_PIRQ);
516 if (VALID_EVTCHN(evtchn))
519 bind_pirq.pirq = pirq_from_irq(irq);
520 /* NB. We are happy to share unless we are probing. */
521 bind_pirq.flags = info->u.pirq.flags & PIRQ_SHAREABLE ?
522 BIND_PIRQ__WILL_SHARE : 0;
523 rc = HYPERVISOR_event_channel_op(EVTCHNOP_bind_pirq, &bind_pirq);
525 if (!probing_irq(irq))
526 printk(KERN_INFO "Failed to obtain physical IRQ %d\n",
530 evtchn = bind_pirq.port;
532 pirq_query_unmask(irq);
534 evtchn_to_irq[evtchn] = irq;
535 bind_evtchn_to_cpu(evtchn, 0);
536 info->evtchn = evtchn;
539 unmask_evtchn(evtchn);
540 pirq_unmask_notify(irq);
545 static unsigned int startup_pirq(struct irq_data *data)
547 return __startup_pirq(data->irq);
550 static void shutdown_pirq(struct irq_data *data)
552 struct evtchn_close close;
553 unsigned int irq = data->irq;
554 struct irq_info *info = info_for_irq(irq);
555 int evtchn = evtchn_from_irq(irq);
557 BUG_ON(info->type != IRQT_PIRQ);
559 if (!VALID_EVTCHN(evtchn))
565 if (HYPERVISOR_event_channel_op(EVTCHNOP_close, &close) != 0)
568 bind_evtchn_to_cpu(evtchn, 0);
569 evtchn_to_irq[evtchn] = -1;
573 static void enable_pirq(struct irq_data *data)
578 static void disable_pirq(struct irq_data *data)
582 static void ack_pirq(struct irq_data *data)
584 int evtchn = evtchn_from_irq(data->irq);
586 move_native_irq(data->irq);
588 if (VALID_EVTCHN(evtchn)) {
590 clear_evtchn(evtchn);
594 static int find_irq_by_gsi(unsigned gsi)
596 struct irq_info *info;
598 list_for_each_entry(info, &xen_irq_list_head, list) {
599 if (info->type != IRQT_PIRQ)
602 if (info->u.pirq.gsi == gsi)
609 int xen_allocate_pirq_gsi(unsigned gsi)
615 * Do not make any assumptions regarding the relationship between the
616 * IRQ number returned here and the Xen pirq argument.
618 * Note: We don't assign an event channel until the irq actually started
619 * up. Return an existing irq if we've already got one for the gsi.
621 int xen_bind_pirq_gsi_to_irq(unsigned gsi,
622 unsigned pirq, int shareable, char *name)
625 struct physdev_irq irq_op;
627 spin_lock(&irq_mapping_update_lock);
629 irq = find_irq_by_gsi(gsi);
631 printk(KERN_INFO "xen_map_pirq_gsi: returning irq %d for gsi %u\n",
633 goto out; /* XXX need refcount? */
636 irq = xen_allocate_irq_gsi(gsi);
640 set_irq_chip_and_handler_name(irq, &xen_pirq_chip,
641 handle_level_irq, name);
646 /* Only the privileged domain can do this. For non-priv, the pcifront
647 * driver provides a PCI bus that does the call to do exactly
648 * this in the priv domain. */
649 if (xen_initial_domain() &&
650 HYPERVISOR_physdev_op(PHYSDEVOP_alloc_irq_vector, &irq_op)) {
656 xen_irq_info_pirq_init(irq, 0, pirq, gsi, irq_op.vector,
657 shareable ? PIRQ_SHAREABLE : 0);
660 spin_unlock(&irq_mapping_update_lock);
665 #ifdef CONFIG_PCI_MSI
666 int xen_allocate_pirq_msi(struct pci_dev *dev, struct msi_desc *msidesc)
669 struct physdev_get_free_pirq op_get_free_pirq;
671 op_get_free_pirq.type = MAP_PIRQ_TYPE_MSI;
672 rc = HYPERVISOR_physdev_op(PHYSDEVOP_get_free_pirq, &op_get_free_pirq);
674 WARN_ONCE(rc == -ENOSYS,
675 "hypervisor does not support the PHYSDEVOP_get_free_pirq interface\n");
677 return rc ? -1 : op_get_free_pirq.pirq;
680 int xen_bind_pirq_msi_to_irq(struct pci_dev *dev, struct msi_desc *msidesc,
681 int pirq, int vector, const char *name)
685 spin_lock(&irq_mapping_update_lock);
687 irq = xen_allocate_irq_dynamic();
691 set_irq_chip_and_handler_name(irq, &xen_pirq_chip,
692 handle_level_irq, name);
694 xen_irq_info_pirq_init(irq, 0, pirq, 0, vector, 0);
695 ret = set_irq_msi(irq, msidesc);
699 spin_unlock(&irq_mapping_update_lock);
702 spin_unlock(&irq_mapping_update_lock);
708 int xen_destroy_irq(int irq)
710 struct irq_desc *desc;
711 struct physdev_unmap_pirq unmap_irq;
712 struct irq_info *info = info_for_irq(irq);
715 spin_lock(&irq_mapping_update_lock);
717 desc = irq_to_desc(irq);
721 if (xen_initial_domain()) {
722 unmap_irq.pirq = info->u.pirq.pirq;
723 unmap_irq.domid = DOMID_SELF;
724 rc = HYPERVISOR_physdev_op(PHYSDEVOP_unmap_pirq, &unmap_irq);
726 printk(KERN_WARNING "unmap irq failed %d\n", rc);
734 spin_unlock(&irq_mapping_update_lock);
738 int xen_irq_from_pirq(unsigned pirq)
742 struct irq_info *info;
744 spin_lock(&irq_mapping_update_lock);
746 list_for_each_entry(info, &xen_irq_list_head, list) {
747 if (info == NULL || info->type != IRQT_PIRQ)
750 if (info->u.pirq.pirq == pirq)
755 spin_unlock(&irq_mapping_update_lock);
760 int bind_evtchn_to_irq(unsigned int evtchn)
764 spin_lock(&irq_mapping_update_lock);
766 irq = evtchn_to_irq[evtchn];
769 irq = xen_allocate_irq_dynamic();
773 set_irq_chip_and_handler_name(irq, &xen_dynamic_chip,
774 handle_fasteoi_irq, "event");
776 xen_irq_info_evtchn_init(irq, evtchn);
780 spin_unlock(&irq_mapping_update_lock);
784 EXPORT_SYMBOL_GPL(bind_evtchn_to_irq);
786 static int bind_ipi_to_irq(unsigned int ipi, unsigned int cpu)
788 struct evtchn_bind_ipi bind_ipi;
791 spin_lock(&irq_mapping_update_lock);
793 irq = per_cpu(ipi_to_irq, cpu)[ipi];
796 irq = xen_allocate_irq_dynamic();
800 set_irq_chip_and_handler_name(irq, &xen_percpu_chip,
801 handle_percpu_irq, "ipi");
804 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_ipi,
807 evtchn = bind_ipi.port;
809 xen_irq_info_ipi_init(cpu, irq, evtchn, ipi);
811 bind_evtchn_to_cpu(evtchn, cpu);
815 spin_unlock(&irq_mapping_update_lock);
820 int bind_virq_to_irq(unsigned int virq, unsigned int cpu)
822 struct evtchn_bind_virq bind_virq;
825 spin_lock(&irq_mapping_update_lock);
827 irq = per_cpu(virq_to_irq, cpu)[virq];
830 irq = xen_allocate_irq_dynamic();
834 set_irq_chip_and_handler_name(irq, &xen_percpu_chip,
835 handle_percpu_irq, "virq");
837 bind_virq.virq = virq;
838 bind_virq.vcpu = cpu;
839 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_virq,
842 evtchn = bind_virq.port;
844 xen_irq_info_virq_init(cpu, irq, evtchn, virq);
846 bind_evtchn_to_cpu(evtchn, cpu);
850 spin_unlock(&irq_mapping_update_lock);
855 static void unbind_from_irq(unsigned int irq)
857 struct evtchn_close close;
858 int evtchn = evtchn_from_irq(irq);
860 spin_lock(&irq_mapping_update_lock);
862 if (VALID_EVTCHN(evtchn)) {
864 if (HYPERVISOR_event_channel_op(EVTCHNOP_close, &close) != 0)
867 switch (type_from_irq(irq)) {
869 per_cpu(virq_to_irq, cpu_from_evtchn(evtchn))
870 [virq_from_irq(irq)] = -1;
873 per_cpu(ipi_to_irq, cpu_from_evtchn(evtchn))
874 [ipi_from_irq(irq)] = -1;
880 /* Closed ports are implicitly re-bound to VCPU0. */
881 bind_evtchn_to_cpu(evtchn, 0);
883 evtchn_to_irq[evtchn] = -1;
886 BUG_ON(info_for_irq(irq)->type == IRQT_UNBOUND);
890 spin_unlock(&irq_mapping_update_lock);
893 int bind_evtchn_to_irqhandler(unsigned int evtchn,
894 irq_handler_t handler,
895 unsigned long irqflags,
896 const char *devname, void *dev_id)
901 irq = bind_evtchn_to_irq(evtchn);
904 retval = request_irq(irq, handler, irqflags, devname, dev_id);
906 unbind_from_irq(irq);
912 EXPORT_SYMBOL_GPL(bind_evtchn_to_irqhandler);
914 int bind_virq_to_irqhandler(unsigned int virq, unsigned int cpu,
915 irq_handler_t handler,
916 unsigned long irqflags, const char *devname, void *dev_id)
921 irq = bind_virq_to_irq(virq, cpu);
924 retval = request_irq(irq, handler, irqflags, devname, dev_id);
926 unbind_from_irq(irq);
932 EXPORT_SYMBOL_GPL(bind_virq_to_irqhandler);
934 int bind_ipi_to_irqhandler(enum ipi_vector ipi,
936 irq_handler_t handler,
937 unsigned long irqflags,
943 irq = bind_ipi_to_irq(ipi, cpu);
947 irqflags |= IRQF_NO_SUSPEND | IRQF_FORCE_RESUME;
948 retval = request_irq(irq, handler, irqflags, devname, dev_id);
950 unbind_from_irq(irq);
957 void unbind_from_irqhandler(unsigned int irq, void *dev_id)
959 free_irq(irq, dev_id);
960 unbind_from_irq(irq);
962 EXPORT_SYMBOL_GPL(unbind_from_irqhandler);
964 void xen_send_IPI_one(unsigned int cpu, enum ipi_vector vector)
966 int irq = per_cpu(ipi_to_irq, cpu)[vector];
968 notify_remote_via_irq(irq);
971 irqreturn_t xen_debug_interrupt(int irq, void *dev_id)
973 struct shared_info *sh = HYPERVISOR_shared_info;
974 int cpu = smp_processor_id();
975 unsigned long *cpu_evtchn = per_cpu(cpu_evtchn_mask, cpu);
978 static DEFINE_SPINLOCK(debug_lock);
981 spin_lock_irqsave(&debug_lock, flags);
983 printk("\nvcpu %d\n ", cpu);
985 for_each_online_cpu(i) {
987 v = per_cpu(xen_vcpu, i);
988 pending = (get_irq_regs() && i == cpu)
989 ? xen_irqs_disabled(get_irq_regs())
990 : v->evtchn_upcall_mask;
991 printk("%d: masked=%d pending=%d event_sel %0*lx\n ", i,
992 pending, v->evtchn_upcall_pending,
993 (int)(sizeof(v->evtchn_pending_sel)*2),
994 v->evtchn_pending_sel);
996 v = per_cpu(xen_vcpu, cpu);
998 printk("\npending:\n ");
999 for (i = ARRAY_SIZE(sh->evtchn_pending)-1; i >= 0; i--)
1000 printk("%0*lx%s", (int)sizeof(sh->evtchn_pending[0])*2,
1001 sh->evtchn_pending[i],
1002 i % 8 == 0 ? "\n " : " ");
1003 printk("\nglobal mask:\n ");
1004 for (i = ARRAY_SIZE(sh->evtchn_mask)-1; i >= 0; i--)
1006 (int)(sizeof(sh->evtchn_mask[0])*2),
1008 i % 8 == 0 ? "\n " : " ");
1010 printk("\nglobally unmasked:\n ");
1011 for (i = ARRAY_SIZE(sh->evtchn_mask)-1; i >= 0; i--)
1012 printk("%0*lx%s", (int)(sizeof(sh->evtchn_mask[0])*2),
1013 sh->evtchn_pending[i] & ~sh->evtchn_mask[i],
1014 i % 8 == 0 ? "\n " : " ");
1016 printk("\nlocal cpu%d mask:\n ", cpu);
1017 for (i = (NR_EVENT_CHANNELS/BITS_PER_LONG)-1; i >= 0; i--)
1018 printk("%0*lx%s", (int)(sizeof(cpu_evtchn[0])*2),
1020 i % 8 == 0 ? "\n " : " ");
1022 printk("\nlocally unmasked:\n ");
1023 for (i = ARRAY_SIZE(sh->evtchn_mask)-1; i >= 0; i--) {
1024 unsigned long pending = sh->evtchn_pending[i]
1025 & ~sh->evtchn_mask[i]
1027 printk("%0*lx%s", (int)(sizeof(sh->evtchn_mask[0])*2),
1028 pending, i % 8 == 0 ? "\n " : " ");
1031 printk("\npending list:\n");
1032 for (i = 0; i < NR_EVENT_CHANNELS; i++) {
1033 if (sync_test_bit(i, sh->evtchn_pending)) {
1034 int word_idx = i / BITS_PER_LONG;
1035 printk(" %d: event %d -> irq %d%s%s%s\n",
1036 cpu_from_evtchn(i), i,
1038 sync_test_bit(word_idx, &v->evtchn_pending_sel)
1040 !sync_test_bit(i, sh->evtchn_mask)
1041 ? "" : " globally-masked",
1042 sync_test_bit(i, cpu_evtchn)
1043 ? "" : " locally-masked");
1047 spin_unlock_irqrestore(&debug_lock, flags);
1052 static DEFINE_PER_CPU(unsigned, xed_nesting_count);
1055 * Search the CPUs pending events bitmasks. For each one found, map
1056 * the event number to an irq, and feed it into do_IRQ() for
1059 * Xen uses a two-level bitmap to speed searching. The first level is
1060 * a bitset of words which contain pending event bits. The second
1061 * level is a bitset of pending events themselves.
1063 static void __xen_evtchn_do_upcall(void)
1065 int cpu = get_cpu();
1066 struct shared_info *s = HYPERVISOR_shared_info;
1067 struct vcpu_info *vcpu_info = __this_cpu_read(xen_vcpu);
1071 unsigned long pending_words;
1073 vcpu_info->evtchn_upcall_pending = 0;
1075 if (__this_cpu_inc_return(xed_nesting_count) - 1)
1078 #ifndef CONFIG_X86 /* No need for a barrier -- XCHG is a barrier on x86. */
1079 /* Clear master flag /before/ clearing selector flag. */
1082 pending_words = xchg(&vcpu_info->evtchn_pending_sel, 0);
1083 while (pending_words != 0) {
1084 unsigned long pending_bits;
1085 int word_idx = __ffs(pending_words);
1086 pending_words &= ~(1UL << word_idx);
1088 while ((pending_bits = active_evtchns(cpu, s, word_idx)) != 0) {
1089 int bit_idx = __ffs(pending_bits);
1090 int port = (word_idx * BITS_PER_LONG) + bit_idx;
1091 int irq = evtchn_to_irq[port];
1092 struct irq_desc *desc;
1098 desc = irq_to_desc(irq);
1100 generic_handle_irq_desc(irq, desc);
1105 BUG_ON(!irqs_disabled());
1107 count = __this_cpu_read(xed_nesting_count);
1108 __this_cpu_write(xed_nesting_count, 0);
1109 } while (count != 1 || vcpu_info->evtchn_upcall_pending);
1116 void xen_evtchn_do_upcall(struct pt_regs *regs)
1118 struct pt_regs *old_regs = set_irq_regs(regs);
1123 __xen_evtchn_do_upcall();
1126 set_irq_regs(old_regs);
1129 void xen_hvm_evtchn_do_upcall(void)
1131 __xen_evtchn_do_upcall();
1133 EXPORT_SYMBOL_GPL(xen_hvm_evtchn_do_upcall);
1135 /* Rebind a new event channel to an existing irq. */
1136 void rebind_evtchn_irq(int evtchn, int irq)
1138 struct irq_info *info = info_for_irq(irq);
1140 /* Make sure the irq is masked, since the new event channel
1141 will also be masked. */
1144 spin_lock(&irq_mapping_update_lock);
1146 /* After resume the irq<->evtchn mappings are all cleared out */
1147 BUG_ON(evtchn_to_irq[evtchn] != -1);
1148 /* Expect irq to have been bound before,
1149 so there should be a proper type */
1150 BUG_ON(info->type == IRQT_UNBOUND);
1152 xen_irq_info_evtchn_init(irq, evtchn);
1154 spin_unlock(&irq_mapping_update_lock);
1156 /* new event channels are always bound to cpu 0 */
1157 irq_set_affinity(irq, cpumask_of(0));
1159 /* Unmask the event channel. */
1163 /* Rebind an evtchn so that it gets delivered to a specific cpu */
1164 static int rebind_irq_to_cpu(unsigned irq, unsigned tcpu)
1166 struct evtchn_bind_vcpu bind_vcpu;
1167 int evtchn = evtchn_from_irq(irq);
1169 if (!VALID_EVTCHN(evtchn))
1173 * Events delivered via platform PCI interrupts are always
1174 * routed to vcpu 0 and hence cannot be rebound.
1176 if (xen_hvm_domain() && !xen_have_vector_callback)
1179 /* Send future instances of this interrupt to other vcpu. */
1180 bind_vcpu.port = evtchn;
1181 bind_vcpu.vcpu = tcpu;
1184 * If this fails, it usually just indicates that we're dealing with a
1185 * virq or IPI channel, which don't actually need to be rebound. Ignore
1186 * it, but don't do the xenlinux-level rebind in that case.
1188 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_vcpu, &bind_vcpu) >= 0)
1189 bind_evtchn_to_cpu(evtchn, tcpu);
1194 static int set_affinity_irq(struct irq_data *data, const struct cpumask *dest,
1197 unsigned tcpu = cpumask_first(dest);
1199 return rebind_irq_to_cpu(data->irq, tcpu);
1202 int resend_irq_on_evtchn(unsigned int irq)
1204 int masked, evtchn = evtchn_from_irq(irq);
1205 struct shared_info *s = HYPERVISOR_shared_info;
1207 if (!VALID_EVTCHN(evtchn))
1210 masked = sync_test_and_set_bit(evtchn, s->evtchn_mask);
1211 sync_set_bit(evtchn, s->evtchn_pending);
1213 unmask_evtchn(evtchn);
1218 static void enable_dynirq(struct irq_data *data)
1220 int evtchn = evtchn_from_irq(data->irq);
1222 if (VALID_EVTCHN(evtchn))
1223 unmask_evtchn(evtchn);
1226 static void disable_dynirq(struct irq_data *data)
1228 int evtchn = evtchn_from_irq(data->irq);
1230 if (VALID_EVTCHN(evtchn))
1231 mask_evtchn(evtchn);
1234 static void ack_dynirq(struct irq_data *data)
1236 int evtchn = evtchn_from_irq(data->irq);
1238 move_masked_irq(data->irq);
1240 if (VALID_EVTCHN(evtchn))
1241 unmask_evtchn(evtchn);
1244 static int retrigger_dynirq(struct irq_data *data)
1246 int evtchn = evtchn_from_irq(data->irq);
1247 struct shared_info *sh = HYPERVISOR_shared_info;
1250 if (VALID_EVTCHN(evtchn)) {
1253 masked = sync_test_and_set_bit(evtchn, sh->evtchn_mask);
1254 sync_set_bit(evtchn, sh->evtchn_pending);
1256 unmask_evtchn(evtchn);
1263 static void restore_pirqs(void)
1265 int pirq, rc, irq, gsi;
1266 struct physdev_map_pirq map_irq;
1267 struct irq_info *info;
1269 list_for_each_entry(info, &xen_irq_list_head, list) {
1270 if (info->type != IRQT_PIRQ)
1273 pirq = info->u.pirq.pirq;
1274 gsi = info->u.pirq.gsi;
1277 /* save/restore of PT devices doesn't work, so at this point the
1278 * only devices present are GSI based emulated devices */
1282 map_irq.domid = DOMID_SELF;
1283 map_irq.type = MAP_PIRQ_TYPE_GSI;
1284 map_irq.index = gsi;
1285 map_irq.pirq = pirq;
1287 rc = HYPERVISOR_physdev_op(PHYSDEVOP_map_pirq, &map_irq);
1289 printk(KERN_WARNING "xen map irq failed gsi=%d irq=%d pirq=%d rc=%d\n",
1290 gsi, irq, pirq, rc);
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 evtchn_to_irq = kcalloc(NR_EVENT_CHANNELS, sizeof(*evtchn_to_irq),
1514 for (i = 0; i < NR_EVENT_CHANNELS; i++)
1515 evtchn_to_irq[i] = -1;
1517 init_evtchn_cpu_bindings();
1519 /* No event channels are 'live' right now. */
1520 for (i = 0; i < NR_EVENT_CHANNELS; i++)
1523 if (xen_hvm_domain()) {
1524 xen_callback_vector();
1526 /* pci_xen_hvm_init must be called after native_init_IRQ so that
1527 * __acpi_register_gsi can point at the right function */
1530 irq_ctx_init(smp_processor_id());
1531 if (xen_initial_domain())