4 * Xen models interrupts with abstract event channels. Because each
5 * domain gets 1024 event channels, but NR_IRQ is not that large, we
6 * must dynamically map irqs<->event channels. The event channels
7 * interface with the rest of the kernel by defining a xen interrupt
8 * chip. When an event is recieved, it is mapped to an irq and sent
9 * through the normal interrupt processing path.
11 * There are four kinds of events which can be mapped to an event
14 * 1. Inter-domain notifications. This includes all the virtual
15 * device events, since they're driven by front-ends in another domain
17 * 2. VIRQs, typically used for timers. These are per-cpu events.
19 * 4. PIRQs - Hardware interrupts.
21 * Jeremy Fitzhardinge <jeremy@xensource.com>, XenSource Inc, 2007
24 #include <linux/linkage.h>
25 #include <linux/interrupt.h>
26 #include <linux/irq.h>
27 #include <linux/module.h>
28 #include <linux/string.h>
29 #include <linux/bootmem.h>
30 #include <linux/slab.h>
31 #include <linux/irqnr.h>
32 #include <linux/pci.h>
35 #include <asm/ptrace.h>
38 #include <asm/io_apic.h>
39 #include <asm/sync_bitops.h>
40 #include <asm/xen/pci.h>
41 #include <asm/xen/hypercall.h>
42 #include <asm/xen/hypervisor.h>
46 #include <xen/xen-ops.h>
47 #include <xen/events.h>
48 #include <xen/interface/xen.h>
49 #include <xen/interface/event_channel.h>
50 #include <xen/interface/hvm/hvm_op.h>
51 #include <xen/interface/hvm/params.h>
54 * This lock protects updates to the following mapping and reference-count
55 * arrays. The lock does not need to be acquired to read the mapping tables.
57 static DEFINE_SPINLOCK(irq_mapping_update_lock);
59 /* IRQ <-> VIRQ mapping. */
60 static DEFINE_PER_CPU(int [NR_VIRQS], virq_to_irq) = {[0 ... NR_VIRQS-1] = -1};
62 /* IRQ <-> IPI mapping */
63 static DEFINE_PER_CPU(int [XEN_NR_IPIS], ipi_to_irq) = {[0 ... XEN_NR_IPIS-1] = -1};
65 /* Interrupt types. */
75 * Packed IRQ information:
76 * type - enum xen_irq_type
77 * event channel - irq->event channel mapping
78 * cpu - cpu this event channel is bound to
79 * index - type-specific information:
80 * PIRQ - vector, with MSB being "needs EIO", or physical IRQ of the HVM
81 * guest, or GSI (real passthrough IRQ) of the device.
88 enum xen_irq_type type; /* type */
89 unsigned short evtchn; /* event channel */
90 unsigned short cpu; /* cpu bound */
103 #define PIRQ_NEEDS_EOI (1 << 0)
104 #define PIRQ_SHAREABLE (1 << 1)
106 static struct irq_info *irq_info;
107 static int *pirq_to_irq;
109 static int *evtchn_to_irq;
110 struct cpu_evtchn_s {
111 unsigned long bits[NR_EVENT_CHANNELS/BITS_PER_LONG];
114 static __initdata struct cpu_evtchn_s init_evtchn_mask = {
115 .bits[0 ... (NR_EVENT_CHANNELS/BITS_PER_LONG)-1] = ~0ul,
117 static struct cpu_evtchn_s *cpu_evtchn_mask_p = &init_evtchn_mask;
119 static inline unsigned long *cpu_evtchn_mask(int cpu)
121 return cpu_evtchn_mask_p[cpu].bits;
124 /* Xen will never allocate port zero for any purpose. */
125 #define VALID_EVTCHN(chn) ((chn) != 0)
127 static struct irq_chip xen_dynamic_chip;
128 static struct irq_chip xen_percpu_chip;
129 static struct irq_chip xen_pirq_chip;
131 /* Constructor for packed IRQ information. */
132 static struct irq_info mk_unbound_info(void)
134 return (struct irq_info) { .type = IRQT_UNBOUND };
137 static struct irq_info mk_evtchn_info(unsigned short evtchn)
139 return (struct irq_info) { .type = IRQT_EVTCHN, .evtchn = evtchn,
143 static struct irq_info mk_ipi_info(unsigned short evtchn, enum ipi_vector ipi)
145 return (struct irq_info) { .type = IRQT_IPI, .evtchn = evtchn,
146 .cpu = 0, .u.ipi = ipi };
149 static struct irq_info mk_virq_info(unsigned short evtchn, unsigned short virq)
151 return (struct irq_info) { .type = IRQT_VIRQ, .evtchn = evtchn,
152 .cpu = 0, .u.virq = virq };
155 static struct irq_info mk_pirq_info(unsigned short evtchn, unsigned short pirq,
156 unsigned short gsi, unsigned short vector)
158 return (struct irq_info) { .type = IRQT_PIRQ, .evtchn = evtchn,
160 .u.pirq = { .pirq = pirq, .gsi = gsi, .vector = vector } };
164 * Accessors for packed IRQ information.
166 static struct irq_info *info_for_irq(unsigned irq)
168 return &irq_info[irq];
171 static unsigned int evtchn_from_irq(unsigned irq)
173 if (unlikely(WARN(irq < 0 || irq >= nr_irqs, "Invalid irq %d!\n", irq)))
176 return info_for_irq(irq)->evtchn;
179 unsigned irq_from_evtchn(unsigned int evtchn)
181 return evtchn_to_irq[evtchn];
183 EXPORT_SYMBOL_GPL(irq_from_evtchn);
185 static enum ipi_vector ipi_from_irq(unsigned irq)
187 struct irq_info *info = info_for_irq(irq);
189 BUG_ON(info == NULL);
190 BUG_ON(info->type != IRQT_IPI);
195 static unsigned virq_from_irq(unsigned irq)
197 struct irq_info *info = info_for_irq(irq);
199 BUG_ON(info == NULL);
200 BUG_ON(info->type != IRQT_VIRQ);
205 static unsigned pirq_from_irq(unsigned irq)
207 struct irq_info *info = info_for_irq(irq);
209 BUG_ON(info == NULL);
210 BUG_ON(info->type != IRQT_PIRQ);
212 return info->u.pirq.pirq;
215 static unsigned gsi_from_irq(unsigned irq)
217 struct irq_info *info = info_for_irq(irq);
219 BUG_ON(info == NULL);
220 BUG_ON(info->type != IRQT_PIRQ);
222 return info->u.pirq.gsi;
225 static unsigned vector_from_irq(unsigned irq)
227 struct irq_info *info = info_for_irq(irq);
229 BUG_ON(info == NULL);
230 BUG_ON(info->type != IRQT_PIRQ);
232 return info->u.pirq.vector;
235 static enum xen_irq_type type_from_irq(unsigned irq)
237 return info_for_irq(irq)->type;
240 static unsigned cpu_from_irq(unsigned irq)
242 return info_for_irq(irq)->cpu;
245 static unsigned int cpu_from_evtchn(unsigned int evtchn)
247 int irq = evtchn_to_irq[evtchn];
251 ret = cpu_from_irq(irq);
256 static bool pirq_needs_eoi(unsigned irq)
258 struct irq_info *info = info_for_irq(irq);
260 BUG_ON(info->type != IRQT_PIRQ);
262 return info->u.pirq.flags & PIRQ_NEEDS_EOI;
265 static inline unsigned long active_evtchns(unsigned int cpu,
266 struct shared_info *sh,
269 return (sh->evtchn_pending[idx] &
270 cpu_evtchn_mask(cpu)[idx] &
271 ~sh->evtchn_mask[idx]);
274 static void bind_evtchn_to_cpu(unsigned int chn, unsigned int cpu)
276 int irq = evtchn_to_irq[chn];
280 cpumask_copy(irq_to_desc(irq)->affinity, cpumask_of(cpu));
283 clear_bit(chn, cpu_evtchn_mask(cpu_from_irq(irq)));
284 set_bit(chn, cpu_evtchn_mask(cpu));
286 irq_info[irq].cpu = cpu;
289 static void init_evtchn_cpu_bindings(void)
293 struct irq_desc *desc;
295 /* By default all event channels notify CPU#0. */
296 for_each_irq_desc(i, desc) {
297 cpumask_copy(desc->affinity, cpumask_of(0));
301 for_each_possible_cpu(i)
302 memset(cpu_evtchn_mask(i),
303 (i == 0) ? ~0 : 0, sizeof(struct cpu_evtchn_s));
307 static inline void clear_evtchn(int port)
309 struct shared_info *s = HYPERVISOR_shared_info;
310 sync_clear_bit(port, &s->evtchn_pending[0]);
313 static inline void set_evtchn(int port)
315 struct shared_info *s = HYPERVISOR_shared_info;
316 sync_set_bit(port, &s->evtchn_pending[0]);
319 static inline int test_evtchn(int port)
321 struct shared_info *s = HYPERVISOR_shared_info;
322 return sync_test_bit(port, &s->evtchn_pending[0]);
327 * notify_remote_via_irq - send event to remote end of event channel via irq
328 * @irq: irq of event channel to send event to
330 * Unlike notify_remote_via_evtchn(), this is safe to use across
331 * save/restore. Notifications on a broken connection are silently
334 void notify_remote_via_irq(int irq)
336 int evtchn = evtchn_from_irq(irq);
338 if (VALID_EVTCHN(evtchn))
339 notify_remote_via_evtchn(evtchn);
341 EXPORT_SYMBOL_GPL(notify_remote_via_irq);
343 static void mask_evtchn(int port)
345 struct shared_info *s = HYPERVISOR_shared_info;
346 sync_set_bit(port, &s->evtchn_mask[0]);
349 static void unmask_evtchn(int port)
351 struct shared_info *s = HYPERVISOR_shared_info;
352 unsigned int cpu = get_cpu();
354 BUG_ON(!irqs_disabled());
356 /* Slow path (hypercall) if this is a non-local port. */
357 if (unlikely(cpu != cpu_from_evtchn(port))) {
358 struct evtchn_unmask unmask = { .port = port };
359 (void)HYPERVISOR_event_channel_op(EVTCHNOP_unmask, &unmask);
361 struct vcpu_info *vcpu_info = __this_cpu_read(xen_vcpu);
363 sync_clear_bit(port, &s->evtchn_mask[0]);
366 * The following is basically the equivalent of
367 * 'hw_resend_irq'. Just like a real IO-APIC we 'lose
368 * the interrupt edge' if the channel is masked.
370 if (sync_test_bit(port, &s->evtchn_pending[0]) &&
371 !sync_test_and_set_bit(port / BITS_PER_LONG,
372 &vcpu_info->evtchn_pending_sel))
373 vcpu_info->evtchn_upcall_pending = 1;
379 static int get_nr_hw_irqs(void)
383 #ifdef CONFIG_X86_IO_APIC
384 ret = get_nr_irqs_gsi();
390 static int find_unbound_pirq(int type)
393 struct physdev_get_free_pirq op_get_free_pirq;
394 op_get_free_pirq.type = type;
396 rc = HYPERVISOR_physdev_op(PHYSDEVOP_get_free_pirq, &op_get_free_pirq);
398 return op_get_free_pirq.pirq;
400 for (i = 0; i < nr_irqs; i++) {
401 if (pirq_to_irq[i] < 0)
407 static int find_unbound_irq(void)
409 struct irq_data *data;
411 int bottom = get_nr_hw_irqs();
414 if (bottom == nr_irqs)
417 /* This loop starts from the top of IRQ space and goes down.
418 * We need this b/c if we have a PCI device in a Xen PV guest
419 * we do not have an IO-APIC (though the backend might have them)
420 * mapped in. To not have a collision of physical IRQs with the Xen
421 * event channels start at the top of the IRQ space for virtual IRQs.
423 for (irq = top; irq > bottom; irq--) {
424 data = irq_get_irq_data(irq);
425 /* only 15->0 have init'd desc; handle irq > 16 */
428 if (data->chip == &no_irq_chip)
430 if (data->chip != &xen_dynamic_chip)
432 if (irq_info[irq].type == IRQT_UNBOUND)
439 res = irq_alloc_desc_at(irq, -1);
441 if (WARN_ON(res != irq))
447 panic("No available IRQ to bind to: increase nr_irqs!\n");
450 static bool identity_mapped_irq(unsigned irq)
452 /* identity map all the hardware irqs */
453 return irq < get_nr_hw_irqs();
456 static void pirq_unmask_notify(int irq)
458 struct physdev_eoi eoi = { .irq = pirq_from_irq(irq) };
460 if (unlikely(pirq_needs_eoi(irq))) {
461 int rc = HYPERVISOR_physdev_op(PHYSDEVOP_eoi, &eoi);
466 static void pirq_query_unmask(int irq)
468 struct physdev_irq_status_query irq_status;
469 struct irq_info *info = info_for_irq(irq);
471 BUG_ON(info->type != IRQT_PIRQ);
473 irq_status.irq = pirq_from_irq(irq);
474 if (HYPERVISOR_physdev_op(PHYSDEVOP_irq_status_query, &irq_status))
475 irq_status.flags = 0;
477 info->u.pirq.flags &= ~PIRQ_NEEDS_EOI;
478 if (irq_status.flags & XENIRQSTAT_needs_eoi)
479 info->u.pirq.flags |= PIRQ_NEEDS_EOI;
482 static bool probing_irq(int irq)
484 struct irq_desc *desc = irq_to_desc(irq);
486 return desc && desc->action == NULL;
489 static unsigned int startup_pirq(unsigned int irq)
491 struct evtchn_bind_pirq bind_pirq;
492 struct irq_info *info = info_for_irq(irq);
493 int evtchn = evtchn_from_irq(irq);
496 BUG_ON(info->type != IRQT_PIRQ);
498 if (VALID_EVTCHN(evtchn))
501 bind_pirq.pirq = pirq_from_irq(irq);
502 /* NB. We are happy to share unless we are probing. */
503 bind_pirq.flags = info->u.pirq.flags & PIRQ_SHAREABLE ?
504 BIND_PIRQ__WILL_SHARE : 0;
505 rc = HYPERVISOR_event_channel_op(EVTCHNOP_bind_pirq, &bind_pirq);
507 if (!probing_irq(irq))
508 printk(KERN_INFO "Failed to obtain physical IRQ %d\n",
512 evtchn = bind_pirq.port;
514 pirq_query_unmask(irq);
516 evtchn_to_irq[evtchn] = irq;
517 bind_evtchn_to_cpu(evtchn, 0);
518 info->evtchn = evtchn;
521 unmask_evtchn(evtchn);
522 pirq_unmask_notify(irq);
527 static void shutdown_pirq(unsigned int irq)
529 struct evtchn_close close;
530 struct irq_info *info = info_for_irq(irq);
531 int evtchn = evtchn_from_irq(irq);
533 BUG_ON(info->type != IRQT_PIRQ);
535 if (!VALID_EVTCHN(evtchn))
541 if (HYPERVISOR_event_channel_op(EVTCHNOP_close, &close) != 0)
544 bind_evtchn_to_cpu(evtchn, 0);
545 evtchn_to_irq[evtchn] = -1;
549 static void enable_pirq(unsigned int irq)
554 static void disable_pirq(unsigned int irq)
558 static void ack_pirq(unsigned int irq)
560 int evtchn = evtchn_from_irq(irq);
562 move_native_irq(irq);
564 if (VALID_EVTCHN(evtchn)) {
566 clear_evtchn(evtchn);
570 static void end_pirq(unsigned int irq)
572 int evtchn = evtchn_from_irq(irq);
573 struct irq_desc *desc = irq_to_desc(irq);
578 if ((desc->status & (IRQ_DISABLED|IRQ_PENDING)) ==
579 (IRQ_DISABLED|IRQ_PENDING)) {
581 } else if (VALID_EVTCHN(evtchn)) {
582 unmask_evtchn(evtchn);
583 pirq_unmask_notify(irq);
587 static int find_irq_by_gsi(unsigned gsi)
591 for (irq = 0; irq < nr_irqs; irq++) {
592 struct irq_info *info = info_for_irq(irq);
594 if (info == NULL || info->type != IRQT_PIRQ)
597 if (gsi_from_irq(irq) == gsi)
604 int xen_allocate_pirq(unsigned gsi, int shareable, char *name)
606 return xen_map_pirq_gsi(gsi, gsi, shareable, name);
609 /* xen_map_pirq_gsi might allocate irqs from the top down, as a
610 * consequence don't assume that the irq number returned has a low value
611 * or can be used as a pirq number unless you know otherwise.
613 * One notable exception is when xen_map_pirq_gsi is called passing an
614 * hardware gsi as argument, in that case the irq number returned
615 * matches the gsi number passed as second argument.
617 * Note: We don't assign an event channel until the irq actually started
618 * up. Return an existing irq if we've already got one for the gsi.
620 int xen_map_pirq_gsi(unsigned pirq, unsigned gsi, int shareable, char *name)
623 struct physdev_irq irq_op;
625 spin_lock(&irq_mapping_update_lock);
627 if ((pirq > nr_irqs) || (gsi > nr_irqs)) {
628 printk(KERN_WARNING "xen_map_pirq_gsi: %s %s is incorrect!\n",
629 pirq > nr_irqs ? "pirq" :"",
630 gsi > nr_irqs ? "gsi" : "");
634 irq = find_irq_by_gsi(gsi);
636 printk(KERN_INFO "xen_map_pirq_gsi: returning irq %d for gsi %u\n",
638 goto out; /* XXX need refcount? */
641 /* If we are a PV guest, we don't have GSIs (no ACPI passed). Therefore
642 * we are using the !xen_initial_domain() to drop in the function.*/
643 if (identity_mapped_irq(gsi) || (!xen_initial_domain() &&
646 irq_alloc_desc_at(irq, -1);
648 irq = find_unbound_irq();
650 set_irq_chip_and_handler_name(irq, &xen_pirq_chip,
651 handle_level_irq, name);
656 /* Only the privileged domain can do this. For non-priv, the pcifront
657 * driver provides a PCI bus that does the call to do exactly
658 * this in the priv domain. */
659 if (xen_initial_domain() &&
660 HYPERVISOR_physdev_op(PHYSDEVOP_alloc_irq_vector, &irq_op)) {
666 irq_info[irq] = mk_pirq_info(0, pirq, gsi, irq_op.vector);
667 irq_info[irq].u.pirq.flags |= shareable ? PIRQ_SHAREABLE : 0;
668 pirq_to_irq[pirq] = irq;
671 spin_unlock(&irq_mapping_update_lock);
676 #ifdef CONFIG_PCI_MSI
677 #include <linux/msi.h>
678 #include "../pci/msi.h"
680 void xen_allocate_pirq_msi(char *name, int *irq, int *pirq, int alloc)
682 spin_lock(&irq_mapping_update_lock);
684 if (alloc & XEN_ALLOC_IRQ) {
685 *irq = find_unbound_irq();
690 if (alloc & XEN_ALLOC_PIRQ) {
691 *pirq = find_unbound_pirq(MAP_PIRQ_TYPE_MSI);
696 set_irq_chip_and_handler_name(*irq, &xen_pirq_chip,
697 handle_level_irq, name);
699 irq_info[*irq] = mk_pirq_info(0, *pirq, 0, 0);
700 pirq_to_irq[*pirq] = *irq;
703 spin_unlock(&irq_mapping_update_lock);
706 int xen_create_msi_irq(struct pci_dev *dev, struct msi_desc *msidesc, int type)
709 struct physdev_map_pirq map_irq;
712 u32 table_offset, bir;
714 memset(&map_irq, 0, sizeof(map_irq));
715 map_irq.domid = DOMID_SELF;
716 map_irq.type = MAP_PIRQ_TYPE_MSI;
719 map_irq.bus = dev->bus->number;
720 map_irq.devfn = dev->devfn;
722 if (type == PCI_CAP_ID_MSIX) {
723 pos = pci_find_capability(dev, PCI_CAP_ID_MSIX);
725 pci_read_config_dword(dev, msix_table_offset_reg(pos),
727 bir = (u8)(table_offset & PCI_MSIX_FLAGS_BIRMASK);
729 map_irq.table_base = pci_resource_start(dev, bir);
730 map_irq.entry_nr = msidesc->msi_attrib.entry_nr;
733 spin_lock(&irq_mapping_update_lock);
735 irq = find_unbound_irq();
740 rc = HYPERVISOR_physdev_op(PHYSDEVOP_map_pirq, &map_irq);
742 printk(KERN_WARNING "xen map irq failed %d\n", rc);
749 irq_info[irq] = mk_pirq_info(0, map_irq.pirq, 0, map_irq.index);
751 set_irq_chip_and_handler_name(irq, &xen_pirq_chip,
753 (type == PCI_CAP_ID_MSIX) ? "msi-x":"msi");
756 spin_unlock(&irq_mapping_update_lock);
761 int xen_destroy_irq(int irq)
763 struct irq_desc *desc;
764 struct physdev_unmap_pirq unmap_irq;
765 struct irq_info *info = info_for_irq(irq);
768 spin_lock(&irq_mapping_update_lock);
770 desc = irq_to_desc(irq);
774 if (xen_initial_domain()) {
775 unmap_irq.pirq = info->u.pirq.pirq;
776 unmap_irq.domid = DOMID_SELF;
777 rc = HYPERVISOR_physdev_op(PHYSDEVOP_unmap_pirq, &unmap_irq);
779 printk(KERN_WARNING "unmap irq failed %d\n", rc);
782 pirq_to_irq[info->u.pirq.pirq] = -1;
784 irq_info[irq] = mk_unbound_info();
789 spin_unlock(&irq_mapping_update_lock);
793 int xen_vector_from_irq(unsigned irq)
795 return vector_from_irq(irq);
798 int xen_gsi_from_irq(unsigned irq)
800 return gsi_from_irq(irq);
803 int xen_irq_from_pirq(unsigned pirq)
805 return pirq_to_irq[pirq];
808 int bind_evtchn_to_irq(unsigned int evtchn)
812 spin_lock(&irq_mapping_update_lock);
814 irq = evtchn_to_irq[evtchn];
817 irq = find_unbound_irq();
819 set_irq_chip_and_handler_name(irq, &xen_dynamic_chip,
820 handle_fasteoi_irq, "event");
822 evtchn_to_irq[evtchn] = irq;
823 irq_info[irq] = mk_evtchn_info(evtchn);
826 spin_unlock(&irq_mapping_update_lock);
830 EXPORT_SYMBOL_GPL(bind_evtchn_to_irq);
832 static int bind_ipi_to_irq(unsigned int ipi, unsigned int cpu)
834 struct evtchn_bind_ipi bind_ipi;
837 spin_lock(&irq_mapping_update_lock);
839 irq = per_cpu(ipi_to_irq, cpu)[ipi];
842 irq = find_unbound_irq();
846 set_irq_chip_and_handler_name(irq, &xen_percpu_chip,
847 handle_percpu_irq, "ipi");
850 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_ipi,
853 evtchn = bind_ipi.port;
855 evtchn_to_irq[evtchn] = irq;
856 irq_info[irq] = mk_ipi_info(evtchn, ipi);
857 per_cpu(ipi_to_irq, cpu)[ipi] = irq;
859 bind_evtchn_to_cpu(evtchn, cpu);
863 spin_unlock(&irq_mapping_update_lock);
868 int bind_virq_to_irq(unsigned int virq, unsigned int cpu)
870 struct evtchn_bind_virq bind_virq;
873 spin_lock(&irq_mapping_update_lock);
875 irq = per_cpu(virq_to_irq, cpu)[virq];
878 irq = find_unbound_irq();
880 set_irq_chip_and_handler_name(irq, &xen_percpu_chip,
881 handle_percpu_irq, "virq");
883 bind_virq.virq = virq;
884 bind_virq.vcpu = cpu;
885 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_virq,
888 evtchn = bind_virq.port;
890 evtchn_to_irq[evtchn] = irq;
891 irq_info[irq] = mk_virq_info(evtchn, virq);
893 per_cpu(virq_to_irq, cpu)[virq] = irq;
895 bind_evtchn_to_cpu(evtchn, cpu);
898 spin_unlock(&irq_mapping_update_lock);
903 static void unbind_from_irq(unsigned int irq)
905 struct evtchn_close close;
906 int evtchn = evtchn_from_irq(irq);
908 spin_lock(&irq_mapping_update_lock);
910 if (VALID_EVTCHN(evtchn)) {
912 if (HYPERVISOR_event_channel_op(EVTCHNOP_close, &close) != 0)
915 switch (type_from_irq(irq)) {
917 per_cpu(virq_to_irq, cpu_from_evtchn(evtchn))
918 [virq_from_irq(irq)] = -1;
921 per_cpu(ipi_to_irq, cpu_from_evtchn(evtchn))
922 [ipi_from_irq(irq)] = -1;
928 /* Closed ports are implicitly re-bound to VCPU0. */
929 bind_evtchn_to_cpu(evtchn, 0);
931 evtchn_to_irq[evtchn] = -1;
934 if (irq_info[irq].type != IRQT_UNBOUND) {
935 irq_info[irq] = mk_unbound_info();
940 spin_unlock(&irq_mapping_update_lock);
943 int bind_evtchn_to_irqhandler(unsigned int evtchn,
944 irq_handler_t handler,
945 unsigned long irqflags,
946 const char *devname, void *dev_id)
951 irq = bind_evtchn_to_irq(evtchn);
952 retval = request_irq(irq, handler, irqflags, devname, dev_id);
954 unbind_from_irq(irq);
960 EXPORT_SYMBOL_GPL(bind_evtchn_to_irqhandler);
962 int bind_virq_to_irqhandler(unsigned int virq, unsigned int cpu,
963 irq_handler_t handler,
964 unsigned long irqflags, const char *devname, void *dev_id)
969 irq = bind_virq_to_irq(virq, cpu);
970 retval = request_irq(irq, handler, irqflags, devname, dev_id);
972 unbind_from_irq(irq);
978 EXPORT_SYMBOL_GPL(bind_virq_to_irqhandler);
980 int bind_ipi_to_irqhandler(enum ipi_vector ipi,
982 irq_handler_t handler,
983 unsigned long irqflags,
989 irq = bind_ipi_to_irq(ipi, cpu);
993 irqflags |= IRQF_NO_SUSPEND;
994 retval = request_irq(irq, handler, irqflags, devname, dev_id);
996 unbind_from_irq(irq);
1003 void unbind_from_irqhandler(unsigned int irq, void *dev_id)
1005 free_irq(irq, dev_id);
1006 unbind_from_irq(irq);
1008 EXPORT_SYMBOL_GPL(unbind_from_irqhandler);
1010 void xen_send_IPI_one(unsigned int cpu, enum ipi_vector vector)
1012 int irq = per_cpu(ipi_to_irq, cpu)[vector];
1014 notify_remote_via_irq(irq);
1017 irqreturn_t xen_debug_interrupt(int irq, void *dev_id)
1019 struct shared_info *sh = HYPERVISOR_shared_info;
1020 int cpu = smp_processor_id();
1021 unsigned long *cpu_evtchn = cpu_evtchn_mask(cpu);
1023 unsigned long flags;
1024 static DEFINE_SPINLOCK(debug_lock);
1025 struct vcpu_info *v;
1027 spin_lock_irqsave(&debug_lock, flags);
1029 printk("\nvcpu %d\n ", cpu);
1031 for_each_online_cpu(i) {
1033 v = per_cpu(xen_vcpu, i);
1034 pending = (get_irq_regs() && i == cpu)
1035 ? xen_irqs_disabled(get_irq_regs())
1036 : v->evtchn_upcall_mask;
1037 printk("%d: masked=%d pending=%d event_sel %0*lx\n ", i,
1038 pending, v->evtchn_upcall_pending,
1039 (int)(sizeof(v->evtchn_pending_sel)*2),
1040 v->evtchn_pending_sel);
1042 v = per_cpu(xen_vcpu, cpu);
1044 printk("\npending:\n ");
1045 for (i = ARRAY_SIZE(sh->evtchn_pending)-1; i >= 0; i--)
1046 printk("%0*lx%s", (int)sizeof(sh->evtchn_pending[0])*2,
1047 sh->evtchn_pending[i],
1048 i % 8 == 0 ? "\n " : " ");
1049 printk("\nglobal mask:\n ");
1050 for (i = ARRAY_SIZE(sh->evtchn_mask)-1; i >= 0; i--)
1052 (int)(sizeof(sh->evtchn_mask[0])*2),
1054 i % 8 == 0 ? "\n " : " ");
1056 printk("\nglobally unmasked:\n ");
1057 for (i = ARRAY_SIZE(sh->evtchn_mask)-1; i >= 0; i--)
1058 printk("%0*lx%s", (int)(sizeof(sh->evtchn_mask[0])*2),
1059 sh->evtchn_pending[i] & ~sh->evtchn_mask[i],
1060 i % 8 == 0 ? "\n " : " ");
1062 printk("\nlocal cpu%d mask:\n ", cpu);
1063 for (i = (NR_EVENT_CHANNELS/BITS_PER_LONG)-1; i >= 0; i--)
1064 printk("%0*lx%s", (int)(sizeof(cpu_evtchn[0])*2),
1066 i % 8 == 0 ? "\n " : " ");
1068 printk("\nlocally unmasked:\n ");
1069 for (i = ARRAY_SIZE(sh->evtchn_mask)-1; i >= 0; i--) {
1070 unsigned long pending = sh->evtchn_pending[i]
1071 & ~sh->evtchn_mask[i]
1073 printk("%0*lx%s", (int)(sizeof(sh->evtchn_mask[0])*2),
1074 pending, i % 8 == 0 ? "\n " : " ");
1077 printk("\npending list:\n");
1078 for (i = 0; i < NR_EVENT_CHANNELS; i++) {
1079 if (sync_test_bit(i, sh->evtchn_pending)) {
1080 int word_idx = i / BITS_PER_LONG;
1081 printk(" %d: event %d -> irq %d%s%s%s\n",
1082 cpu_from_evtchn(i), i,
1084 sync_test_bit(word_idx, &v->evtchn_pending_sel)
1086 !sync_test_bit(i, sh->evtchn_mask)
1087 ? "" : " globally-masked",
1088 sync_test_bit(i, cpu_evtchn)
1089 ? "" : " locally-masked");
1093 spin_unlock_irqrestore(&debug_lock, flags);
1098 static DEFINE_PER_CPU(unsigned, xed_nesting_count);
1101 * Search the CPUs pending events bitmasks. For each one found, map
1102 * the event number to an irq, and feed it into do_IRQ() for
1105 * Xen uses a two-level bitmap to speed searching. The first level is
1106 * a bitset of words which contain pending event bits. The second
1107 * level is a bitset of pending events themselves.
1109 static void __xen_evtchn_do_upcall(void)
1111 int cpu = get_cpu();
1112 struct shared_info *s = HYPERVISOR_shared_info;
1113 struct vcpu_info *vcpu_info = __this_cpu_read(xen_vcpu);
1117 unsigned long pending_words;
1119 vcpu_info->evtchn_upcall_pending = 0;
1121 if (__this_cpu_inc_return(xed_nesting_count) - 1)
1124 #ifndef CONFIG_X86 /* No need for a barrier -- XCHG is a barrier on x86. */
1125 /* Clear master flag /before/ clearing selector flag. */
1128 pending_words = xchg(&vcpu_info->evtchn_pending_sel, 0);
1129 while (pending_words != 0) {
1130 unsigned long pending_bits;
1131 int word_idx = __ffs(pending_words);
1132 pending_words &= ~(1UL << word_idx);
1134 while ((pending_bits = active_evtchns(cpu, s, word_idx)) != 0) {
1135 int bit_idx = __ffs(pending_bits);
1136 int port = (word_idx * BITS_PER_LONG) + bit_idx;
1137 int irq = evtchn_to_irq[port];
1138 struct irq_desc *desc;
1144 desc = irq_to_desc(irq);
1146 generic_handle_irq_desc(irq, desc);
1151 BUG_ON(!irqs_disabled());
1153 count = __this_cpu_read(xed_nesting_count);
1154 __this_cpu_write(xed_nesting_count, 0);
1155 } while (count != 1 || vcpu_info->evtchn_upcall_pending);
1162 void xen_evtchn_do_upcall(struct pt_regs *regs)
1164 struct pt_regs *old_regs = set_irq_regs(regs);
1169 __xen_evtchn_do_upcall();
1172 set_irq_regs(old_regs);
1175 void xen_hvm_evtchn_do_upcall(void)
1177 __xen_evtchn_do_upcall();
1179 EXPORT_SYMBOL_GPL(xen_hvm_evtchn_do_upcall);
1181 /* Rebind a new event channel to an existing irq. */
1182 void rebind_evtchn_irq(int evtchn, int irq)
1184 struct irq_info *info = info_for_irq(irq);
1186 /* Make sure the irq is masked, since the new event channel
1187 will also be masked. */
1190 spin_lock(&irq_mapping_update_lock);
1192 /* After resume the irq<->evtchn mappings are all cleared out */
1193 BUG_ON(evtchn_to_irq[evtchn] != -1);
1194 /* Expect irq to have been bound before,
1195 so there should be a proper type */
1196 BUG_ON(info->type == IRQT_UNBOUND);
1198 evtchn_to_irq[evtchn] = irq;
1199 irq_info[irq] = mk_evtchn_info(evtchn);
1201 spin_unlock(&irq_mapping_update_lock);
1203 /* new event channels are always bound to cpu 0 */
1204 irq_set_affinity(irq, cpumask_of(0));
1206 /* Unmask the event channel. */
1210 /* Rebind an evtchn so that it gets delivered to a specific cpu */
1211 static int rebind_irq_to_cpu(unsigned irq, unsigned tcpu)
1213 struct evtchn_bind_vcpu bind_vcpu;
1214 int evtchn = evtchn_from_irq(irq);
1216 /* events delivered via platform PCI interrupts are always
1217 * routed to vcpu 0 */
1218 if (!VALID_EVTCHN(evtchn) ||
1219 (xen_hvm_domain() && !xen_have_vector_callback))
1222 /* Send future instances of this interrupt to other vcpu. */
1223 bind_vcpu.port = evtchn;
1224 bind_vcpu.vcpu = tcpu;
1227 * If this fails, it usually just indicates that we're dealing with a
1228 * virq or IPI channel, which don't actually need to be rebound. Ignore
1229 * it, but don't do the xenlinux-level rebind in that case.
1231 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_vcpu, &bind_vcpu) >= 0)
1232 bind_evtchn_to_cpu(evtchn, tcpu);
1237 static int set_affinity_irq(unsigned irq, const struct cpumask *dest)
1239 unsigned tcpu = cpumask_first(dest);
1241 return rebind_irq_to_cpu(irq, tcpu);
1244 int resend_irq_on_evtchn(unsigned int irq)
1246 int masked, evtchn = evtchn_from_irq(irq);
1247 struct shared_info *s = HYPERVISOR_shared_info;
1249 if (!VALID_EVTCHN(evtchn))
1252 masked = sync_test_and_set_bit(evtchn, s->evtchn_mask);
1253 sync_set_bit(evtchn, s->evtchn_pending);
1255 unmask_evtchn(evtchn);
1260 static void enable_dynirq(unsigned int irq)
1262 int evtchn = evtchn_from_irq(irq);
1264 if (VALID_EVTCHN(evtchn))
1265 unmask_evtchn(evtchn);
1268 static void disable_dynirq(unsigned int irq)
1270 int evtchn = evtchn_from_irq(irq);
1272 if (VALID_EVTCHN(evtchn))
1273 mask_evtchn(evtchn);
1276 static void ack_dynirq(unsigned int irq)
1278 int evtchn = evtchn_from_irq(irq);
1280 move_masked_irq(irq);
1282 if (VALID_EVTCHN(evtchn))
1283 unmask_evtchn(evtchn);
1286 static int retrigger_dynirq(unsigned int irq)
1288 int evtchn = evtchn_from_irq(irq);
1289 struct shared_info *sh = HYPERVISOR_shared_info;
1292 if (VALID_EVTCHN(evtchn)) {
1295 masked = sync_test_and_set_bit(evtchn, sh->evtchn_mask);
1296 sync_set_bit(evtchn, sh->evtchn_pending);
1298 unmask_evtchn(evtchn);
1305 static void restore_cpu_pirqs(void)
1307 int pirq, rc, irq, gsi;
1308 struct physdev_map_pirq map_irq;
1310 for (pirq = 0; pirq < nr_irqs; pirq++) {
1311 irq = pirq_to_irq[pirq];
1315 /* save/restore of PT devices doesn't work, so at this point the
1316 * only devices present are GSI based emulated devices */
1317 gsi = gsi_from_irq(irq);
1321 map_irq.domid = DOMID_SELF;
1322 map_irq.type = MAP_PIRQ_TYPE_GSI;
1323 map_irq.index = gsi;
1324 map_irq.pirq = pirq;
1326 rc = HYPERVISOR_physdev_op(PHYSDEVOP_map_pirq, &map_irq);
1328 printk(KERN_WARNING "xen map irq failed gsi=%d irq=%d pirq=%d rc=%d\n",
1329 gsi, irq, pirq, rc);
1330 irq_info[irq] = mk_unbound_info();
1331 pirq_to_irq[pirq] = -1;
1335 printk(KERN_DEBUG "xen: --> irq=%d, pirq=%d\n", irq, map_irq.pirq);
1341 static void restore_cpu_virqs(unsigned int cpu)
1343 struct evtchn_bind_virq bind_virq;
1344 int virq, irq, evtchn;
1346 for (virq = 0; virq < NR_VIRQS; virq++) {
1347 if ((irq = per_cpu(virq_to_irq, cpu)[virq]) == -1)
1350 BUG_ON(virq_from_irq(irq) != virq);
1352 /* Get a new binding from Xen. */
1353 bind_virq.virq = virq;
1354 bind_virq.vcpu = cpu;
1355 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_virq,
1358 evtchn = bind_virq.port;
1360 /* Record the new mapping. */
1361 evtchn_to_irq[evtchn] = irq;
1362 irq_info[irq] = mk_virq_info(evtchn, virq);
1363 bind_evtchn_to_cpu(evtchn, cpu);
1367 static void restore_cpu_ipis(unsigned int cpu)
1369 struct evtchn_bind_ipi bind_ipi;
1370 int ipi, irq, evtchn;
1372 for (ipi = 0; ipi < XEN_NR_IPIS; ipi++) {
1373 if ((irq = per_cpu(ipi_to_irq, cpu)[ipi]) == -1)
1376 BUG_ON(ipi_from_irq(irq) != ipi);
1378 /* Get a new binding from Xen. */
1379 bind_ipi.vcpu = cpu;
1380 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_ipi,
1383 evtchn = bind_ipi.port;
1385 /* Record the new mapping. */
1386 evtchn_to_irq[evtchn] = irq;
1387 irq_info[irq] = mk_ipi_info(evtchn, ipi);
1388 bind_evtchn_to_cpu(evtchn, cpu);
1392 /* Clear an irq's pending state, in preparation for polling on it */
1393 void xen_clear_irq_pending(int irq)
1395 int evtchn = evtchn_from_irq(irq);
1397 if (VALID_EVTCHN(evtchn))
1398 clear_evtchn(evtchn);
1400 EXPORT_SYMBOL(xen_clear_irq_pending);
1401 void xen_set_irq_pending(int irq)
1403 int evtchn = evtchn_from_irq(irq);
1405 if (VALID_EVTCHN(evtchn))
1409 bool xen_test_irq_pending(int irq)
1411 int evtchn = evtchn_from_irq(irq);
1414 if (VALID_EVTCHN(evtchn))
1415 ret = test_evtchn(evtchn);
1420 /* Poll waiting for an irq to become pending with timeout. In the usual case,
1421 * the irq will be disabled so it won't deliver an interrupt. */
1422 void xen_poll_irq_timeout(int irq, u64 timeout)
1424 evtchn_port_t evtchn = evtchn_from_irq(irq);
1426 if (VALID_EVTCHN(evtchn)) {
1427 struct sched_poll poll;
1430 poll.timeout = timeout;
1431 set_xen_guest_handle(poll.ports, &evtchn);
1433 if (HYPERVISOR_sched_op(SCHEDOP_poll, &poll) != 0)
1437 EXPORT_SYMBOL(xen_poll_irq_timeout);
1438 /* Poll waiting for an irq to become pending. In the usual case, the
1439 * irq will be disabled so it won't deliver an interrupt. */
1440 void xen_poll_irq(int irq)
1442 xen_poll_irq_timeout(irq, 0 /* no timeout */);
1445 void xen_irq_resume(void)
1447 unsigned int cpu, irq, evtchn;
1448 struct irq_desc *desc;
1450 init_evtchn_cpu_bindings();
1452 /* New event-channel space is not 'live' yet. */
1453 for (evtchn = 0; evtchn < NR_EVENT_CHANNELS; evtchn++)
1454 mask_evtchn(evtchn);
1456 /* No IRQ <-> event-channel mappings. */
1457 for (irq = 0; irq < nr_irqs; irq++)
1458 irq_info[irq].evtchn = 0; /* zap event-channel binding */
1460 for (evtchn = 0; evtchn < NR_EVENT_CHANNELS; evtchn++)
1461 evtchn_to_irq[evtchn] = -1;
1463 for_each_possible_cpu(cpu) {
1464 restore_cpu_virqs(cpu);
1465 restore_cpu_ipis(cpu);
1469 * Unmask any IRQF_NO_SUSPEND IRQs which are enabled. These
1470 * are not handled by the IRQ core.
1472 for_each_irq_desc(irq, desc) {
1473 if (!desc->action || !(desc->action->flags & IRQF_NO_SUSPEND))
1475 if (desc->status & IRQ_DISABLED)
1478 evtchn = evtchn_from_irq(irq);
1482 unmask_evtchn(evtchn);
1485 restore_cpu_pirqs();
1488 static struct irq_chip xen_dynamic_chip __read_mostly = {
1491 .disable = disable_dynirq,
1492 .mask = disable_dynirq,
1493 .unmask = enable_dynirq,
1496 .set_affinity = set_affinity_irq,
1497 .retrigger = retrigger_dynirq,
1500 static struct irq_chip xen_pirq_chip __read_mostly = {
1503 .startup = startup_pirq,
1504 .shutdown = shutdown_pirq,
1506 .enable = enable_pirq,
1507 .unmask = enable_pirq,
1509 .disable = disable_pirq,
1510 .mask = disable_pirq,
1515 .set_affinity = set_affinity_irq,
1517 .retrigger = retrigger_dynirq,
1520 static struct irq_chip xen_percpu_chip __read_mostly = {
1521 .name = "xen-percpu",
1523 .disable = disable_dynirq,
1524 .mask = disable_dynirq,
1525 .unmask = enable_dynirq,
1530 int xen_set_callback_via(uint64_t via)
1532 struct xen_hvm_param a;
1533 a.domid = DOMID_SELF;
1534 a.index = HVM_PARAM_CALLBACK_IRQ;
1536 return HYPERVISOR_hvm_op(HVMOP_set_param, &a);
1538 EXPORT_SYMBOL_GPL(xen_set_callback_via);
1540 #ifdef CONFIG_XEN_PVHVM
1541 /* Vector callbacks are better than PCI interrupts to receive event
1542 * channel notifications because we can receive vector callbacks on any
1543 * vcpu and we don't need PCI support or APIC interactions. */
1544 void xen_callback_vector(void)
1547 uint64_t callback_via;
1548 if (xen_have_vector_callback) {
1549 callback_via = HVM_CALLBACK_VECTOR(XEN_HVM_EVTCHN_CALLBACK);
1550 rc = xen_set_callback_via(callback_via);
1552 printk(KERN_ERR "Request for Xen HVM callback vector"
1554 xen_have_vector_callback = 0;
1557 printk(KERN_INFO "Xen HVM callback vector for event delivery is "
1559 /* in the restore case the vector has already been allocated */
1560 if (!test_bit(XEN_HVM_EVTCHN_CALLBACK, used_vectors))
1561 alloc_intr_gate(XEN_HVM_EVTCHN_CALLBACK, xen_hvm_callback_vector);
1565 void xen_callback_vector(void) {}
1568 void __init xen_init_IRQ(void)
1572 cpu_evtchn_mask_p = kcalloc(nr_cpu_ids, sizeof(struct cpu_evtchn_s),
1574 irq_info = kcalloc(nr_irqs, sizeof(*irq_info), GFP_KERNEL);
1576 /* We are using nr_irqs as the maximum number of pirq available but
1577 * that number is actually chosen by Xen and we don't know exactly
1578 * what it is. Be careful choosing high pirq numbers. */
1579 pirq_to_irq = kcalloc(nr_irqs, sizeof(*pirq_to_irq), GFP_KERNEL);
1580 for (i = 0; i < nr_irqs; i++)
1581 pirq_to_irq[i] = -1;
1583 evtchn_to_irq = kcalloc(NR_EVENT_CHANNELS, sizeof(*evtchn_to_irq),
1585 for (i = 0; i < NR_EVENT_CHANNELS; i++)
1586 evtchn_to_irq[i] = -1;
1588 init_evtchn_cpu_bindings();
1590 /* No event channels are 'live' right now. */
1591 for (i = 0; i < NR_EVENT_CHANNELS; i++)
1594 if (xen_hvm_domain()) {
1595 xen_callback_vector();
1597 /* pci_xen_hvm_init must be called after native_init_IRQ so that
1598 * __acpi_register_gsi can point at the right function */
1601 irq_ctx_init(smp_processor_id());
1602 if (xen_initial_domain())