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 return info_for_irq(irq)->evtchn;
176 unsigned irq_from_evtchn(unsigned int evtchn)
178 return evtchn_to_irq[evtchn];
180 EXPORT_SYMBOL_GPL(irq_from_evtchn);
182 static enum ipi_vector ipi_from_irq(unsigned irq)
184 struct irq_info *info = info_for_irq(irq);
186 BUG_ON(info == NULL);
187 BUG_ON(info->type != IRQT_IPI);
192 static unsigned virq_from_irq(unsigned irq)
194 struct irq_info *info = info_for_irq(irq);
196 BUG_ON(info == NULL);
197 BUG_ON(info->type != IRQT_VIRQ);
202 static unsigned pirq_from_irq(unsigned irq)
204 struct irq_info *info = info_for_irq(irq);
206 BUG_ON(info == NULL);
207 BUG_ON(info->type != IRQT_PIRQ);
209 return info->u.pirq.pirq;
212 static unsigned gsi_from_irq(unsigned irq)
214 struct irq_info *info = info_for_irq(irq);
216 BUG_ON(info == NULL);
217 BUG_ON(info->type != IRQT_PIRQ);
219 return info->u.pirq.gsi;
222 static unsigned vector_from_irq(unsigned irq)
224 struct irq_info *info = info_for_irq(irq);
226 BUG_ON(info == NULL);
227 BUG_ON(info->type != IRQT_PIRQ);
229 return info->u.pirq.vector;
232 static enum xen_irq_type type_from_irq(unsigned irq)
234 return info_for_irq(irq)->type;
237 static unsigned cpu_from_irq(unsigned irq)
239 return info_for_irq(irq)->cpu;
242 static unsigned int cpu_from_evtchn(unsigned int evtchn)
244 int irq = evtchn_to_irq[evtchn];
248 ret = cpu_from_irq(irq);
253 static bool pirq_needs_eoi(unsigned irq)
255 struct irq_info *info = info_for_irq(irq);
257 BUG_ON(info->type != IRQT_PIRQ);
259 return info->u.pirq.flags & PIRQ_NEEDS_EOI;
262 static inline unsigned long active_evtchns(unsigned int cpu,
263 struct shared_info *sh,
266 return (sh->evtchn_pending[idx] &
267 cpu_evtchn_mask(cpu)[idx] &
268 ~sh->evtchn_mask[idx]);
271 static void bind_evtchn_to_cpu(unsigned int chn, unsigned int cpu)
273 int irq = evtchn_to_irq[chn];
277 cpumask_copy(irq_to_desc(irq)->affinity, cpumask_of(cpu));
280 clear_bit(chn, cpu_evtchn_mask(cpu_from_irq(irq)));
281 set_bit(chn, cpu_evtchn_mask(cpu));
283 irq_info[irq].cpu = cpu;
286 static void init_evtchn_cpu_bindings(void)
290 struct irq_desc *desc;
292 /* By default all event channels notify CPU#0. */
293 for_each_irq_desc(i, desc) {
294 cpumask_copy(desc->affinity, cpumask_of(0));
298 for_each_possible_cpu(i)
299 memset(cpu_evtchn_mask(i),
300 (i == 0) ? ~0 : 0, sizeof(struct cpu_evtchn_s));
304 static inline void clear_evtchn(int port)
306 struct shared_info *s = HYPERVISOR_shared_info;
307 sync_clear_bit(port, &s->evtchn_pending[0]);
310 static inline void set_evtchn(int port)
312 struct shared_info *s = HYPERVISOR_shared_info;
313 sync_set_bit(port, &s->evtchn_pending[0]);
316 static inline int test_evtchn(int port)
318 struct shared_info *s = HYPERVISOR_shared_info;
319 return sync_test_bit(port, &s->evtchn_pending[0]);
324 * notify_remote_via_irq - send event to remote end of event channel via irq
325 * @irq: irq of event channel to send event to
327 * Unlike notify_remote_via_evtchn(), this is safe to use across
328 * save/restore. Notifications on a broken connection are silently
331 void notify_remote_via_irq(int irq)
333 int evtchn = evtchn_from_irq(irq);
335 if (VALID_EVTCHN(evtchn))
336 notify_remote_via_evtchn(evtchn);
338 EXPORT_SYMBOL_GPL(notify_remote_via_irq);
340 static void mask_evtchn(int port)
342 struct shared_info *s = HYPERVISOR_shared_info;
343 sync_set_bit(port, &s->evtchn_mask[0]);
346 static void unmask_evtchn(int port)
348 struct shared_info *s = HYPERVISOR_shared_info;
349 unsigned int cpu = get_cpu();
351 BUG_ON(!irqs_disabled());
353 /* Slow path (hypercall) if this is a non-local port. */
354 if (unlikely(cpu != cpu_from_evtchn(port))) {
355 struct evtchn_unmask unmask = { .port = port };
356 (void)HYPERVISOR_event_channel_op(EVTCHNOP_unmask, &unmask);
358 struct vcpu_info *vcpu_info = __get_cpu_var(xen_vcpu);
360 sync_clear_bit(port, &s->evtchn_mask[0]);
363 * The following is basically the equivalent of
364 * 'hw_resend_irq'. Just like a real IO-APIC we 'lose
365 * the interrupt edge' if the channel is masked.
367 if (sync_test_bit(port, &s->evtchn_pending[0]) &&
368 !sync_test_and_set_bit(port / BITS_PER_LONG,
369 &vcpu_info->evtchn_pending_sel))
370 vcpu_info->evtchn_upcall_pending = 1;
376 static int get_nr_hw_irqs(void)
380 #ifdef CONFIG_X86_IO_APIC
381 ret = get_nr_irqs_gsi();
387 static int find_unbound_pirq(int type)
390 struct physdev_get_free_pirq op_get_free_pirq;
391 op_get_free_pirq.type = type;
393 rc = HYPERVISOR_physdev_op(PHYSDEVOP_get_free_pirq, &op_get_free_pirq);
395 return op_get_free_pirq.pirq;
397 for (i = 0; i < nr_irqs; i++) {
398 if (pirq_to_irq[i] < 0)
404 static int find_unbound_irq(void)
406 struct irq_data *data;
408 int start = get_nr_hw_irqs();
410 if (start == nr_irqs)
413 /* nr_irqs is a magic value. Must not use it.*/
414 for (irq = nr_irqs-1; irq > start; irq--) {
415 data = irq_get_irq_data(irq);
416 /* only 0->15 have init'd desc; handle irq > 16 */
419 if (data->chip == &no_irq_chip)
421 if (data->chip != &xen_dynamic_chip)
423 if (irq_info[irq].type == IRQT_UNBOUND)
430 res = irq_alloc_desc_at(irq, 0);
432 if (WARN_ON(res != irq))
438 panic("No available IRQ to bind to: increase nr_irqs!\n");
441 static bool identity_mapped_irq(unsigned irq)
443 /* identity map all the hardware irqs */
444 return irq < get_nr_hw_irqs();
447 static void pirq_unmask_notify(int irq)
449 struct physdev_eoi eoi = { .irq = pirq_from_irq(irq) };
451 if (unlikely(pirq_needs_eoi(irq))) {
452 int rc = HYPERVISOR_physdev_op(PHYSDEVOP_eoi, &eoi);
457 static void pirq_query_unmask(int irq)
459 struct physdev_irq_status_query irq_status;
460 struct irq_info *info = info_for_irq(irq);
462 BUG_ON(info->type != IRQT_PIRQ);
464 irq_status.irq = pirq_from_irq(irq);
465 if (HYPERVISOR_physdev_op(PHYSDEVOP_irq_status_query, &irq_status))
466 irq_status.flags = 0;
468 info->u.pirq.flags &= ~PIRQ_NEEDS_EOI;
469 if (irq_status.flags & XENIRQSTAT_needs_eoi)
470 info->u.pirq.flags |= PIRQ_NEEDS_EOI;
473 static bool probing_irq(int irq)
475 struct irq_desc *desc = irq_to_desc(irq);
477 return desc && desc->action == NULL;
480 static unsigned int startup_pirq(unsigned int irq)
482 struct evtchn_bind_pirq bind_pirq;
483 struct irq_info *info = info_for_irq(irq);
484 int evtchn = evtchn_from_irq(irq);
487 BUG_ON(info->type != IRQT_PIRQ);
489 if (VALID_EVTCHN(evtchn))
492 bind_pirq.pirq = pirq_from_irq(irq);
493 /* NB. We are happy to share unless we are probing. */
494 bind_pirq.flags = info->u.pirq.flags & PIRQ_SHAREABLE ?
495 BIND_PIRQ__WILL_SHARE : 0;
496 rc = HYPERVISOR_event_channel_op(EVTCHNOP_bind_pirq, &bind_pirq);
498 if (!probing_irq(irq))
499 printk(KERN_INFO "Failed to obtain physical IRQ %d\n",
503 evtchn = bind_pirq.port;
505 pirq_query_unmask(irq);
507 evtchn_to_irq[evtchn] = irq;
508 bind_evtchn_to_cpu(evtchn, 0);
509 info->evtchn = evtchn;
512 unmask_evtchn(evtchn);
513 pirq_unmask_notify(irq);
518 static void shutdown_pirq(unsigned int irq)
520 struct evtchn_close close;
521 struct irq_info *info = info_for_irq(irq);
522 int evtchn = evtchn_from_irq(irq);
524 BUG_ON(info->type != IRQT_PIRQ);
526 if (!VALID_EVTCHN(evtchn))
532 if (HYPERVISOR_event_channel_op(EVTCHNOP_close, &close) != 0)
535 bind_evtchn_to_cpu(evtchn, 0);
536 evtchn_to_irq[evtchn] = -1;
540 static void enable_pirq(unsigned int irq)
545 static void disable_pirq(unsigned int irq)
549 static void ack_pirq(unsigned int irq)
551 int evtchn = evtchn_from_irq(irq);
553 move_native_irq(irq);
555 if (VALID_EVTCHN(evtchn)) {
557 clear_evtchn(evtchn);
561 static void end_pirq(unsigned int irq)
563 int evtchn = evtchn_from_irq(irq);
564 struct irq_desc *desc = irq_to_desc(irq);
569 if ((desc->status & (IRQ_DISABLED|IRQ_PENDING)) ==
570 (IRQ_DISABLED|IRQ_PENDING)) {
572 } else if (VALID_EVTCHN(evtchn)) {
573 unmask_evtchn(evtchn);
574 pirq_unmask_notify(irq);
578 static int find_irq_by_gsi(unsigned gsi)
582 for (irq = 0; irq < nr_irqs; irq++) {
583 struct irq_info *info = info_for_irq(irq);
585 if (info == NULL || info->type != IRQT_PIRQ)
588 if (gsi_from_irq(irq) == gsi)
595 int xen_allocate_pirq(unsigned gsi, int shareable, char *name)
597 return xen_map_pirq_gsi(gsi, gsi, shareable, name);
600 /* xen_map_pirq_gsi might allocate irqs from the top down, as a
601 * consequence don't assume that the irq number returned has a low value
602 * or can be used as a pirq number unless you know otherwise.
604 * One notable exception is when xen_map_pirq_gsi is called passing an
605 * hardware gsi as argument, in that case the irq number returned
606 * matches the gsi number passed as second argument.
608 * Note: We don't assign an event channel until the irq actually started
609 * up. Return an existing irq if we've already got one for the gsi.
611 int xen_map_pirq_gsi(unsigned pirq, unsigned gsi, int shareable, char *name)
614 struct physdev_irq irq_op;
616 spin_lock(&irq_mapping_update_lock);
618 if ((pirq > nr_irqs) || (gsi > nr_irqs)) {
619 printk(KERN_WARNING "xen_map_pirq_gsi: %s %s is incorrect!\n",
620 pirq > nr_irqs ? "pirq" :"",
621 gsi > nr_irqs ? "gsi" : "");
625 irq = find_irq_by_gsi(gsi);
627 printk(KERN_INFO "xen_map_pirq_gsi: returning irq %d for gsi %u\n",
629 goto out; /* XXX need refcount? */
632 /* If we are a PV guest, we don't have GSIs (no ACPI passed). Therefore
633 * we are using the !xen_initial_domain() to drop in the function.*/
634 if (identity_mapped_irq(gsi) || (!xen_initial_domain() &&
637 irq_alloc_desc_at(irq, 0);
639 irq = find_unbound_irq();
641 set_irq_chip_and_handler_name(irq, &xen_pirq_chip,
642 handle_level_irq, name);
647 /* Only the privileged domain can do this. For non-priv, the pcifront
648 * driver provides a PCI bus that does the call to do exactly
649 * this in the priv domain. */
650 if (xen_initial_domain() &&
651 HYPERVISOR_physdev_op(PHYSDEVOP_alloc_irq_vector, &irq_op)) {
657 irq_info[irq] = mk_pirq_info(0, pirq, gsi, irq_op.vector);
658 irq_info[irq].u.pirq.flags |= shareable ? PIRQ_SHAREABLE : 0;
659 pirq_to_irq[pirq] = irq;
662 spin_unlock(&irq_mapping_update_lock);
667 #ifdef CONFIG_PCI_MSI
668 #include <linux/msi.h>
669 #include "../pci/msi.h"
671 void xen_allocate_pirq_msi(char *name, int *irq, int *pirq)
673 spin_lock(&irq_mapping_update_lock);
675 *irq = find_unbound_irq();
679 *pirq = find_unbound_pirq(MAP_PIRQ_TYPE_MSI);
683 set_irq_chip_and_handler_name(*irq, &xen_pirq_chip,
684 handle_level_irq, name);
686 irq_info[*irq] = mk_pirq_info(0, *pirq, 0, 0);
687 pirq_to_irq[*pirq] = *irq;
690 spin_unlock(&irq_mapping_update_lock);
693 int xen_create_msi_irq(struct pci_dev *dev, struct msi_desc *msidesc, int type)
696 struct physdev_map_pirq map_irq;
699 u32 table_offset, bir;
701 memset(&map_irq, 0, sizeof(map_irq));
702 map_irq.domid = DOMID_SELF;
703 map_irq.type = MAP_PIRQ_TYPE_MSI;
706 map_irq.bus = dev->bus->number;
707 map_irq.devfn = dev->devfn;
709 if (type == PCI_CAP_ID_MSIX) {
710 pos = pci_find_capability(dev, PCI_CAP_ID_MSIX);
712 pci_read_config_dword(dev, msix_table_offset_reg(pos),
714 bir = (u8)(table_offset & PCI_MSIX_FLAGS_BIRMASK);
716 map_irq.table_base = pci_resource_start(dev, bir);
717 map_irq.entry_nr = msidesc->msi_attrib.entry_nr;
720 spin_lock(&irq_mapping_update_lock);
722 irq = find_unbound_irq();
727 rc = HYPERVISOR_physdev_op(PHYSDEVOP_map_pirq, &map_irq);
729 printk(KERN_WARNING "xen map irq failed %d\n", rc);
736 irq_info[irq] = mk_pirq_info(0, map_irq.pirq, 0, map_irq.index);
738 set_irq_chip_and_handler_name(irq, &xen_pirq_chip,
740 (type == PCI_CAP_ID_MSIX) ? "msi-x":"msi");
743 spin_unlock(&irq_mapping_update_lock);
748 int xen_destroy_irq(int irq)
750 struct irq_desc *desc;
751 struct physdev_unmap_pirq unmap_irq;
752 struct irq_info *info = info_for_irq(irq);
755 spin_lock(&irq_mapping_update_lock);
757 desc = irq_to_desc(irq);
761 if (xen_initial_domain()) {
762 unmap_irq.pirq = info->u.pirq.pirq;
763 unmap_irq.domid = DOMID_SELF;
764 rc = HYPERVISOR_physdev_op(PHYSDEVOP_unmap_pirq, &unmap_irq);
766 printk(KERN_WARNING "unmap irq failed %d\n", rc);
770 irq_info[irq] = mk_unbound_info();
775 spin_unlock(&irq_mapping_update_lock);
779 int xen_vector_from_irq(unsigned irq)
781 return vector_from_irq(irq);
784 int xen_gsi_from_irq(unsigned irq)
786 return gsi_from_irq(irq);
789 int bind_evtchn_to_irq(unsigned int evtchn)
793 spin_lock(&irq_mapping_update_lock);
795 irq = evtchn_to_irq[evtchn];
798 irq = find_unbound_irq();
800 set_irq_chip_and_handler_name(irq, &xen_dynamic_chip,
801 handle_fasteoi_irq, "event");
803 evtchn_to_irq[evtchn] = irq;
804 irq_info[irq] = mk_evtchn_info(evtchn);
807 spin_unlock(&irq_mapping_update_lock);
811 EXPORT_SYMBOL_GPL(bind_evtchn_to_irq);
813 static int bind_ipi_to_irq(unsigned int ipi, unsigned int cpu)
815 struct evtchn_bind_ipi bind_ipi;
818 spin_lock(&irq_mapping_update_lock);
820 irq = per_cpu(ipi_to_irq, cpu)[ipi];
823 irq = find_unbound_irq();
827 set_irq_chip_and_handler_name(irq, &xen_percpu_chip,
828 handle_percpu_irq, "ipi");
831 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_ipi,
834 evtchn = bind_ipi.port;
836 evtchn_to_irq[evtchn] = irq;
837 irq_info[irq] = mk_ipi_info(evtchn, ipi);
838 per_cpu(ipi_to_irq, cpu)[ipi] = irq;
840 bind_evtchn_to_cpu(evtchn, cpu);
844 spin_unlock(&irq_mapping_update_lock);
849 int bind_virq_to_irq(unsigned int virq, unsigned int cpu)
851 struct evtchn_bind_virq bind_virq;
854 spin_lock(&irq_mapping_update_lock);
856 irq = per_cpu(virq_to_irq, cpu)[virq];
859 irq = find_unbound_irq();
861 set_irq_chip_and_handler_name(irq, &xen_percpu_chip,
862 handle_percpu_irq, "virq");
864 bind_virq.virq = virq;
865 bind_virq.vcpu = cpu;
866 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_virq,
869 evtchn = bind_virq.port;
871 evtchn_to_irq[evtchn] = irq;
872 irq_info[irq] = mk_virq_info(evtchn, virq);
874 per_cpu(virq_to_irq, cpu)[virq] = irq;
876 bind_evtchn_to_cpu(evtchn, cpu);
879 spin_unlock(&irq_mapping_update_lock);
884 static void unbind_from_irq(unsigned int irq)
886 struct evtchn_close close;
887 int evtchn = evtchn_from_irq(irq);
889 spin_lock(&irq_mapping_update_lock);
891 if (VALID_EVTCHN(evtchn)) {
893 if (HYPERVISOR_event_channel_op(EVTCHNOP_close, &close) != 0)
896 switch (type_from_irq(irq)) {
898 per_cpu(virq_to_irq, cpu_from_evtchn(evtchn))
899 [virq_from_irq(irq)] = -1;
902 per_cpu(ipi_to_irq, cpu_from_evtchn(evtchn))
903 [ipi_from_irq(irq)] = -1;
909 /* Closed ports are implicitly re-bound to VCPU0. */
910 bind_evtchn_to_cpu(evtchn, 0);
912 evtchn_to_irq[evtchn] = -1;
915 if (irq_info[irq].type != IRQT_UNBOUND) {
916 irq_info[irq] = mk_unbound_info();
921 spin_unlock(&irq_mapping_update_lock);
924 int bind_evtchn_to_irqhandler(unsigned int evtchn,
925 irq_handler_t handler,
926 unsigned long irqflags,
927 const char *devname, void *dev_id)
932 irq = bind_evtchn_to_irq(evtchn);
933 retval = request_irq(irq, handler, irqflags, devname, dev_id);
935 unbind_from_irq(irq);
941 EXPORT_SYMBOL_GPL(bind_evtchn_to_irqhandler);
943 int bind_virq_to_irqhandler(unsigned int virq, unsigned int cpu,
944 irq_handler_t handler,
945 unsigned long irqflags, const char *devname, void *dev_id)
950 irq = bind_virq_to_irq(virq, cpu);
951 retval = request_irq(irq, handler, irqflags, devname, dev_id);
953 unbind_from_irq(irq);
959 EXPORT_SYMBOL_GPL(bind_virq_to_irqhandler);
961 int bind_ipi_to_irqhandler(enum ipi_vector ipi,
963 irq_handler_t handler,
964 unsigned long irqflags,
970 irq = bind_ipi_to_irq(ipi, cpu);
974 irqflags |= IRQF_NO_SUSPEND;
975 retval = request_irq(irq, handler, irqflags, devname, dev_id);
977 unbind_from_irq(irq);
984 void unbind_from_irqhandler(unsigned int irq, void *dev_id)
986 free_irq(irq, dev_id);
987 unbind_from_irq(irq);
989 EXPORT_SYMBOL_GPL(unbind_from_irqhandler);
991 void xen_send_IPI_one(unsigned int cpu, enum ipi_vector vector)
993 int irq = per_cpu(ipi_to_irq, cpu)[vector];
995 notify_remote_via_irq(irq);
998 irqreturn_t xen_debug_interrupt(int irq, void *dev_id)
1000 struct shared_info *sh = HYPERVISOR_shared_info;
1001 int cpu = smp_processor_id();
1002 unsigned long *cpu_evtchn = cpu_evtchn_mask(cpu);
1004 unsigned long flags;
1005 static DEFINE_SPINLOCK(debug_lock);
1006 struct vcpu_info *v;
1008 spin_lock_irqsave(&debug_lock, flags);
1010 printk("\nvcpu %d\n ", cpu);
1012 for_each_online_cpu(i) {
1014 v = per_cpu(xen_vcpu, i);
1015 pending = (get_irq_regs() && i == cpu)
1016 ? xen_irqs_disabled(get_irq_regs())
1017 : v->evtchn_upcall_mask;
1018 printk("%d: masked=%d pending=%d event_sel %0*lx\n ", i,
1019 pending, v->evtchn_upcall_pending,
1020 (int)(sizeof(v->evtchn_pending_sel)*2),
1021 v->evtchn_pending_sel);
1023 v = per_cpu(xen_vcpu, cpu);
1025 printk("\npending:\n ");
1026 for (i = ARRAY_SIZE(sh->evtchn_pending)-1; i >= 0; i--)
1027 printk("%0*lx%s", (int)sizeof(sh->evtchn_pending[0])*2,
1028 sh->evtchn_pending[i],
1029 i % 8 == 0 ? "\n " : " ");
1030 printk("\nglobal mask:\n ");
1031 for (i = ARRAY_SIZE(sh->evtchn_mask)-1; i >= 0; i--)
1033 (int)(sizeof(sh->evtchn_mask[0])*2),
1035 i % 8 == 0 ? "\n " : " ");
1037 printk("\nglobally unmasked:\n ");
1038 for (i = ARRAY_SIZE(sh->evtchn_mask)-1; i >= 0; i--)
1039 printk("%0*lx%s", (int)(sizeof(sh->evtchn_mask[0])*2),
1040 sh->evtchn_pending[i] & ~sh->evtchn_mask[i],
1041 i % 8 == 0 ? "\n " : " ");
1043 printk("\nlocal cpu%d mask:\n ", cpu);
1044 for (i = (NR_EVENT_CHANNELS/BITS_PER_LONG)-1; i >= 0; i--)
1045 printk("%0*lx%s", (int)(sizeof(cpu_evtchn[0])*2),
1047 i % 8 == 0 ? "\n " : " ");
1049 printk("\nlocally unmasked:\n ");
1050 for (i = ARRAY_SIZE(sh->evtchn_mask)-1; i >= 0; i--) {
1051 unsigned long pending = sh->evtchn_pending[i]
1052 & ~sh->evtchn_mask[i]
1054 printk("%0*lx%s", (int)(sizeof(sh->evtchn_mask[0])*2),
1055 pending, i % 8 == 0 ? "\n " : " ");
1058 printk("\npending list:\n");
1059 for (i = 0; i < NR_EVENT_CHANNELS; i++) {
1060 if (sync_test_bit(i, sh->evtchn_pending)) {
1061 int word_idx = i / BITS_PER_LONG;
1062 printk(" %d: event %d -> irq %d%s%s%s\n",
1063 cpu_from_evtchn(i), i,
1065 sync_test_bit(word_idx, &v->evtchn_pending_sel)
1067 !sync_test_bit(i, sh->evtchn_mask)
1068 ? "" : " globally-masked",
1069 sync_test_bit(i, cpu_evtchn)
1070 ? "" : " locally-masked");
1074 spin_unlock_irqrestore(&debug_lock, flags);
1079 static DEFINE_PER_CPU(unsigned, xed_nesting_count);
1082 * Search the CPUs pending events bitmasks. For each one found, map
1083 * the event number to an irq, and feed it into do_IRQ() for
1086 * Xen uses a two-level bitmap to speed searching. The first level is
1087 * a bitset of words which contain pending event bits. The second
1088 * level is a bitset of pending events themselves.
1090 static void __xen_evtchn_do_upcall(void)
1092 int cpu = get_cpu();
1093 struct shared_info *s = HYPERVISOR_shared_info;
1094 struct vcpu_info *vcpu_info = __get_cpu_var(xen_vcpu);
1098 unsigned long pending_words;
1100 vcpu_info->evtchn_upcall_pending = 0;
1102 if (__get_cpu_var(xed_nesting_count)++)
1105 #ifndef CONFIG_X86 /* No need for a barrier -- XCHG is a barrier on x86. */
1106 /* Clear master flag /before/ clearing selector flag. */
1109 pending_words = xchg(&vcpu_info->evtchn_pending_sel, 0);
1110 while (pending_words != 0) {
1111 unsigned long pending_bits;
1112 int word_idx = __ffs(pending_words);
1113 pending_words &= ~(1UL << word_idx);
1115 while ((pending_bits = active_evtchns(cpu, s, word_idx)) != 0) {
1116 int bit_idx = __ffs(pending_bits);
1117 int port = (word_idx * BITS_PER_LONG) + bit_idx;
1118 int irq = evtchn_to_irq[port];
1119 struct irq_desc *desc;
1125 desc = irq_to_desc(irq);
1127 generic_handle_irq_desc(irq, desc);
1132 BUG_ON(!irqs_disabled());
1134 count = __get_cpu_var(xed_nesting_count);
1135 __get_cpu_var(xed_nesting_count) = 0;
1136 } while (count != 1 || vcpu_info->evtchn_upcall_pending);
1143 void xen_evtchn_do_upcall(struct pt_regs *regs)
1145 struct pt_regs *old_regs = set_irq_regs(regs);
1150 __xen_evtchn_do_upcall();
1153 set_irq_regs(old_regs);
1156 void xen_hvm_evtchn_do_upcall(void)
1158 __xen_evtchn_do_upcall();
1160 EXPORT_SYMBOL_GPL(xen_hvm_evtchn_do_upcall);
1162 /* Rebind a new event channel to an existing irq. */
1163 void rebind_evtchn_irq(int evtchn, int irq)
1165 struct irq_info *info = info_for_irq(irq);
1167 /* Make sure the irq is masked, since the new event channel
1168 will also be masked. */
1171 spin_lock(&irq_mapping_update_lock);
1173 /* After resume the irq<->evtchn mappings are all cleared out */
1174 BUG_ON(evtchn_to_irq[evtchn] != -1);
1175 /* Expect irq to have been bound before,
1176 so there should be a proper type */
1177 BUG_ON(info->type == IRQT_UNBOUND);
1179 evtchn_to_irq[evtchn] = irq;
1180 irq_info[irq] = mk_evtchn_info(evtchn);
1182 spin_unlock(&irq_mapping_update_lock);
1184 /* new event channels are always bound to cpu 0 */
1185 irq_set_affinity(irq, cpumask_of(0));
1187 /* Unmask the event channel. */
1191 /* Rebind an evtchn so that it gets delivered to a specific cpu */
1192 static int rebind_irq_to_cpu(unsigned irq, unsigned tcpu)
1194 struct evtchn_bind_vcpu bind_vcpu;
1195 int evtchn = evtchn_from_irq(irq);
1197 /* events delivered via platform PCI interrupts are always
1198 * routed to vcpu 0 */
1199 if (!VALID_EVTCHN(evtchn) ||
1200 (xen_hvm_domain() && !xen_have_vector_callback))
1203 /* Send future instances of this interrupt to other vcpu. */
1204 bind_vcpu.port = evtchn;
1205 bind_vcpu.vcpu = tcpu;
1208 * If this fails, it usually just indicates that we're dealing with a
1209 * virq or IPI channel, which don't actually need to be rebound. Ignore
1210 * it, but don't do the xenlinux-level rebind in that case.
1212 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_vcpu, &bind_vcpu) >= 0)
1213 bind_evtchn_to_cpu(evtchn, tcpu);
1218 static int set_affinity_irq(unsigned irq, const struct cpumask *dest)
1220 unsigned tcpu = cpumask_first(dest);
1222 return rebind_irq_to_cpu(irq, tcpu);
1225 int resend_irq_on_evtchn(unsigned int irq)
1227 int masked, evtchn = evtchn_from_irq(irq);
1228 struct shared_info *s = HYPERVISOR_shared_info;
1230 if (!VALID_EVTCHN(evtchn))
1233 masked = sync_test_and_set_bit(evtchn, s->evtchn_mask);
1234 sync_set_bit(evtchn, s->evtchn_pending);
1236 unmask_evtchn(evtchn);
1241 static void enable_dynirq(unsigned int irq)
1243 int evtchn = evtchn_from_irq(irq);
1245 if (VALID_EVTCHN(evtchn))
1246 unmask_evtchn(evtchn);
1249 static void disable_dynirq(unsigned int irq)
1251 int evtchn = evtchn_from_irq(irq);
1253 if (VALID_EVTCHN(evtchn))
1254 mask_evtchn(evtchn);
1257 static void ack_dynirq(unsigned int irq)
1259 int evtchn = evtchn_from_irq(irq);
1261 move_masked_irq(irq);
1263 if (VALID_EVTCHN(evtchn))
1264 unmask_evtchn(evtchn);
1267 static int retrigger_dynirq(unsigned int irq)
1269 int evtchn = evtchn_from_irq(irq);
1270 struct shared_info *sh = HYPERVISOR_shared_info;
1273 if (VALID_EVTCHN(evtchn)) {
1276 masked = sync_test_and_set_bit(evtchn, sh->evtchn_mask);
1277 sync_set_bit(evtchn, sh->evtchn_pending);
1279 unmask_evtchn(evtchn);
1286 static void restore_cpu_virqs(unsigned int cpu)
1288 struct evtchn_bind_virq bind_virq;
1289 int virq, irq, evtchn;
1291 for (virq = 0; virq < NR_VIRQS; virq++) {
1292 if ((irq = per_cpu(virq_to_irq, cpu)[virq]) == -1)
1295 BUG_ON(virq_from_irq(irq) != virq);
1297 /* Get a new binding from Xen. */
1298 bind_virq.virq = virq;
1299 bind_virq.vcpu = cpu;
1300 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_virq,
1303 evtchn = bind_virq.port;
1305 /* Record the new mapping. */
1306 evtchn_to_irq[evtchn] = irq;
1307 irq_info[irq] = mk_virq_info(evtchn, virq);
1308 bind_evtchn_to_cpu(evtchn, cpu);
1312 static void restore_cpu_ipis(unsigned int cpu)
1314 struct evtchn_bind_ipi bind_ipi;
1315 int ipi, irq, evtchn;
1317 for (ipi = 0; ipi < XEN_NR_IPIS; ipi++) {
1318 if ((irq = per_cpu(ipi_to_irq, cpu)[ipi]) == -1)
1321 BUG_ON(ipi_from_irq(irq) != ipi);
1323 /* Get a new binding from Xen. */
1324 bind_ipi.vcpu = cpu;
1325 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_ipi,
1328 evtchn = bind_ipi.port;
1330 /* Record the new mapping. */
1331 evtchn_to_irq[evtchn] = irq;
1332 irq_info[irq] = mk_ipi_info(evtchn, ipi);
1333 bind_evtchn_to_cpu(evtchn, cpu);
1337 /* Clear an irq's pending state, in preparation for polling on it */
1338 void xen_clear_irq_pending(int irq)
1340 int evtchn = evtchn_from_irq(irq);
1342 if (VALID_EVTCHN(evtchn))
1343 clear_evtchn(evtchn);
1345 EXPORT_SYMBOL(xen_clear_irq_pending);
1346 void xen_set_irq_pending(int irq)
1348 int evtchn = evtchn_from_irq(irq);
1350 if (VALID_EVTCHN(evtchn))
1354 bool xen_test_irq_pending(int irq)
1356 int evtchn = evtchn_from_irq(irq);
1359 if (VALID_EVTCHN(evtchn))
1360 ret = test_evtchn(evtchn);
1365 /* Poll waiting for an irq to become pending with timeout. In the usual case,
1366 * the irq will be disabled so it won't deliver an interrupt. */
1367 void xen_poll_irq_timeout(int irq, u64 timeout)
1369 evtchn_port_t evtchn = evtchn_from_irq(irq);
1371 if (VALID_EVTCHN(evtchn)) {
1372 struct sched_poll poll;
1375 poll.timeout = timeout;
1376 set_xen_guest_handle(poll.ports, &evtchn);
1378 if (HYPERVISOR_sched_op(SCHEDOP_poll, &poll) != 0)
1382 EXPORT_SYMBOL(xen_poll_irq_timeout);
1383 /* Poll waiting for an irq to become pending. In the usual case, the
1384 * irq will be disabled so it won't deliver an interrupt. */
1385 void xen_poll_irq(int irq)
1387 xen_poll_irq_timeout(irq, 0 /* no timeout */);
1390 void xen_irq_resume(void)
1392 unsigned int cpu, irq, evtchn;
1393 struct irq_desc *desc;
1395 init_evtchn_cpu_bindings();
1397 /* New event-channel space is not 'live' yet. */
1398 for (evtchn = 0; evtchn < NR_EVENT_CHANNELS; evtchn++)
1399 mask_evtchn(evtchn);
1401 /* No IRQ <-> event-channel mappings. */
1402 for (irq = 0; irq < nr_irqs; irq++)
1403 irq_info[irq].evtchn = 0; /* zap event-channel binding */
1405 for (evtchn = 0; evtchn < NR_EVENT_CHANNELS; evtchn++)
1406 evtchn_to_irq[evtchn] = -1;
1408 for_each_possible_cpu(cpu) {
1409 restore_cpu_virqs(cpu);
1410 restore_cpu_ipis(cpu);
1414 * Unmask any IRQF_NO_SUSPEND IRQs which are enabled. These
1415 * are not handled by the IRQ core.
1417 for_each_irq_desc(irq, desc) {
1418 if (!desc->action || !(desc->action->flags & IRQF_NO_SUSPEND))
1420 if (desc->status & IRQ_DISABLED)
1423 evtchn = evtchn_from_irq(irq);
1427 unmask_evtchn(evtchn);
1431 static struct irq_chip xen_dynamic_chip __read_mostly = {
1434 .disable = disable_dynirq,
1435 .mask = disable_dynirq,
1436 .unmask = enable_dynirq,
1439 .set_affinity = set_affinity_irq,
1440 .retrigger = retrigger_dynirq,
1443 static struct irq_chip xen_pirq_chip __read_mostly = {
1446 .startup = startup_pirq,
1447 .shutdown = shutdown_pirq,
1449 .enable = enable_pirq,
1450 .unmask = enable_pirq,
1452 .disable = disable_pirq,
1453 .mask = disable_pirq,
1458 .set_affinity = set_affinity_irq,
1460 .retrigger = retrigger_dynirq,
1463 static struct irq_chip xen_percpu_chip __read_mostly = {
1464 .name = "xen-percpu",
1466 .disable = disable_dynirq,
1467 .mask = disable_dynirq,
1468 .unmask = enable_dynirq,
1473 int xen_set_callback_via(uint64_t via)
1475 struct xen_hvm_param a;
1476 a.domid = DOMID_SELF;
1477 a.index = HVM_PARAM_CALLBACK_IRQ;
1479 return HYPERVISOR_hvm_op(HVMOP_set_param, &a);
1481 EXPORT_SYMBOL_GPL(xen_set_callback_via);
1483 #ifdef CONFIG_XEN_PVHVM
1484 /* Vector callbacks are better than PCI interrupts to receive event
1485 * channel notifications because we can receive vector callbacks on any
1486 * vcpu and we don't need PCI support or APIC interactions. */
1487 void xen_callback_vector(void)
1490 uint64_t callback_via;
1491 if (xen_have_vector_callback) {
1492 callback_via = HVM_CALLBACK_VECTOR(XEN_HVM_EVTCHN_CALLBACK);
1493 rc = xen_set_callback_via(callback_via);
1495 printk(KERN_ERR "Request for Xen HVM callback vector"
1497 xen_have_vector_callback = 0;
1500 printk(KERN_INFO "Xen HVM callback vector for event delivery is "
1502 /* in the restore case the vector has already been allocated */
1503 if (!test_bit(XEN_HVM_EVTCHN_CALLBACK, used_vectors))
1504 alloc_intr_gate(XEN_HVM_EVTCHN_CALLBACK, xen_hvm_callback_vector);
1508 void xen_callback_vector(void) {}
1511 void __init xen_init_IRQ(void)
1515 cpu_evtchn_mask_p = kcalloc(nr_cpu_ids, sizeof(struct cpu_evtchn_s),
1517 irq_info = kcalloc(nr_irqs, sizeof(*irq_info), GFP_KERNEL);
1519 /* We are using nr_irqs as the maximum number of pirq available but
1520 * that number is actually chosen by Xen and we don't know exactly
1521 * what it is. Be careful choosing high pirq numbers. */
1522 pirq_to_irq = kcalloc(nr_irqs, sizeof(*pirq_to_irq), GFP_KERNEL);
1523 for (i = 0; i < nr_irqs; i++)
1524 pirq_to_irq[i] = -1;
1526 evtchn_to_irq = kcalloc(NR_EVENT_CHANNELS, sizeof(*evtchn_to_irq),
1528 for (i = 0; i < NR_EVENT_CHANNELS; i++)
1529 evtchn_to_irq[i] = -1;
1531 init_evtchn_cpu_bindings();
1533 /* No event channels are 'live' right now. */
1534 for (i = 0; i < NR_EVENT_CHANNELS; i++)
1537 if (xen_hvm_domain()) {
1538 xen_callback_vector();
1540 /* pci_xen_hvm_init must be called after native_init_IRQ so that
1541 * __acpi_register_gsi can point at the right function */
1544 irq_ctx_init(smp_processor_id());
1545 if (xen_initial_domain())