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)->irq_data.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->irq_data.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 xen_allocate_irq_dynamic(void)
384 #ifdef CONFIG_X86_IO_APIC
386 * For an HVM guest or domain 0 which see "real" (emulated or
387 * actual repectively) GSIs we allocate dynamic IRQs
388 * e.g. those corresponding to event channels or MSIs
389 * etc. from the range above those "real" GSIs to avoid
392 if (xen_initial_domain() || xen_hvm_domain())
393 first = get_nr_irqs_gsi();
397 irq = irq_alloc_desc_from(first, -1);
399 if (irq == -ENOMEM && first > NR_IRQS_LEGACY) {
400 printk(KERN_ERR "Out of dynamic IRQ space and eating into GSI space. You should increase nr_irqs\n");
401 first = max(NR_IRQS_LEGACY, first - NR_IRQS_LEGACY);
406 panic("No available IRQ to bind to: increase nr_irqs!\n");
411 static int xen_allocate_irq_gsi(unsigned gsi)
416 * A PV guest has no concept of a GSI (since it has no ACPI
417 * nor access to/knowledge of the physical APICs). Therefore
418 * all IRQs are dynamically allocated from the entire IRQ
421 if (xen_pv_domain() && !xen_initial_domain())
422 return xen_allocate_irq_dynamic();
424 /* Legacy IRQ descriptors are already allocated by the arch. */
425 if (gsi < NR_IRQS_LEGACY)
428 irq = irq_alloc_desc_at(gsi, -1);
430 panic("Unable to allocate to IRQ%d (%d)\n", gsi, irq);
435 static void xen_free_irq(unsigned irq)
437 /* Legacy IRQ descriptors are managed by the arch. */
438 if (irq < NR_IRQS_LEGACY)
444 static void pirq_unmask_notify(int irq)
446 struct physdev_eoi eoi = { .irq = pirq_from_irq(irq) };
448 if (unlikely(pirq_needs_eoi(irq))) {
449 int rc = HYPERVISOR_physdev_op(PHYSDEVOP_eoi, &eoi);
454 static void pirq_query_unmask(int irq)
456 struct physdev_irq_status_query irq_status;
457 struct irq_info *info = info_for_irq(irq);
459 BUG_ON(info->type != IRQT_PIRQ);
461 irq_status.irq = pirq_from_irq(irq);
462 if (HYPERVISOR_physdev_op(PHYSDEVOP_irq_status_query, &irq_status))
463 irq_status.flags = 0;
465 info->u.pirq.flags &= ~PIRQ_NEEDS_EOI;
466 if (irq_status.flags & XENIRQSTAT_needs_eoi)
467 info->u.pirq.flags |= PIRQ_NEEDS_EOI;
470 static bool probing_irq(int irq)
472 struct irq_desc *desc = irq_to_desc(irq);
474 return desc && desc->action == NULL;
477 static unsigned int __startup_pirq(unsigned int irq)
479 struct evtchn_bind_pirq bind_pirq;
480 struct irq_info *info = info_for_irq(irq);
481 int evtchn = evtchn_from_irq(irq);
484 BUG_ON(info->type != IRQT_PIRQ);
486 if (VALID_EVTCHN(evtchn))
489 bind_pirq.pirq = pirq_from_irq(irq);
490 /* NB. We are happy to share unless we are probing. */
491 bind_pirq.flags = info->u.pirq.flags & PIRQ_SHAREABLE ?
492 BIND_PIRQ__WILL_SHARE : 0;
493 rc = HYPERVISOR_event_channel_op(EVTCHNOP_bind_pirq, &bind_pirq);
495 if (!probing_irq(irq))
496 printk(KERN_INFO "Failed to obtain physical IRQ %d\n",
500 evtchn = bind_pirq.port;
502 pirq_query_unmask(irq);
504 evtchn_to_irq[evtchn] = irq;
505 bind_evtchn_to_cpu(evtchn, 0);
506 info->evtchn = evtchn;
509 unmask_evtchn(evtchn);
510 pirq_unmask_notify(irq);
515 static unsigned int startup_pirq(struct irq_data *data)
517 return __startup_pirq(data->irq);
520 static void shutdown_pirq(struct irq_data *data)
522 struct evtchn_close close;
523 unsigned int irq = data->irq;
524 struct irq_info *info = info_for_irq(irq);
525 int evtchn = evtchn_from_irq(irq);
527 BUG_ON(info->type != IRQT_PIRQ);
529 if (!VALID_EVTCHN(evtchn))
535 if (HYPERVISOR_event_channel_op(EVTCHNOP_close, &close) != 0)
538 bind_evtchn_to_cpu(evtchn, 0);
539 evtchn_to_irq[evtchn] = -1;
543 static void enable_pirq(struct irq_data *data)
548 static void disable_pirq(struct irq_data *data)
552 static void ack_pirq(struct irq_data *data)
554 int evtchn = evtchn_from_irq(data->irq);
556 move_native_irq(data->irq);
558 if (VALID_EVTCHN(evtchn)) {
560 clear_evtchn(evtchn);
564 static int find_irq_by_gsi(unsigned gsi)
568 for (irq = 0; irq < nr_irqs; irq++) {
569 struct irq_info *info = info_for_irq(irq);
571 if (info == NULL || info->type != IRQT_PIRQ)
574 if (gsi_from_irq(irq) == gsi)
581 int xen_allocate_pirq(unsigned gsi, int shareable, char *name)
583 return xen_map_pirq_gsi(gsi, gsi, shareable, name);
586 /* xen_map_pirq_gsi might allocate irqs from the top down, as a
587 * consequence don't assume that the irq number returned has a low value
588 * or can be used as a pirq number unless you know otherwise.
590 * One notable exception is when xen_map_pirq_gsi is called passing an
591 * hardware gsi as argument, in that case the irq number returned
592 * matches the gsi number passed as second argument.
594 * Note: We don't assign an event channel until the irq actually started
595 * up. Return an existing irq if we've already got one for the gsi.
597 int xen_map_pirq_gsi(unsigned pirq, unsigned gsi, int shareable, char *name)
600 struct physdev_irq irq_op;
602 spin_lock(&irq_mapping_update_lock);
604 if ((pirq > nr_irqs) || (gsi > nr_irqs)) {
605 printk(KERN_WARNING "xen_map_pirq_gsi: %s %s is incorrect!\n",
606 pirq > nr_irqs ? "pirq" :"",
607 gsi > nr_irqs ? "gsi" : "");
611 irq = find_irq_by_gsi(gsi);
613 printk(KERN_INFO "xen_map_pirq_gsi: returning irq %d for gsi %u\n",
615 goto out; /* XXX need refcount? */
618 irq = xen_allocate_irq_gsi(gsi);
620 set_irq_chip_and_handler_name(irq, &xen_pirq_chip,
621 handle_level_irq, name);
626 /* Only the privileged domain can do this. For non-priv, the pcifront
627 * driver provides a PCI bus that does the call to do exactly
628 * this in the priv domain. */
629 if (xen_initial_domain() &&
630 HYPERVISOR_physdev_op(PHYSDEVOP_alloc_irq_vector, &irq_op)) {
636 irq_info[irq] = mk_pirq_info(0, pirq, gsi, irq_op.vector);
637 irq_info[irq].u.pirq.flags |= shareable ? PIRQ_SHAREABLE : 0;
638 pirq_to_irq[pirq] = irq;
641 spin_unlock(&irq_mapping_update_lock);
646 #ifdef CONFIG_PCI_MSI
647 int xen_allocate_pirq_msi(struct pci_dev *dev, struct msi_desc *msidesc)
650 struct physdev_get_free_pirq op_get_free_pirq;
652 op_get_free_pirq.type = MAP_PIRQ_TYPE_MSI;
653 rc = HYPERVISOR_physdev_op(PHYSDEVOP_get_free_pirq, &op_get_free_pirq);
655 WARN_ONCE(rc == -ENOSYS,
656 "hypervisor does not support the PHYSDEVOP_get_free_pirq interface\n");
658 return rc ? -1 : op_get_free_pirq.pirq;
661 int xen_bind_pirq_msi_to_irq(struct pci_dev *dev, struct msi_desc *msidesc,
662 int pirq, int vector, const char *name)
666 spin_lock(&irq_mapping_update_lock);
668 irq = xen_allocate_irq_dynamic();
672 set_irq_chip_and_handler_name(irq, &xen_pirq_chip,
673 handle_level_irq, name);
675 irq_info[irq] = mk_pirq_info(0, pirq, 0, vector);
676 pirq_to_irq[pirq] = irq;
677 ret = set_irq_msi(irq, msidesc);
681 spin_unlock(&irq_mapping_update_lock);
684 spin_unlock(&irq_mapping_update_lock);
690 int xen_destroy_irq(int irq)
692 struct irq_desc *desc;
693 struct physdev_unmap_pirq unmap_irq;
694 struct irq_info *info = info_for_irq(irq);
697 spin_lock(&irq_mapping_update_lock);
699 desc = irq_to_desc(irq);
703 if (xen_initial_domain()) {
704 unmap_irq.pirq = info->u.pirq.pirq;
705 unmap_irq.domid = DOMID_SELF;
706 rc = HYPERVISOR_physdev_op(PHYSDEVOP_unmap_pirq, &unmap_irq);
708 printk(KERN_WARNING "unmap irq failed %d\n", rc);
712 pirq_to_irq[info->u.pirq.pirq] = -1;
714 irq_info[irq] = mk_unbound_info();
719 spin_unlock(&irq_mapping_update_lock);
723 int xen_vector_from_irq(unsigned irq)
725 return vector_from_irq(irq);
728 int xen_gsi_from_irq(unsigned irq)
730 return gsi_from_irq(irq);
733 int xen_irq_from_pirq(unsigned pirq)
735 return pirq_to_irq[pirq];
738 int bind_evtchn_to_irq(unsigned int evtchn)
742 spin_lock(&irq_mapping_update_lock);
744 irq = evtchn_to_irq[evtchn];
747 irq = xen_allocate_irq_dynamic();
749 set_irq_chip_and_handler_name(irq, &xen_dynamic_chip,
750 handle_fasteoi_irq, "event");
752 evtchn_to_irq[evtchn] = irq;
753 irq_info[irq] = mk_evtchn_info(evtchn);
756 spin_unlock(&irq_mapping_update_lock);
760 EXPORT_SYMBOL_GPL(bind_evtchn_to_irq);
762 static int bind_ipi_to_irq(unsigned int ipi, unsigned int cpu)
764 struct evtchn_bind_ipi bind_ipi;
767 spin_lock(&irq_mapping_update_lock);
769 irq = per_cpu(ipi_to_irq, cpu)[ipi];
772 irq = xen_allocate_irq_dynamic();
776 set_irq_chip_and_handler_name(irq, &xen_percpu_chip,
777 handle_percpu_irq, "ipi");
780 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_ipi,
783 evtchn = bind_ipi.port;
785 evtchn_to_irq[evtchn] = irq;
786 irq_info[irq] = mk_ipi_info(evtchn, ipi);
787 per_cpu(ipi_to_irq, cpu)[ipi] = irq;
789 bind_evtchn_to_cpu(evtchn, cpu);
793 spin_unlock(&irq_mapping_update_lock);
798 int bind_virq_to_irq(unsigned int virq, unsigned int cpu)
800 struct evtchn_bind_virq bind_virq;
803 spin_lock(&irq_mapping_update_lock);
805 irq = per_cpu(virq_to_irq, cpu)[virq];
808 irq = xen_allocate_irq_dynamic();
810 set_irq_chip_and_handler_name(irq, &xen_percpu_chip,
811 handle_percpu_irq, "virq");
813 bind_virq.virq = virq;
814 bind_virq.vcpu = cpu;
815 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_virq,
818 evtchn = bind_virq.port;
820 evtchn_to_irq[evtchn] = irq;
821 irq_info[irq] = mk_virq_info(evtchn, virq);
823 per_cpu(virq_to_irq, cpu)[virq] = irq;
825 bind_evtchn_to_cpu(evtchn, cpu);
828 spin_unlock(&irq_mapping_update_lock);
833 static void unbind_from_irq(unsigned int irq)
835 struct evtchn_close close;
836 int evtchn = evtchn_from_irq(irq);
838 spin_lock(&irq_mapping_update_lock);
840 if (VALID_EVTCHN(evtchn)) {
842 if (HYPERVISOR_event_channel_op(EVTCHNOP_close, &close) != 0)
845 switch (type_from_irq(irq)) {
847 per_cpu(virq_to_irq, cpu_from_evtchn(evtchn))
848 [virq_from_irq(irq)] = -1;
851 per_cpu(ipi_to_irq, cpu_from_evtchn(evtchn))
852 [ipi_from_irq(irq)] = -1;
858 /* Closed ports are implicitly re-bound to VCPU0. */
859 bind_evtchn_to_cpu(evtchn, 0);
861 evtchn_to_irq[evtchn] = -1;
864 if (irq_info[irq].type != IRQT_UNBOUND) {
865 irq_info[irq] = mk_unbound_info();
870 spin_unlock(&irq_mapping_update_lock);
873 int bind_evtchn_to_irqhandler(unsigned int evtchn,
874 irq_handler_t handler,
875 unsigned long irqflags,
876 const char *devname, void *dev_id)
881 irq = bind_evtchn_to_irq(evtchn);
882 retval = request_irq(irq, handler, irqflags, devname, dev_id);
884 unbind_from_irq(irq);
890 EXPORT_SYMBOL_GPL(bind_evtchn_to_irqhandler);
892 int bind_virq_to_irqhandler(unsigned int virq, unsigned int cpu,
893 irq_handler_t handler,
894 unsigned long irqflags, const char *devname, void *dev_id)
899 irq = bind_virq_to_irq(virq, cpu);
900 retval = request_irq(irq, handler, irqflags, devname, dev_id);
902 unbind_from_irq(irq);
908 EXPORT_SYMBOL_GPL(bind_virq_to_irqhandler);
910 int bind_ipi_to_irqhandler(enum ipi_vector ipi,
912 irq_handler_t handler,
913 unsigned long irqflags,
919 irq = bind_ipi_to_irq(ipi, cpu);
923 irqflags |= IRQF_NO_SUSPEND | IRQF_FORCE_RESUME;
924 retval = request_irq(irq, handler, irqflags, devname, dev_id);
926 unbind_from_irq(irq);
933 void unbind_from_irqhandler(unsigned int irq, void *dev_id)
935 free_irq(irq, dev_id);
936 unbind_from_irq(irq);
938 EXPORT_SYMBOL_GPL(unbind_from_irqhandler);
940 void xen_send_IPI_one(unsigned int cpu, enum ipi_vector vector)
942 int irq = per_cpu(ipi_to_irq, cpu)[vector];
944 notify_remote_via_irq(irq);
947 irqreturn_t xen_debug_interrupt(int irq, void *dev_id)
949 struct shared_info *sh = HYPERVISOR_shared_info;
950 int cpu = smp_processor_id();
951 unsigned long *cpu_evtchn = cpu_evtchn_mask(cpu);
954 static DEFINE_SPINLOCK(debug_lock);
957 spin_lock_irqsave(&debug_lock, flags);
959 printk("\nvcpu %d\n ", cpu);
961 for_each_online_cpu(i) {
963 v = per_cpu(xen_vcpu, i);
964 pending = (get_irq_regs() && i == cpu)
965 ? xen_irqs_disabled(get_irq_regs())
966 : v->evtchn_upcall_mask;
967 printk("%d: masked=%d pending=%d event_sel %0*lx\n ", i,
968 pending, v->evtchn_upcall_pending,
969 (int)(sizeof(v->evtchn_pending_sel)*2),
970 v->evtchn_pending_sel);
972 v = per_cpu(xen_vcpu, cpu);
974 printk("\npending:\n ");
975 for (i = ARRAY_SIZE(sh->evtchn_pending)-1; i >= 0; i--)
976 printk("%0*lx%s", (int)sizeof(sh->evtchn_pending[0])*2,
977 sh->evtchn_pending[i],
978 i % 8 == 0 ? "\n " : " ");
979 printk("\nglobal mask:\n ");
980 for (i = ARRAY_SIZE(sh->evtchn_mask)-1; i >= 0; i--)
982 (int)(sizeof(sh->evtchn_mask[0])*2),
984 i % 8 == 0 ? "\n " : " ");
986 printk("\nglobally unmasked:\n ");
987 for (i = ARRAY_SIZE(sh->evtchn_mask)-1; i >= 0; i--)
988 printk("%0*lx%s", (int)(sizeof(sh->evtchn_mask[0])*2),
989 sh->evtchn_pending[i] & ~sh->evtchn_mask[i],
990 i % 8 == 0 ? "\n " : " ");
992 printk("\nlocal cpu%d mask:\n ", cpu);
993 for (i = (NR_EVENT_CHANNELS/BITS_PER_LONG)-1; i >= 0; i--)
994 printk("%0*lx%s", (int)(sizeof(cpu_evtchn[0])*2),
996 i % 8 == 0 ? "\n " : " ");
998 printk("\nlocally unmasked:\n ");
999 for (i = ARRAY_SIZE(sh->evtchn_mask)-1; i >= 0; i--) {
1000 unsigned long pending = sh->evtchn_pending[i]
1001 & ~sh->evtchn_mask[i]
1003 printk("%0*lx%s", (int)(sizeof(sh->evtchn_mask[0])*2),
1004 pending, i % 8 == 0 ? "\n " : " ");
1007 printk("\npending list:\n");
1008 for (i = 0; i < NR_EVENT_CHANNELS; i++) {
1009 if (sync_test_bit(i, sh->evtchn_pending)) {
1010 int word_idx = i / BITS_PER_LONG;
1011 printk(" %d: event %d -> irq %d%s%s%s\n",
1012 cpu_from_evtchn(i), i,
1014 sync_test_bit(word_idx, &v->evtchn_pending_sel)
1016 !sync_test_bit(i, sh->evtchn_mask)
1017 ? "" : " globally-masked",
1018 sync_test_bit(i, cpu_evtchn)
1019 ? "" : " locally-masked");
1023 spin_unlock_irqrestore(&debug_lock, flags);
1028 static DEFINE_PER_CPU(unsigned, xed_nesting_count);
1029 static DEFINE_PER_CPU(unsigned int, current_word_idx);
1030 static DEFINE_PER_CPU(unsigned int, current_bit_idx);
1033 * Mask out the i least significant bits of w
1035 #define MASK_LSBS(w, i) (w & ((~0UL) << i))
1038 * Search the CPUs pending events bitmasks. For each one found, map
1039 * the event number to an irq, and feed it into do_IRQ() for
1042 * Xen uses a two-level bitmap to speed searching. The first level is
1043 * a bitset of words which contain pending event bits. The second
1044 * level is a bitset of pending events themselves.
1046 static void __xen_evtchn_do_upcall(void)
1048 int start_word_idx, start_bit_idx;
1049 int word_idx, bit_idx;
1051 int cpu = get_cpu();
1052 struct shared_info *s = HYPERVISOR_shared_info;
1053 struct vcpu_info *vcpu_info = __this_cpu_read(xen_vcpu);
1057 unsigned long pending_words;
1059 vcpu_info->evtchn_upcall_pending = 0;
1061 if (__this_cpu_inc_return(xed_nesting_count) - 1)
1064 #ifndef CONFIG_X86 /* No need for a barrier -- XCHG is a barrier on x86. */
1065 /* Clear master flag /before/ clearing selector flag. */
1068 pending_words = xchg(&vcpu_info->evtchn_pending_sel, 0);
1070 start_word_idx = __this_cpu_read(current_word_idx);
1071 start_bit_idx = __this_cpu_read(current_bit_idx);
1073 word_idx = start_word_idx;
1075 for (i = 0; pending_words != 0; i++) {
1076 unsigned long pending_bits;
1077 unsigned long words;
1079 words = MASK_LSBS(pending_words, word_idx);
1082 * If we masked out all events, wrap to beginning.
1089 word_idx = __ffs(words);
1091 pending_bits = active_evtchns(cpu, s, word_idx);
1092 bit_idx = 0; /* usually scan entire word from start */
1093 if (word_idx == start_word_idx) {
1094 /* We scan the starting word in two parts */
1096 /* 1st time: start in the middle */
1097 bit_idx = start_bit_idx;
1099 /* 2nd time: mask bits done already */
1100 bit_idx &= (1UL << start_bit_idx) - 1;
1106 struct irq_desc *desc;
1108 bits = MASK_LSBS(pending_bits, bit_idx);
1110 /* If we masked out all events, move on. */
1114 bit_idx = __ffs(bits);
1117 port = (word_idx * BITS_PER_LONG) + bit_idx;
1118 irq = evtchn_to_irq[port];
1124 desc = irq_to_desc(irq);
1126 generic_handle_irq_desc(irq, desc);
1129 bit_idx = (bit_idx + 1) % BITS_PER_LONG;
1131 /* Next caller starts at last processed + 1 */
1132 __this_cpu_write(current_word_idx,
1133 bit_idx ? word_idx :
1134 (word_idx+1) % BITS_PER_LONG);
1135 __this_cpu_write(current_bit_idx, bit_idx);
1136 } while (bit_idx != 0);
1138 /* Scan start_l1i twice; all others once. */
1139 if ((word_idx != start_word_idx) || (i != 0))
1140 pending_words &= ~(1UL << word_idx);
1142 word_idx = (word_idx + 1) % BITS_PER_LONG;
1145 BUG_ON(!irqs_disabled());
1147 count = __this_cpu_read(xed_nesting_count);
1148 __this_cpu_write(xed_nesting_count, 0);
1149 } while (count != 1 || vcpu_info->evtchn_upcall_pending);
1156 void xen_evtchn_do_upcall(struct pt_regs *regs)
1158 struct pt_regs *old_regs = set_irq_regs(regs);
1163 __xen_evtchn_do_upcall();
1166 set_irq_regs(old_regs);
1169 void xen_hvm_evtchn_do_upcall(void)
1171 __xen_evtchn_do_upcall();
1173 EXPORT_SYMBOL_GPL(xen_hvm_evtchn_do_upcall);
1175 /* Rebind a new event channel to an existing irq. */
1176 void rebind_evtchn_irq(int evtchn, int irq)
1178 struct irq_info *info = info_for_irq(irq);
1180 /* Make sure the irq is masked, since the new event channel
1181 will also be masked. */
1184 spin_lock(&irq_mapping_update_lock);
1186 /* After resume the irq<->evtchn mappings are all cleared out */
1187 BUG_ON(evtchn_to_irq[evtchn] != -1);
1188 /* Expect irq to have been bound before,
1189 so there should be a proper type */
1190 BUG_ON(info->type == IRQT_UNBOUND);
1192 evtchn_to_irq[evtchn] = irq;
1193 irq_info[irq] = mk_evtchn_info(evtchn);
1195 spin_unlock(&irq_mapping_update_lock);
1197 /* new event channels are always bound to cpu 0 */
1198 irq_set_affinity(irq, cpumask_of(0));
1200 /* Unmask the event channel. */
1204 /* Rebind an evtchn so that it gets delivered to a specific cpu */
1205 static int rebind_irq_to_cpu(unsigned irq, unsigned tcpu)
1207 struct evtchn_bind_vcpu bind_vcpu;
1208 int evtchn = evtchn_from_irq(irq);
1210 /* events delivered via platform PCI interrupts are always
1211 * routed to vcpu 0 */
1212 if (!VALID_EVTCHN(evtchn) ||
1213 (xen_hvm_domain() && !xen_have_vector_callback))
1216 /* Send future instances of this interrupt to other vcpu. */
1217 bind_vcpu.port = evtchn;
1218 bind_vcpu.vcpu = tcpu;
1221 * If this fails, it usually just indicates that we're dealing with a
1222 * virq or IPI channel, which don't actually need to be rebound. Ignore
1223 * it, but don't do the xenlinux-level rebind in that case.
1225 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_vcpu, &bind_vcpu) >= 0)
1226 bind_evtchn_to_cpu(evtchn, tcpu);
1231 static int set_affinity_irq(struct irq_data *data, const struct cpumask *dest,
1234 unsigned tcpu = cpumask_first(dest);
1236 return rebind_irq_to_cpu(data->irq, tcpu);
1239 int resend_irq_on_evtchn(unsigned int irq)
1241 int masked, evtchn = evtchn_from_irq(irq);
1242 struct shared_info *s = HYPERVISOR_shared_info;
1244 if (!VALID_EVTCHN(evtchn))
1247 masked = sync_test_and_set_bit(evtchn, s->evtchn_mask);
1248 sync_set_bit(evtchn, s->evtchn_pending);
1250 unmask_evtchn(evtchn);
1255 static void enable_dynirq(struct irq_data *data)
1257 int evtchn = evtchn_from_irq(data->irq);
1259 if (VALID_EVTCHN(evtchn))
1260 unmask_evtchn(evtchn);
1263 static void disable_dynirq(struct irq_data *data)
1265 int evtchn = evtchn_from_irq(data->irq);
1267 if (VALID_EVTCHN(evtchn))
1268 mask_evtchn(evtchn);
1271 static void ack_dynirq(struct irq_data *data)
1273 int evtchn = evtchn_from_irq(data->irq);
1275 move_masked_irq(data->irq);
1277 if (VALID_EVTCHN(evtchn))
1278 unmask_evtchn(evtchn);
1281 static int retrigger_dynirq(struct irq_data *data)
1283 int evtchn = evtchn_from_irq(data->irq);
1284 struct shared_info *sh = HYPERVISOR_shared_info;
1287 if (VALID_EVTCHN(evtchn)) {
1290 masked = sync_test_and_set_bit(evtchn, sh->evtchn_mask);
1291 sync_set_bit(evtchn, sh->evtchn_pending);
1293 unmask_evtchn(evtchn);
1300 static void restore_cpu_pirqs(void)
1302 int pirq, rc, irq, gsi;
1303 struct physdev_map_pirq map_irq;
1305 for (pirq = 0; pirq < nr_irqs; pirq++) {
1306 irq = pirq_to_irq[pirq];
1310 /* save/restore of PT devices doesn't work, so at this point the
1311 * only devices present are GSI based emulated devices */
1312 gsi = gsi_from_irq(irq);
1316 map_irq.domid = DOMID_SELF;
1317 map_irq.type = MAP_PIRQ_TYPE_GSI;
1318 map_irq.index = gsi;
1319 map_irq.pirq = pirq;
1321 rc = HYPERVISOR_physdev_op(PHYSDEVOP_map_pirq, &map_irq);
1323 printk(KERN_WARNING "xen map irq failed gsi=%d irq=%d pirq=%d rc=%d\n",
1324 gsi, irq, pirq, rc);
1325 irq_info[irq] = mk_unbound_info();
1326 pirq_to_irq[pirq] = -1;
1330 printk(KERN_DEBUG "xen: --> irq=%d, pirq=%d\n", irq, map_irq.pirq);
1332 __startup_pirq(irq);
1336 static void restore_cpu_virqs(unsigned int cpu)
1338 struct evtchn_bind_virq bind_virq;
1339 int virq, irq, evtchn;
1341 for (virq = 0; virq < NR_VIRQS; virq++) {
1342 if ((irq = per_cpu(virq_to_irq, cpu)[virq]) == -1)
1345 BUG_ON(virq_from_irq(irq) != virq);
1347 /* Get a new binding from Xen. */
1348 bind_virq.virq = virq;
1349 bind_virq.vcpu = cpu;
1350 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_virq,
1353 evtchn = bind_virq.port;
1355 /* Record the new mapping. */
1356 evtchn_to_irq[evtchn] = irq;
1357 irq_info[irq] = mk_virq_info(evtchn, virq);
1358 bind_evtchn_to_cpu(evtchn, cpu);
1362 static void restore_cpu_ipis(unsigned int cpu)
1364 struct evtchn_bind_ipi bind_ipi;
1365 int ipi, irq, evtchn;
1367 for (ipi = 0; ipi < XEN_NR_IPIS; ipi++) {
1368 if ((irq = per_cpu(ipi_to_irq, cpu)[ipi]) == -1)
1371 BUG_ON(ipi_from_irq(irq) != ipi);
1373 /* Get a new binding from Xen. */
1374 bind_ipi.vcpu = cpu;
1375 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_ipi,
1378 evtchn = bind_ipi.port;
1380 /* Record the new mapping. */
1381 evtchn_to_irq[evtchn] = irq;
1382 irq_info[irq] = mk_ipi_info(evtchn, ipi);
1383 bind_evtchn_to_cpu(evtchn, cpu);
1387 /* Clear an irq's pending state, in preparation for polling on it */
1388 void xen_clear_irq_pending(int irq)
1390 int evtchn = evtchn_from_irq(irq);
1392 if (VALID_EVTCHN(evtchn))
1393 clear_evtchn(evtchn);
1395 EXPORT_SYMBOL(xen_clear_irq_pending);
1396 void xen_set_irq_pending(int irq)
1398 int evtchn = evtchn_from_irq(irq);
1400 if (VALID_EVTCHN(evtchn))
1404 bool xen_test_irq_pending(int irq)
1406 int evtchn = evtchn_from_irq(irq);
1409 if (VALID_EVTCHN(evtchn))
1410 ret = test_evtchn(evtchn);
1415 /* Poll waiting for an irq to become pending with timeout. In the usual case,
1416 * the irq will be disabled so it won't deliver an interrupt. */
1417 void xen_poll_irq_timeout(int irq, u64 timeout)
1419 evtchn_port_t evtchn = evtchn_from_irq(irq);
1421 if (VALID_EVTCHN(evtchn)) {
1422 struct sched_poll poll;
1425 poll.timeout = timeout;
1426 set_xen_guest_handle(poll.ports, &evtchn);
1428 if (HYPERVISOR_sched_op(SCHEDOP_poll, &poll) != 0)
1432 EXPORT_SYMBOL(xen_poll_irq_timeout);
1433 /* Poll waiting for an irq to become pending. In the usual case, the
1434 * irq will be disabled so it won't deliver an interrupt. */
1435 void xen_poll_irq(int irq)
1437 xen_poll_irq_timeout(irq, 0 /* no timeout */);
1440 void xen_irq_resume(void)
1442 unsigned int cpu, irq, evtchn;
1444 init_evtchn_cpu_bindings();
1446 /* New event-channel space is not 'live' yet. */
1447 for (evtchn = 0; evtchn < NR_EVENT_CHANNELS; evtchn++)
1448 mask_evtchn(evtchn);
1450 /* No IRQ <-> event-channel mappings. */
1451 for (irq = 0; irq < nr_irqs; irq++)
1452 irq_info[irq].evtchn = 0; /* zap event-channel binding */
1454 for (evtchn = 0; evtchn < NR_EVENT_CHANNELS; evtchn++)
1455 evtchn_to_irq[evtchn] = -1;
1457 for_each_possible_cpu(cpu) {
1458 restore_cpu_virqs(cpu);
1459 restore_cpu_ipis(cpu);
1462 restore_cpu_pirqs();
1465 static struct irq_chip xen_dynamic_chip __read_mostly = {
1468 .irq_disable = disable_dynirq,
1469 .irq_mask = disable_dynirq,
1470 .irq_unmask = enable_dynirq,
1472 .irq_eoi = ack_dynirq,
1473 .irq_set_affinity = set_affinity_irq,
1474 .irq_retrigger = retrigger_dynirq,
1477 static struct irq_chip xen_pirq_chip __read_mostly = {
1480 .irq_startup = startup_pirq,
1481 .irq_shutdown = shutdown_pirq,
1483 .irq_enable = enable_pirq,
1484 .irq_unmask = enable_pirq,
1486 .irq_disable = disable_pirq,
1487 .irq_mask = disable_pirq,
1489 .irq_ack = ack_pirq,
1491 .irq_set_affinity = set_affinity_irq,
1493 .irq_retrigger = retrigger_dynirq,
1496 static struct irq_chip xen_percpu_chip __read_mostly = {
1497 .name = "xen-percpu",
1499 .irq_disable = disable_dynirq,
1500 .irq_mask = disable_dynirq,
1501 .irq_unmask = enable_dynirq,
1503 .irq_ack = ack_dynirq,
1506 int xen_set_callback_via(uint64_t via)
1508 struct xen_hvm_param a;
1509 a.domid = DOMID_SELF;
1510 a.index = HVM_PARAM_CALLBACK_IRQ;
1512 return HYPERVISOR_hvm_op(HVMOP_set_param, &a);
1514 EXPORT_SYMBOL_GPL(xen_set_callback_via);
1516 #ifdef CONFIG_XEN_PVHVM
1517 /* Vector callbacks are better than PCI interrupts to receive event
1518 * channel notifications because we can receive vector callbacks on any
1519 * vcpu and we don't need PCI support or APIC interactions. */
1520 void xen_callback_vector(void)
1523 uint64_t callback_via;
1524 if (xen_have_vector_callback) {
1525 callback_via = HVM_CALLBACK_VECTOR(XEN_HVM_EVTCHN_CALLBACK);
1526 rc = xen_set_callback_via(callback_via);
1528 printk(KERN_ERR "Request for Xen HVM callback vector"
1530 xen_have_vector_callback = 0;
1533 printk(KERN_INFO "Xen HVM callback vector for event delivery is "
1535 /* in the restore case the vector has already been allocated */
1536 if (!test_bit(XEN_HVM_EVTCHN_CALLBACK, used_vectors))
1537 alloc_intr_gate(XEN_HVM_EVTCHN_CALLBACK, xen_hvm_callback_vector);
1541 void xen_callback_vector(void) {}
1544 void __init xen_init_IRQ(void)
1548 cpu_evtchn_mask_p = kcalloc(nr_cpu_ids, sizeof(struct cpu_evtchn_s),
1550 irq_info = kcalloc(nr_irqs, sizeof(*irq_info), GFP_KERNEL);
1552 /* We are using nr_irqs as the maximum number of pirq available but
1553 * that number is actually chosen by Xen and we don't know exactly
1554 * what it is. Be careful choosing high pirq numbers. */
1555 pirq_to_irq = kcalloc(nr_irqs, sizeof(*pirq_to_irq), GFP_KERNEL);
1556 for (i = 0; i < nr_irqs; i++)
1557 pirq_to_irq[i] = -1;
1559 evtchn_to_irq = kcalloc(NR_EVENT_CHANNELS, sizeof(*evtchn_to_irq),
1561 for (i = 0; i < NR_EVENT_CHANNELS; i++)
1562 evtchn_to_irq[i] = -1;
1564 init_evtchn_cpu_bindings();
1566 /* No event channels are 'live' right now. */
1567 for (i = 0; i < NR_EVENT_CHANNELS; i++)
1570 if (xen_hvm_domain()) {
1571 xen_callback_vector();
1573 /* pci_xen_hvm_init must be called after native_init_IRQ so that
1574 * __acpi_register_gsi can point at the right function */
1577 irq_ctx_init(smp_processor_id());
1578 if (xen_initial_domain())