1 // SPDX-License-Identifier: GPL-2.0-only
5 * Xen models interrupts with abstract event channels. Because each
6 * domain gets 1024 event channels, but NR_IRQ is not that large, we
7 * must dynamically map irqs<->event channels. The event channels
8 * interface with the rest of the kernel by defining a xen interrupt
9 * chip. When an event is received, it is mapped to an irq and sent
10 * through the normal interrupt processing path.
12 * There are four kinds of events which can be mapped to an event
15 * 1. Inter-domain notifications. This includes all the virtual
16 * device events, since they're driven by front-ends in another domain
18 * 2. VIRQs, typically used for timers. These are per-cpu events.
20 * 4. PIRQs - Hardware interrupts.
22 * Jeremy Fitzhardinge <jeremy@xensource.com>, XenSource Inc, 2007
25 #define pr_fmt(fmt) "xen:" KBUILD_MODNAME ": " fmt
27 #include <linux/linkage.h>
28 #include <linux/interrupt.h>
29 #include <linux/irq.h>
30 #include <linux/moduleparam.h>
31 #include <linux/string.h>
32 #include <linux/memblock.h>
33 #include <linux/slab.h>
34 #include <linux/irqnr.h>
35 #include <linux/pci.h>
39 #include <asm/ptrace.h>
41 #include <asm/io_apic.h>
42 #include <asm/i8259.h>
43 #include <asm/xen/pci.h>
45 #include <asm/sync_bitops.h>
46 #include <asm/xen/hypercall.h>
47 #include <asm/xen/hypervisor.h>
52 #include <xen/xen-ops.h>
53 #include <xen/events.h>
54 #include <xen/interface/xen.h>
55 #include <xen/interface/event_channel.h>
56 #include <xen/interface/hvm/hvm_op.h>
57 #include <xen/interface/hvm/params.h>
58 #include <xen/interface/physdev.h>
59 #include <xen/interface/sched.h>
60 #include <xen/interface/vcpu.h>
61 #include <asm/hw_irq.h>
63 #include "events_internal.h"
65 const struct evtchn_ops *evtchn_ops;
68 * This lock protects updates to the following mapping and reference-count
69 * arrays. The lock does not need to be acquired to read the mapping tables.
71 static DEFINE_MUTEX(irq_mapping_update_lock);
73 static LIST_HEAD(xen_irq_list_head);
75 /* IRQ <-> VIRQ mapping. */
76 static DEFINE_PER_CPU(int [NR_VIRQS], virq_to_irq) = {[0 ... NR_VIRQS-1] = -1};
78 /* IRQ <-> IPI mapping */
79 static DEFINE_PER_CPU(int [XEN_NR_IPIS], ipi_to_irq) = {[0 ... XEN_NR_IPIS-1] = -1};
83 static unsigned long *pirq_eoi_map;
85 static bool (*pirq_needs_eoi)(unsigned irq);
87 #define EVTCHN_ROW(e) (e / (PAGE_SIZE/sizeof(**evtchn_to_irq)))
88 #define EVTCHN_COL(e) (e % (PAGE_SIZE/sizeof(**evtchn_to_irq)))
89 #define EVTCHN_PER_ROW (PAGE_SIZE / sizeof(**evtchn_to_irq))
91 /* Xen will never allocate port zero for any purpose. */
92 #define VALID_EVTCHN(chn) ((chn) != 0)
94 static struct irq_chip xen_dynamic_chip;
95 static struct irq_chip xen_percpu_chip;
96 static struct irq_chip xen_pirq_chip;
97 static void enable_dynirq(struct irq_data *data);
98 static void disable_dynirq(struct irq_data *data);
100 static void clear_evtchn_to_irq_row(unsigned row)
104 for (col = 0; col < EVTCHN_PER_ROW; col++)
105 evtchn_to_irq[row][col] = -1;
108 static void clear_evtchn_to_irq_all(void)
112 for (row = 0; row < EVTCHN_ROW(xen_evtchn_max_channels()); row++) {
113 if (evtchn_to_irq[row] == NULL)
115 clear_evtchn_to_irq_row(row);
119 static int set_evtchn_to_irq(unsigned evtchn, unsigned irq)
124 if (evtchn >= xen_evtchn_max_channels())
127 row = EVTCHN_ROW(evtchn);
128 col = EVTCHN_COL(evtchn);
130 if (evtchn_to_irq[row] == NULL) {
131 /* Unallocated irq entries return -1 anyway */
135 evtchn_to_irq[row] = (int *)get_zeroed_page(GFP_KERNEL);
136 if (evtchn_to_irq[row] == NULL)
139 clear_evtchn_to_irq_row(row);
142 evtchn_to_irq[row][col] = irq;
146 int get_evtchn_to_irq(unsigned evtchn)
148 if (evtchn >= xen_evtchn_max_channels())
150 if (evtchn_to_irq[EVTCHN_ROW(evtchn)] == NULL)
152 return evtchn_to_irq[EVTCHN_ROW(evtchn)][EVTCHN_COL(evtchn)];
155 /* Get info for IRQ */
156 struct irq_info *info_for_irq(unsigned irq)
158 return irq_get_handler_data(irq);
161 /* Constructors for packed IRQ information. */
162 static int xen_irq_info_common_setup(struct irq_info *info,
164 enum xen_irq_type type,
170 BUG_ON(info->type != IRQT_UNBOUND && info->type != type);
174 info->evtchn = evtchn;
177 ret = set_evtchn_to_irq(evtchn, irq);
181 irq_clear_status_flags(irq, IRQ_NOREQUEST|IRQ_NOAUTOEN);
183 return xen_evtchn_port_setup(info);
186 static int xen_irq_info_evtchn_setup(unsigned irq,
189 struct irq_info *info = info_for_irq(irq);
191 return xen_irq_info_common_setup(info, irq, IRQT_EVTCHN, evtchn, 0);
194 static int xen_irq_info_ipi_setup(unsigned cpu,
199 struct irq_info *info = info_for_irq(irq);
203 per_cpu(ipi_to_irq, cpu)[ipi] = irq;
205 return xen_irq_info_common_setup(info, irq, IRQT_IPI, evtchn, 0);
208 static int xen_irq_info_virq_setup(unsigned cpu,
213 struct irq_info *info = info_for_irq(irq);
217 per_cpu(virq_to_irq, cpu)[virq] = irq;
219 return xen_irq_info_common_setup(info, irq, IRQT_VIRQ, evtchn, 0);
222 static int xen_irq_info_pirq_setup(unsigned irq,
229 struct irq_info *info = info_for_irq(irq);
231 info->u.pirq.pirq = pirq;
232 info->u.pirq.gsi = gsi;
233 info->u.pirq.domid = domid;
234 info->u.pirq.flags = flags;
236 return xen_irq_info_common_setup(info, irq, IRQT_PIRQ, evtchn, 0);
239 static void xen_irq_info_cleanup(struct irq_info *info)
241 set_evtchn_to_irq(info->evtchn, -1);
246 * Accessors for packed IRQ information.
248 unsigned int evtchn_from_irq(unsigned irq)
250 if (unlikely(WARN(irq >= nr_irqs, "Invalid irq %d!\n", irq)))
253 return info_for_irq(irq)->evtchn;
256 unsigned irq_from_evtchn(unsigned int evtchn)
258 return get_evtchn_to_irq(evtchn);
260 EXPORT_SYMBOL_GPL(irq_from_evtchn);
262 int irq_from_virq(unsigned int cpu, unsigned int virq)
264 return per_cpu(virq_to_irq, cpu)[virq];
267 static enum ipi_vector ipi_from_irq(unsigned irq)
269 struct irq_info *info = info_for_irq(irq);
271 BUG_ON(info == NULL);
272 BUG_ON(info->type != IRQT_IPI);
277 static unsigned virq_from_irq(unsigned irq)
279 struct irq_info *info = info_for_irq(irq);
281 BUG_ON(info == NULL);
282 BUG_ON(info->type != IRQT_VIRQ);
287 static unsigned pirq_from_irq(unsigned irq)
289 struct irq_info *info = info_for_irq(irq);
291 BUG_ON(info == NULL);
292 BUG_ON(info->type != IRQT_PIRQ);
294 return info->u.pirq.pirq;
297 static enum xen_irq_type type_from_irq(unsigned irq)
299 return info_for_irq(irq)->type;
302 unsigned cpu_from_irq(unsigned irq)
304 return info_for_irq(irq)->cpu;
307 unsigned int cpu_from_evtchn(unsigned int evtchn)
309 int irq = get_evtchn_to_irq(evtchn);
313 ret = cpu_from_irq(irq);
319 static bool pirq_check_eoi_map(unsigned irq)
321 return test_bit(pirq_from_irq(irq), pirq_eoi_map);
325 static bool pirq_needs_eoi_flag(unsigned irq)
327 struct irq_info *info = info_for_irq(irq);
328 BUG_ON(info->type != IRQT_PIRQ);
330 return info->u.pirq.flags & PIRQ_NEEDS_EOI;
333 static void bind_evtchn_to_cpu(unsigned int chn, unsigned int cpu)
335 int irq = get_evtchn_to_irq(chn);
336 struct irq_info *info = info_for_irq(irq);
340 cpumask_copy(irq_get_affinity_mask(irq), cpumask_of(cpu));
342 xen_evtchn_port_bind_to_cpu(info, cpu);
348 * notify_remote_via_irq - send event to remote end of event channel via irq
349 * @irq: irq of event channel to send event to
351 * Unlike notify_remote_via_evtchn(), this is safe to use across
352 * save/restore. Notifications on a broken connection are silently
355 void notify_remote_via_irq(int irq)
357 int evtchn = evtchn_from_irq(irq);
359 if (VALID_EVTCHN(evtchn))
360 notify_remote_via_evtchn(evtchn);
362 EXPORT_SYMBOL_GPL(notify_remote_via_irq);
364 static void xen_irq_init(unsigned irq)
366 struct irq_info *info;
368 /* By default all event channels notify CPU#0. */
369 cpumask_copy(irq_get_affinity_mask(irq), cpumask_of(0));
372 info = kzalloc(sizeof(*info), GFP_KERNEL);
374 panic("Unable to allocate metadata for IRQ%d\n", irq);
376 info->type = IRQT_UNBOUND;
379 irq_set_handler_data(irq, info);
381 list_add_tail(&info->list, &xen_irq_list_head);
384 static int __must_check xen_allocate_irqs_dynamic(int nvec)
386 int i, irq = irq_alloc_descs(-1, 0, nvec, -1);
389 for (i = 0; i < nvec; i++)
390 xen_irq_init(irq + i);
396 static inline int __must_check xen_allocate_irq_dynamic(void)
399 return xen_allocate_irqs_dynamic(1);
402 static int __must_check xen_allocate_irq_gsi(unsigned gsi)
407 * A PV guest has no concept of a GSI (since it has no ACPI
408 * nor access to/knowledge of the physical APICs). Therefore
409 * all IRQs are dynamically allocated from the entire IRQ
412 if (xen_pv_domain() && !xen_initial_domain())
413 return xen_allocate_irq_dynamic();
415 /* Legacy IRQ descriptors are already allocated by the arch. */
416 if (gsi < nr_legacy_irqs())
419 irq = irq_alloc_desc_at(gsi, -1);
426 static void xen_free_irq(unsigned irq)
428 struct irq_info *info = irq_get_handler_data(irq);
433 list_del(&info->list);
435 irq_set_handler_data(irq, NULL);
437 WARN_ON(info->refcnt > 0);
441 /* Legacy IRQ descriptors are managed by the arch. */
442 if (irq < nr_legacy_irqs())
448 static void xen_evtchn_close(unsigned int port)
450 struct evtchn_close close;
453 if (HYPERVISOR_event_channel_op(EVTCHNOP_close, &close) != 0)
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 void eoi_pirq(struct irq_data *data)
475 int evtchn = evtchn_from_irq(data->irq);
476 struct physdev_eoi eoi = { .irq = pirq_from_irq(data->irq) };
479 if (!VALID_EVTCHN(evtchn))
482 if (unlikely(irqd_is_setaffinity_pending(data)) &&
483 likely(!irqd_irq_disabled(data))) {
484 int masked = test_and_set_mask(evtchn);
486 clear_evtchn(evtchn);
488 irq_move_masked_irq(data);
491 unmask_evtchn(evtchn);
493 clear_evtchn(evtchn);
495 if (pirq_needs_eoi(data->irq)) {
496 rc = HYPERVISOR_physdev_op(PHYSDEVOP_eoi, &eoi);
501 static void mask_ack_pirq(struct irq_data *data)
503 disable_dynirq(data);
507 static unsigned int __startup_pirq(unsigned int irq)
509 struct evtchn_bind_pirq bind_pirq;
510 struct irq_info *info = info_for_irq(irq);
511 int evtchn = evtchn_from_irq(irq);
514 BUG_ON(info->type != IRQT_PIRQ);
516 if (VALID_EVTCHN(evtchn))
519 bind_pirq.pirq = pirq_from_irq(irq);
520 /* NB. We are happy to share unless we are probing. */
521 bind_pirq.flags = info->u.pirq.flags & PIRQ_SHAREABLE ?
522 BIND_PIRQ__WILL_SHARE : 0;
523 rc = HYPERVISOR_event_channel_op(EVTCHNOP_bind_pirq, &bind_pirq);
525 pr_warn("Failed to obtain physical IRQ %d\n", irq);
528 evtchn = bind_pirq.port;
530 pirq_query_unmask(irq);
532 rc = set_evtchn_to_irq(evtchn, irq);
536 info->evtchn = evtchn;
537 bind_evtchn_to_cpu(evtchn, 0);
539 rc = xen_evtchn_port_setup(info);
544 unmask_evtchn(evtchn);
545 eoi_pirq(irq_get_irq_data(irq));
550 pr_err("irq%d: Failed to set port to irq mapping (%d)\n", irq, rc);
551 xen_evtchn_close(evtchn);
555 static unsigned int startup_pirq(struct irq_data *data)
557 return __startup_pirq(data->irq);
560 static void shutdown_pirq(struct irq_data *data)
562 unsigned int irq = data->irq;
563 struct irq_info *info = info_for_irq(irq);
564 unsigned evtchn = evtchn_from_irq(irq);
566 BUG_ON(info->type != IRQT_PIRQ);
568 if (!VALID_EVTCHN(evtchn))
572 xen_evtchn_close(evtchn);
573 xen_irq_info_cleanup(info);
576 static void enable_pirq(struct irq_data *data)
581 static void disable_pirq(struct irq_data *data)
583 disable_dynirq(data);
586 int xen_irq_from_gsi(unsigned gsi)
588 struct irq_info *info;
590 list_for_each_entry(info, &xen_irq_list_head, list) {
591 if (info->type != IRQT_PIRQ)
594 if (info->u.pirq.gsi == gsi)
600 EXPORT_SYMBOL_GPL(xen_irq_from_gsi);
602 static void __unbind_from_irq(unsigned int irq)
604 int evtchn = evtchn_from_irq(irq);
605 struct irq_info *info = irq_get_handler_data(irq);
607 if (info->refcnt > 0) {
609 if (info->refcnt != 0)
613 if (VALID_EVTCHN(evtchn)) {
614 unsigned int cpu = cpu_from_irq(irq);
616 xen_evtchn_close(evtchn);
618 switch (type_from_irq(irq)) {
620 per_cpu(virq_to_irq, cpu)[virq_from_irq(irq)] = -1;
623 per_cpu(ipi_to_irq, cpu)[ipi_from_irq(irq)] = -1;
629 xen_irq_info_cleanup(info);
636 * Do not make any assumptions regarding the relationship between the
637 * IRQ number returned here and the Xen pirq argument.
639 * Note: We don't assign an event channel until the irq actually started
640 * up. Return an existing irq if we've already got one for the gsi.
642 * Shareable implies level triggered, not shareable implies edge
645 int xen_bind_pirq_gsi_to_irq(unsigned gsi,
646 unsigned pirq, int shareable, char *name)
649 struct physdev_irq irq_op;
652 mutex_lock(&irq_mapping_update_lock);
654 irq = xen_irq_from_gsi(gsi);
656 pr_info("%s: returning irq %d for gsi %u\n",
661 irq = xen_allocate_irq_gsi(gsi);
668 /* Only the privileged domain can do this. For non-priv, the pcifront
669 * driver provides a PCI bus that does the call to do exactly
670 * this in the priv domain. */
671 if (xen_initial_domain() &&
672 HYPERVISOR_physdev_op(PHYSDEVOP_alloc_irq_vector, &irq_op)) {
678 ret = xen_irq_info_pirq_setup(irq, 0, pirq, gsi, DOMID_SELF,
679 shareable ? PIRQ_SHAREABLE : 0);
681 __unbind_from_irq(irq);
686 pirq_query_unmask(irq);
687 /* We try to use the handler with the appropriate semantic for the
688 * type of interrupt: if the interrupt is an edge triggered
689 * interrupt we use handle_edge_irq.
691 * On the other hand if the interrupt is level triggered we use
692 * handle_fasteoi_irq like the native code does for this kind of
695 * Depending on the Xen version, pirq_needs_eoi might return true
696 * not only for level triggered interrupts but for edge triggered
697 * interrupts too. In any case Xen always honors the eoi mechanism,
698 * not injecting any more pirqs of the same kind if the first one
699 * hasn't received an eoi yet. Therefore using the fasteoi handler
700 * is the right choice either way.
703 irq_set_chip_and_handler_name(irq, &xen_pirq_chip,
704 handle_fasteoi_irq, name);
706 irq_set_chip_and_handler_name(irq, &xen_pirq_chip,
707 handle_edge_irq, name);
710 mutex_unlock(&irq_mapping_update_lock);
715 #ifdef CONFIG_PCI_MSI
716 int xen_allocate_pirq_msi(struct pci_dev *dev, struct msi_desc *msidesc)
719 struct physdev_get_free_pirq op_get_free_pirq;
721 op_get_free_pirq.type = MAP_PIRQ_TYPE_MSI;
722 rc = HYPERVISOR_physdev_op(PHYSDEVOP_get_free_pirq, &op_get_free_pirq);
724 WARN_ONCE(rc == -ENOSYS,
725 "hypervisor does not support the PHYSDEVOP_get_free_pirq interface\n");
727 return rc ? -1 : op_get_free_pirq.pirq;
730 int xen_bind_pirq_msi_to_irq(struct pci_dev *dev, struct msi_desc *msidesc,
731 int pirq, int nvec, const char *name, domid_t domid)
735 mutex_lock(&irq_mapping_update_lock);
737 irq = xen_allocate_irqs_dynamic(nvec);
741 for (i = 0; i < nvec; i++) {
742 irq_set_chip_and_handler_name(irq + i, &xen_pirq_chip, handle_edge_irq, name);
744 ret = xen_irq_info_pirq_setup(irq + i, 0, pirq + i, 0, domid,
745 i == 0 ? 0 : PIRQ_MSI_GROUP);
750 ret = irq_set_msi_desc(irq, msidesc);
754 mutex_unlock(&irq_mapping_update_lock);
758 __unbind_from_irq(irq + nvec);
759 mutex_unlock(&irq_mapping_update_lock);
764 int xen_destroy_irq(int irq)
766 struct physdev_unmap_pirq unmap_irq;
767 struct irq_info *info = info_for_irq(irq);
770 mutex_lock(&irq_mapping_update_lock);
773 * If trying to remove a vector in a MSI group different
774 * than the first one skip the PIRQ unmap unless this vector
775 * is the first one in the group.
777 if (xen_initial_domain() && !(info->u.pirq.flags & PIRQ_MSI_GROUP)) {
778 unmap_irq.pirq = info->u.pirq.pirq;
779 unmap_irq.domid = info->u.pirq.domid;
780 rc = HYPERVISOR_physdev_op(PHYSDEVOP_unmap_pirq, &unmap_irq);
781 /* If another domain quits without making the pci_disable_msix
782 * call, the Xen hypervisor takes care of freeing the PIRQs
783 * (free_domain_pirqs).
785 if ((rc == -ESRCH && info->u.pirq.domid != DOMID_SELF))
786 pr_info("domain %d does not have %d anymore\n",
787 info->u.pirq.domid, info->u.pirq.pirq);
789 pr_warn("unmap irq failed %d\n", rc);
797 mutex_unlock(&irq_mapping_update_lock);
801 int xen_irq_from_pirq(unsigned pirq)
805 struct irq_info *info;
807 mutex_lock(&irq_mapping_update_lock);
809 list_for_each_entry(info, &xen_irq_list_head, list) {
810 if (info->type != IRQT_PIRQ)
813 if (info->u.pirq.pirq == pirq)
818 mutex_unlock(&irq_mapping_update_lock);
824 int xen_pirq_from_irq(unsigned irq)
826 return pirq_from_irq(irq);
828 EXPORT_SYMBOL_GPL(xen_pirq_from_irq);
830 int bind_evtchn_to_irq(unsigned int evtchn)
835 if (evtchn >= xen_evtchn_max_channels())
838 mutex_lock(&irq_mapping_update_lock);
840 irq = get_evtchn_to_irq(evtchn);
843 irq = xen_allocate_irq_dynamic();
847 irq_set_chip_and_handler_name(irq, &xen_dynamic_chip,
848 handle_edge_irq, "event");
850 ret = xen_irq_info_evtchn_setup(irq, evtchn);
852 __unbind_from_irq(irq);
856 /* New interdomain events are bound to VCPU 0. */
857 bind_evtchn_to_cpu(evtchn, 0);
859 struct irq_info *info = info_for_irq(irq);
860 WARN_ON(info == NULL || info->type != IRQT_EVTCHN);
864 mutex_unlock(&irq_mapping_update_lock);
868 EXPORT_SYMBOL_GPL(bind_evtchn_to_irq);
870 static int bind_ipi_to_irq(unsigned int ipi, unsigned int cpu)
872 struct evtchn_bind_ipi bind_ipi;
876 mutex_lock(&irq_mapping_update_lock);
878 irq = per_cpu(ipi_to_irq, cpu)[ipi];
881 irq = xen_allocate_irq_dynamic();
885 irq_set_chip_and_handler_name(irq, &xen_percpu_chip,
886 handle_percpu_irq, "ipi");
888 bind_ipi.vcpu = xen_vcpu_nr(cpu);
889 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_ipi,
892 evtchn = bind_ipi.port;
894 ret = xen_irq_info_ipi_setup(cpu, irq, evtchn, ipi);
896 __unbind_from_irq(irq);
900 bind_evtchn_to_cpu(evtchn, cpu);
902 struct irq_info *info = info_for_irq(irq);
903 WARN_ON(info == NULL || info->type != IRQT_IPI);
907 mutex_unlock(&irq_mapping_update_lock);
911 int bind_interdomain_evtchn_to_irq(unsigned int remote_domain,
912 unsigned int remote_port)
914 struct evtchn_bind_interdomain bind_interdomain;
917 bind_interdomain.remote_dom = remote_domain;
918 bind_interdomain.remote_port = remote_port;
920 err = HYPERVISOR_event_channel_op(EVTCHNOP_bind_interdomain,
923 return err ? : bind_evtchn_to_irq(bind_interdomain.local_port);
925 EXPORT_SYMBOL_GPL(bind_interdomain_evtchn_to_irq);
927 static int find_virq(unsigned int virq, unsigned int cpu)
929 struct evtchn_status status;
930 int port, rc = -ENOENT;
932 memset(&status, 0, sizeof(status));
933 for (port = 0; port < xen_evtchn_max_channels(); port++) {
934 status.dom = DOMID_SELF;
936 rc = HYPERVISOR_event_channel_op(EVTCHNOP_status, &status);
939 if (status.status != EVTCHNSTAT_virq)
941 if (status.u.virq == virq && status.vcpu == xen_vcpu_nr(cpu)) {
950 * xen_evtchn_nr_channels - number of usable event channel ports
952 * This may be less than the maximum supported by the current
953 * hypervisor ABI. Use xen_evtchn_max_channels() for the maximum
956 unsigned xen_evtchn_nr_channels(void)
958 return evtchn_ops->nr_channels();
960 EXPORT_SYMBOL_GPL(xen_evtchn_nr_channels);
962 int bind_virq_to_irq(unsigned int virq, unsigned int cpu, bool percpu)
964 struct evtchn_bind_virq bind_virq;
965 int evtchn, irq, ret;
967 mutex_lock(&irq_mapping_update_lock);
969 irq = per_cpu(virq_to_irq, cpu)[virq];
972 irq = xen_allocate_irq_dynamic();
977 irq_set_chip_and_handler_name(irq, &xen_percpu_chip,
978 handle_percpu_irq, "virq");
980 irq_set_chip_and_handler_name(irq, &xen_dynamic_chip,
981 handle_edge_irq, "virq");
983 bind_virq.virq = virq;
984 bind_virq.vcpu = xen_vcpu_nr(cpu);
985 ret = HYPERVISOR_event_channel_op(EVTCHNOP_bind_virq,
988 evtchn = bind_virq.port;
991 ret = find_virq(virq, cpu);
996 ret = xen_irq_info_virq_setup(cpu, irq, evtchn, virq);
998 __unbind_from_irq(irq);
1003 bind_evtchn_to_cpu(evtchn, cpu);
1005 struct irq_info *info = info_for_irq(irq);
1006 WARN_ON(info == NULL || info->type != IRQT_VIRQ);
1010 mutex_unlock(&irq_mapping_update_lock);
1015 static void unbind_from_irq(unsigned int irq)
1017 mutex_lock(&irq_mapping_update_lock);
1018 __unbind_from_irq(irq);
1019 mutex_unlock(&irq_mapping_update_lock);
1022 int bind_evtchn_to_irqhandler(unsigned int evtchn,
1023 irq_handler_t handler,
1024 unsigned long irqflags,
1025 const char *devname, void *dev_id)
1029 irq = bind_evtchn_to_irq(evtchn);
1032 retval = request_irq(irq, handler, irqflags, devname, dev_id);
1034 unbind_from_irq(irq);
1040 EXPORT_SYMBOL_GPL(bind_evtchn_to_irqhandler);
1042 int bind_interdomain_evtchn_to_irqhandler(unsigned int remote_domain,
1043 unsigned int remote_port,
1044 irq_handler_t handler,
1045 unsigned long irqflags,
1046 const char *devname,
1051 irq = bind_interdomain_evtchn_to_irq(remote_domain, remote_port);
1055 retval = request_irq(irq, handler, irqflags, devname, dev_id);
1057 unbind_from_irq(irq);
1063 EXPORT_SYMBOL_GPL(bind_interdomain_evtchn_to_irqhandler);
1065 int bind_virq_to_irqhandler(unsigned int virq, unsigned int cpu,
1066 irq_handler_t handler,
1067 unsigned long irqflags, const char *devname, void *dev_id)
1071 irq = bind_virq_to_irq(virq, cpu, irqflags & IRQF_PERCPU);
1074 retval = request_irq(irq, handler, irqflags, devname, dev_id);
1076 unbind_from_irq(irq);
1082 EXPORT_SYMBOL_GPL(bind_virq_to_irqhandler);
1084 int bind_ipi_to_irqhandler(enum ipi_vector ipi,
1086 irq_handler_t handler,
1087 unsigned long irqflags,
1088 const char *devname,
1093 irq = bind_ipi_to_irq(ipi, cpu);
1097 irqflags |= IRQF_NO_SUSPEND | IRQF_FORCE_RESUME | IRQF_EARLY_RESUME;
1098 retval = request_irq(irq, handler, irqflags, devname, dev_id);
1100 unbind_from_irq(irq);
1107 void unbind_from_irqhandler(unsigned int irq, void *dev_id)
1109 struct irq_info *info = irq_get_handler_data(irq);
1113 free_irq(irq, dev_id);
1114 unbind_from_irq(irq);
1116 EXPORT_SYMBOL_GPL(unbind_from_irqhandler);
1119 * xen_set_irq_priority() - set an event channel priority.
1120 * @irq:irq bound to an event channel.
1121 * @priority: priority between XEN_IRQ_PRIORITY_MAX and XEN_IRQ_PRIORITY_MIN.
1123 int xen_set_irq_priority(unsigned irq, unsigned priority)
1125 struct evtchn_set_priority set_priority;
1127 set_priority.port = evtchn_from_irq(irq);
1128 set_priority.priority = priority;
1130 return HYPERVISOR_event_channel_op(EVTCHNOP_set_priority,
1133 EXPORT_SYMBOL_GPL(xen_set_irq_priority);
1135 int evtchn_make_refcounted(unsigned int evtchn)
1137 int irq = get_evtchn_to_irq(evtchn);
1138 struct irq_info *info;
1143 info = irq_get_handler_data(irq);
1148 WARN_ON(info->refcnt != -1);
1154 EXPORT_SYMBOL_GPL(evtchn_make_refcounted);
1156 int evtchn_get(unsigned int evtchn)
1159 struct irq_info *info;
1162 if (evtchn >= xen_evtchn_max_channels())
1165 mutex_lock(&irq_mapping_update_lock);
1167 irq = get_evtchn_to_irq(evtchn);
1171 info = irq_get_handler_data(irq);
1177 if (info->refcnt <= 0)
1183 mutex_unlock(&irq_mapping_update_lock);
1187 EXPORT_SYMBOL_GPL(evtchn_get);
1189 void evtchn_put(unsigned int evtchn)
1191 int irq = get_evtchn_to_irq(evtchn);
1192 if (WARN_ON(irq == -1))
1194 unbind_from_irq(irq);
1196 EXPORT_SYMBOL_GPL(evtchn_put);
1198 void xen_send_IPI_one(unsigned int cpu, enum ipi_vector vector)
1203 if (unlikely(vector == XEN_NMI_VECTOR)) {
1204 int rc = HYPERVISOR_vcpu_op(VCPUOP_send_nmi, xen_vcpu_nr(cpu),
1207 printk(KERN_WARNING "Sending nmi to CPU%d failed (rc:%d)\n", cpu, rc);
1211 irq = per_cpu(ipi_to_irq, cpu)[vector];
1213 notify_remote_via_irq(irq);
1216 static DEFINE_PER_CPU(unsigned, xed_nesting_count);
1218 static void __xen_evtchn_do_upcall(void)
1220 struct vcpu_info *vcpu_info = __this_cpu_read(xen_vcpu);
1221 int cpu = get_cpu();
1225 vcpu_info->evtchn_upcall_pending = 0;
1227 if (__this_cpu_inc_return(xed_nesting_count) - 1)
1230 xen_evtchn_handle_events(cpu);
1232 BUG_ON(!irqs_disabled());
1234 count = __this_cpu_read(xed_nesting_count);
1235 __this_cpu_write(xed_nesting_count, 0);
1236 } while (count != 1 || vcpu_info->evtchn_upcall_pending);
1243 void xen_evtchn_do_upcall(struct pt_regs *regs)
1245 struct pt_regs *old_regs = set_irq_regs(regs);
1249 inc_irq_stat(irq_hv_callback_count);
1252 __xen_evtchn_do_upcall();
1255 set_irq_regs(old_regs);
1258 void xen_hvm_evtchn_do_upcall(void)
1260 __xen_evtchn_do_upcall();
1262 EXPORT_SYMBOL_GPL(xen_hvm_evtchn_do_upcall);
1264 /* Rebind a new event channel to an existing irq. */
1265 void rebind_evtchn_irq(int evtchn, int irq)
1267 struct irq_info *info = info_for_irq(irq);
1272 /* Make sure the irq is masked, since the new event channel
1273 will also be masked. */
1276 mutex_lock(&irq_mapping_update_lock);
1278 /* After resume the irq<->evtchn mappings are all cleared out */
1279 BUG_ON(get_evtchn_to_irq(evtchn) != -1);
1280 /* Expect irq to have been bound before,
1281 so there should be a proper type */
1282 BUG_ON(info->type == IRQT_UNBOUND);
1284 (void)xen_irq_info_evtchn_setup(irq, evtchn);
1286 mutex_unlock(&irq_mapping_update_lock);
1288 bind_evtchn_to_cpu(evtchn, info->cpu);
1289 /* This will be deferred until interrupt is processed */
1290 irq_set_affinity(irq, cpumask_of(info->cpu));
1292 /* Unmask the event channel. */
1296 /* Rebind an evtchn so that it gets delivered to a specific cpu */
1297 static int xen_rebind_evtchn_to_cpu(int evtchn, unsigned int tcpu)
1299 struct evtchn_bind_vcpu bind_vcpu;
1302 if (!VALID_EVTCHN(evtchn))
1305 if (!xen_support_evtchn_rebind())
1308 /* Send future instances of this interrupt to other vcpu. */
1309 bind_vcpu.port = evtchn;
1310 bind_vcpu.vcpu = xen_vcpu_nr(tcpu);
1313 * Mask the event while changing the VCPU binding to prevent
1314 * it being delivered on an unexpected VCPU.
1316 masked = test_and_set_mask(evtchn);
1319 * If this fails, it usually just indicates that we're dealing with a
1320 * virq or IPI channel, which don't actually need to be rebound. Ignore
1321 * it, but don't do the xenlinux-level rebind in that case.
1323 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_vcpu, &bind_vcpu) >= 0)
1324 bind_evtchn_to_cpu(evtchn, tcpu);
1327 unmask_evtchn(evtchn);
1332 static int set_affinity_irq(struct irq_data *data, const struct cpumask *dest,
1335 unsigned tcpu = cpumask_first_and(dest, cpu_online_mask);
1336 int ret = xen_rebind_evtchn_to_cpu(evtchn_from_irq(data->irq), tcpu);
1339 irq_data_update_effective_affinity(data, cpumask_of(tcpu));
1344 /* To be called with desc->lock held. */
1345 int xen_set_affinity_evtchn(struct irq_desc *desc, unsigned int tcpu)
1347 struct irq_data *d = irq_desc_get_irq_data(desc);
1349 return set_affinity_irq(d, cpumask_of(tcpu), false);
1351 EXPORT_SYMBOL_GPL(xen_set_affinity_evtchn);
1353 static void enable_dynirq(struct irq_data *data)
1355 int evtchn = evtchn_from_irq(data->irq);
1357 if (VALID_EVTCHN(evtchn))
1358 unmask_evtchn(evtchn);
1361 static void disable_dynirq(struct irq_data *data)
1363 int evtchn = evtchn_from_irq(data->irq);
1365 if (VALID_EVTCHN(evtchn))
1366 mask_evtchn(evtchn);
1369 static void ack_dynirq(struct irq_data *data)
1371 int evtchn = evtchn_from_irq(data->irq);
1373 if (!VALID_EVTCHN(evtchn))
1376 if (unlikely(irqd_is_setaffinity_pending(data)) &&
1377 likely(!irqd_irq_disabled(data))) {
1378 int masked = test_and_set_mask(evtchn);
1380 clear_evtchn(evtchn);
1382 irq_move_masked_irq(data);
1385 unmask_evtchn(evtchn);
1387 clear_evtchn(evtchn);
1390 static void mask_ack_dynirq(struct irq_data *data)
1392 disable_dynirq(data);
1396 static int retrigger_dynirq(struct irq_data *data)
1398 unsigned int evtchn = evtchn_from_irq(data->irq);
1401 if (!VALID_EVTCHN(evtchn))
1404 masked = test_and_set_mask(evtchn);
1407 unmask_evtchn(evtchn);
1412 static void restore_pirqs(void)
1414 int pirq, rc, irq, gsi;
1415 struct physdev_map_pirq map_irq;
1416 struct irq_info *info;
1418 list_for_each_entry(info, &xen_irq_list_head, list) {
1419 if (info->type != IRQT_PIRQ)
1422 pirq = info->u.pirq.pirq;
1423 gsi = info->u.pirq.gsi;
1426 /* save/restore of PT devices doesn't work, so at this point the
1427 * only devices present are GSI based emulated devices */
1431 map_irq.domid = DOMID_SELF;
1432 map_irq.type = MAP_PIRQ_TYPE_GSI;
1433 map_irq.index = gsi;
1434 map_irq.pirq = pirq;
1436 rc = HYPERVISOR_physdev_op(PHYSDEVOP_map_pirq, &map_irq);
1438 pr_warn("xen map irq failed gsi=%d irq=%d pirq=%d rc=%d\n",
1439 gsi, irq, pirq, rc);
1444 printk(KERN_DEBUG "xen: --> irq=%d, pirq=%d\n", irq, map_irq.pirq);
1446 __startup_pirq(irq);
1450 static void restore_cpu_virqs(unsigned int cpu)
1452 struct evtchn_bind_virq bind_virq;
1453 int virq, irq, evtchn;
1455 for (virq = 0; virq < NR_VIRQS; virq++) {
1456 if ((irq = per_cpu(virq_to_irq, cpu)[virq]) == -1)
1459 BUG_ON(virq_from_irq(irq) != virq);
1461 /* Get a new binding from Xen. */
1462 bind_virq.virq = virq;
1463 bind_virq.vcpu = xen_vcpu_nr(cpu);
1464 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_virq,
1467 evtchn = bind_virq.port;
1469 /* Record the new mapping. */
1470 (void)xen_irq_info_virq_setup(cpu, irq, evtchn, virq);
1471 bind_evtchn_to_cpu(evtchn, cpu);
1475 static void restore_cpu_ipis(unsigned int cpu)
1477 struct evtchn_bind_ipi bind_ipi;
1478 int ipi, irq, evtchn;
1480 for (ipi = 0; ipi < XEN_NR_IPIS; ipi++) {
1481 if ((irq = per_cpu(ipi_to_irq, cpu)[ipi]) == -1)
1484 BUG_ON(ipi_from_irq(irq) != ipi);
1486 /* Get a new binding from Xen. */
1487 bind_ipi.vcpu = xen_vcpu_nr(cpu);
1488 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_ipi,
1491 evtchn = bind_ipi.port;
1493 /* Record the new mapping. */
1494 (void)xen_irq_info_ipi_setup(cpu, irq, evtchn, ipi);
1495 bind_evtchn_to_cpu(evtchn, cpu);
1499 /* Clear an irq's pending state, in preparation for polling on it */
1500 void xen_clear_irq_pending(int irq)
1502 int evtchn = evtchn_from_irq(irq);
1504 if (VALID_EVTCHN(evtchn))
1505 clear_evtchn(evtchn);
1507 EXPORT_SYMBOL(xen_clear_irq_pending);
1508 void xen_set_irq_pending(int irq)
1510 int evtchn = evtchn_from_irq(irq);
1512 if (VALID_EVTCHN(evtchn))
1516 bool xen_test_irq_pending(int irq)
1518 int evtchn = evtchn_from_irq(irq);
1521 if (VALID_EVTCHN(evtchn))
1522 ret = test_evtchn(evtchn);
1527 /* Poll waiting for an irq to become pending with timeout. In the usual case,
1528 * the irq will be disabled so it won't deliver an interrupt. */
1529 void xen_poll_irq_timeout(int irq, u64 timeout)
1531 evtchn_port_t evtchn = evtchn_from_irq(irq);
1533 if (VALID_EVTCHN(evtchn)) {
1534 struct sched_poll poll;
1537 poll.timeout = timeout;
1538 set_xen_guest_handle(poll.ports, &evtchn);
1540 if (HYPERVISOR_sched_op(SCHEDOP_poll, &poll) != 0)
1544 EXPORT_SYMBOL(xen_poll_irq_timeout);
1545 /* Poll waiting for an irq to become pending. In the usual case, the
1546 * irq will be disabled so it won't deliver an interrupt. */
1547 void xen_poll_irq(int irq)
1549 xen_poll_irq_timeout(irq, 0 /* no timeout */);
1552 /* Check whether the IRQ line is shared with other guests. */
1553 int xen_test_irq_shared(int irq)
1555 struct irq_info *info = info_for_irq(irq);
1556 struct physdev_irq_status_query irq_status;
1561 irq_status.irq = info->u.pirq.pirq;
1563 if (HYPERVISOR_physdev_op(PHYSDEVOP_irq_status_query, &irq_status))
1565 return !(irq_status.flags & XENIRQSTAT_shared);
1567 EXPORT_SYMBOL_GPL(xen_test_irq_shared);
1569 void xen_irq_resume(void)
1572 struct irq_info *info;
1574 /* New event-channel space is not 'live' yet. */
1575 xen_evtchn_resume();
1577 /* No IRQ <-> event-channel mappings. */
1578 list_for_each_entry(info, &xen_irq_list_head, list)
1579 info->evtchn = 0; /* zap event-channel binding */
1581 clear_evtchn_to_irq_all();
1583 for_each_possible_cpu(cpu) {
1584 restore_cpu_virqs(cpu);
1585 restore_cpu_ipis(cpu);
1591 static struct irq_chip xen_dynamic_chip __read_mostly = {
1594 .irq_disable = disable_dynirq,
1595 .irq_mask = disable_dynirq,
1596 .irq_unmask = enable_dynirq,
1598 .irq_ack = ack_dynirq,
1599 .irq_mask_ack = mask_ack_dynirq,
1601 .irq_set_affinity = set_affinity_irq,
1602 .irq_retrigger = retrigger_dynirq,
1605 static struct irq_chip xen_pirq_chip __read_mostly = {
1608 .irq_startup = startup_pirq,
1609 .irq_shutdown = shutdown_pirq,
1610 .irq_enable = enable_pirq,
1611 .irq_disable = disable_pirq,
1613 .irq_mask = disable_dynirq,
1614 .irq_unmask = enable_dynirq,
1616 .irq_ack = eoi_pirq,
1617 .irq_eoi = eoi_pirq,
1618 .irq_mask_ack = mask_ack_pirq,
1620 .irq_set_affinity = set_affinity_irq,
1622 .irq_retrigger = retrigger_dynirq,
1625 static struct irq_chip xen_percpu_chip __read_mostly = {
1626 .name = "xen-percpu",
1628 .irq_disable = disable_dynirq,
1629 .irq_mask = disable_dynirq,
1630 .irq_unmask = enable_dynirq,
1632 .irq_ack = ack_dynirq,
1635 int xen_set_callback_via(uint64_t via)
1637 struct xen_hvm_param a;
1638 a.domid = DOMID_SELF;
1639 a.index = HVM_PARAM_CALLBACK_IRQ;
1641 return HYPERVISOR_hvm_op(HVMOP_set_param, &a);
1643 EXPORT_SYMBOL_GPL(xen_set_callback_via);
1645 #ifdef CONFIG_XEN_PVHVM
1646 /* Vector callbacks are better than PCI interrupts to receive event
1647 * channel notifications because we can receive vector callbacks on any
1648 * vcpu and we don't need PCI support or APIC interactions. */
1649 void xen_callback_vector(void)
1652 uint64_t callback_via;
1654 if (xen_have_vector_callback) {
1655 callback_via = HVM_CALLBACK_VECTOR(HYPERVISOR_CALLBACK_VECTOR);
1656 rc = xen_set_callback_via(callback_via);
1658 pr_err("Request for Xen HVM callback vector failed\n");
1659 xen_have_vector_callback = 0;
1662 pr_info_once("Xen HVM callback vector for event delivery is enabled\n");
1663 alloc_intr_gate(HYPERVISOR_CALLBACK_VECTOR,
1664 xen_hvm_callback_vector);
1668 void xen_callback_vector(void) {}
1671 #undef MODULE_PARAM_PREFIX
1672 #define MODULE_PARAM_PREFIX "xen."
1674 static bool fifo_events = true;
1675 module_param(fifo_events, bool, 0);
1677 void __init xen_init_IRQ(void)
1680 unsigned int evtchn;
1683 ret = xen_evtchn_fifo_init();
1685 xen_evtchn_2l_init();
1687 evtchn_to_irq = kcalloc(EVTCHN_ROW(xen_evtchn_max_channels()),
1688 sizeof(*evtchn_to_irq), GFP_KERNEL);
1689 BUG_ON(!evtchn_to_irq);
1691 /* No event channels are 'live' right now. */
1692 for (evtchn = 0; evtchn < xen_evtchn_nr_channels(); evtchn++)
1693 mask_evtchn(evtchn);
1695 pirq_needs_eoi = pirq_needs_eoi_flag;
1698 if (xen_pv_domain()) {
1699 if (xen_initial_domain())
1700 pci_xen_initial_domain();
1702 if (xen_feature(XENFEAT_hvm_callback_vector))
1703 xen_callback_vector();
1705 if (xen_hvm_domain()) {
1707 /* pci_xen_hvm_init must be called after native_init_IRQ so that
1708 * __acpi_register_gsi can point at the right function */
1712 struct physdev_pirq_eoi_gmfn eoi_gmfn;
1714 pirq_eoi_map = (void *)__get_free_page(GFP_KERNEL|__GFP_ZERO);
1715 eoi_gmfn.gmfn = virt_to_gfn(pirq_eoi_map);
1716 rc = HYPERVISOR_physdev_op(PHYSDEVOP_pirq_eoi_gmfn_v2, &eoi_gmfn);
1718 free_page((unsigned long) pirq_eoi_map);
1719 pirq_eoi_map = NULL;
1721 pirq_needs_eoi = pirq_check_eoi_map;