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 received, 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 #define pr_fmt(fmt) "xen:" KBUILD_MODNAME ": " fmt
26 #include <linux/linkage.h>
27 #include <linux/interrupt.h>
28 #include <linux/irq.h>
29 #include <linux/module.h>
30 #include <linux/string.h>
31 #include <linux/bootmem.h>
32 #include <linux/slab.h>
33 #include <linux/irqnr.h>
34 #include <linux/pci.h>
38 #include <asm/ptrace.h>
41 #include <asm/io_apic.h>
42 #include <asm/xen/page.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>
51 #include <xen/xen-ops.h>
52 #include <xen/events.h>
53 #include <xen/interface/xen.h>
54 #include <xen/interface/event_channel.h>
55 #include <xen/interface/hvm/hvm_op.h>
56 #include <xen/interface/hvm/params.h>
57 #include <xen/interface/physdev.h>
58 #include <xen/interface/sched.h>
59 #include <xen/interface/vcpu.h>
60 #include <asm/hw_irq.h>
62 #include "events_internal.h"
64 const struct evtchn_ops *evtchn_ops;
67 * This lock protects updates to the following mapping and reference-count
68 * arrays. The lock does not need to be acquired to read the mapping tables.
70 static DEFINE_MUTEX(irq_mapping_update_lock);
72 static LIST_HEAD(xen_irq_list_head);
74 /* IRQ <-> VIRQ mapping. */
75 static DEFINE_PER_CPU(int [NR_VIRQS], virq_to_irq) = {[0 ... NR_VIRQS-1] = -1};
77 /* IRQ <-> IPI mapping */
78 static DEFINE_PER_CPU(int [XEN_NR_IPIS], ipi_to_irq) = {[0 ... XEN_NR_IPIS-1] = -1};
82 static unsigned long *pirq_eoi_map;
84 static bool (*pirq_needs_eoi)(unsigned irq);
86 #define EVTCHN_ROW(e) (e / (PAGE_SIZE/sizeof(**evtchn_to_irq)))
87 #define EVTCHN_COL(e) (e % (PAGE_SIZE/sizeof(**evtchn_to_irq)))
88 #define EVTCHN_PER_ROW (PAGE_SIZE / sizeof(**evtchn_to_irq))
90 /* Xen will never allocate port zero for any purpose. */
91 #define VALID_EVTCHN(chn) ((chn) != 0)
93 static struct irq_chip xen_dynamic_chip;
94 static struct irq_chip xen_percpu_chip;
95 static struct irq_chip xen_pirq_chip;
96 static void enable_dynirq(struct irq_data *data);
97 static void disable_dynirq(struct irq_data *data);
99 static void clear_evtchn_to_irq_row(unsigned row)
103 for (col = 0; col < EVTCHN_PER_ROW; col++)
104 evtchn_to_irq[row][col] = -1;
107 static void clear_evtchn_to_irq_all(void)
111 for (row = 0; row < EVTCHN_ROW(xen_evtchn_max_channels()); row++) {
112 if (evtchn_to_irq[row] == NULL)
114 clear_evtchn_to_irq_row(row);
118 static int set_evtchn_to_irq(unsigned evtchn, unsigned irq)
123 if (evtchn >= xen_evtchn_max_channels())
126 row = EVTCHN_ROW(evtchn);
127 col = EVTCHN_COL(evtchn);
129 if (evtchn_to_irq[row] == NULL) {
130 /* Unallocated irq entries return -1 anyway */
134 evtchn_to_irq[row] = (int *)get_zeroed_page(GFP_KERNEL);
135 if (evtchn_to_irq[row] == NULL)
138 clear_evtchn_to_irq_row(row);
141 evtchn_to_irq[EVTCHN_ROW(evtchn)][EVTCHN_COL(evtchn)] = irq;
145 int get_evtchn_to_irq(unsigned evtchn)
147 if (evtchn >= xen_evtchn_max_channels())
149 if (evtchn_to_irq[EVTCHN_ROW(evtchn)] == NULL)
151 return evtchn_to_irq[EVTCHN_ROW(evtchn)][EVTCHN_COL(evtchn)];
154 /* Get info for IRQ */
155 struct irq_info *info_for_irq(unsigned irq)
157 return irq_get_handler_data(irq);
160 /* Constructors for packed IRQ information. */
161 static int xen_irq_info_common_setup(struct irq_info *info,
163 enum xen_irq_type type,
169 BUG_ON(info->type != IRQT_UNBOUND && info->type != type);
173 info->evtchn = evtchn;
176 ret = set_evtchn_to_irq(evtchn, irq);
180 irq_clear_status_flags(irq, IRQ_NOREQUEST|IRQ_NOAUTOEN);
182 return xen_evtchn_port_setup(info);
185 static int xen_irq_info_evtchn_setup(unsigned irq,
188 struct irq_info *info = info_for_irq(irq);
190 return xen_irq_info_common_setup(info, irq, IRQT_EVTCHN, evtchn, 0);
193 static int xen_irq_info_ipi_setup(unsigned cpu,
198 struct irq_info *info = info_for_irq(irq);
202 per_cpu(ipi_to_irq, cpu)[ipi] = irq;
204 return xen_irq_info_common_setup(info, irq, IRQT_IPI, evtchn, 0);
207 static int xen_irq_info_virq_setup(unsigned cpu,
212 struct irq_info *info = info_for_irq(irq);
216 per_cpu(virq_to_irq, cpu)[virq] = irq;
218 return xen_irq_info_common_setup(info, irq, IRQT_VIRQ, evtchn, 0);
221 static int xen_irq_info_pirq_setup(unsigned irq,
228 struct irq_info *info = info_for_irq(irq);
230 info->u.pirq.pirq = pirq;
231 info->u.pirq.gsi = gsi;
232 info->u.pirq.domid = domid;
233 info->u.pirq.flags = flags;
235 return xen_irq_info_common_setup(info, irq, IRQT_PIRQ, evtchn, 0);
238 static void xen_irq_info_cleanup(struct irq_info *info)
240 set_evtchn_to_irq(info->evtchn, -1);
245 * Accessors for packed IRQ information.
247 unsigned int evtchn_from_irq(unsigned irq)
249 if (unlikely(WARN(irq < 0 || irq >= nr_irqs, "Invalid irq %d!\n", irq)))
252 return info_for_irq(irq)->evtchn;
255 unsigned irq_from_evtchn(unsigned int evtchn)
257 return get_evtchn_to_irq(evtchn);
259 EXPORT_SYMBOL_GPL(irq_from_evtchn);
261 int irq_from_virq(unsigned int cpu, unsigned int virq)
263 return per_cpu(virq_to_irq, cpu)[virq];
266 static enum ipi_vector ipi_from_irq(unsigned irq)
268 struct irq_info *info = info_for_irq(irq);
270 BUG_ON(info == NULL);
271 BUG_ON(info->type != IRQT_IPI);
276 static unsigned virq_from_irq(unsigned irq)
278 struct irq_info *info = info_for_irq(irq);
280 BUG_ON(info == NULL);
281 BUG_ON(info->type != IRQT_VIRQ);
286 static unsigned pirq_from_irq(unsigned irq)
288 struct irq_info *info = info_for_irq(irq);
290 BUG_ON(info == NULL);
291 BUG_ON(info->type != IRQT_PIRQ);
293 return info->u.pirq.pirq;
296 static enum xen_irq_type type_from_irq(unsigned irq)
298 return info_for_irq(irq)->type;
301 unsigned cpu_from_irq(unsigned irq)
303 return info_for_irq(irq)->cpu;
306 unsigned int cpu_from_evtchn(unsigned int evtchn)
308 int irq = get_evtchn_to_irq(evtchn);
312 ret = cpu_from_irq(irq);
318 static bool pirq_check_eoi_map(unsigned irq)
320 return test_bit(pirq_from_irq(irq), pirq_eoi_map);
324 static bool pirq_needs_eoi_flag(unsigned irq)
326 struct irq_info *info = info_for_irq(irq);
327 BUG_ON(info->type != IRQT_PIRQ);
329 return info->u.pirq.flags & PIRQ_NEEDS_EOI;
332 static void bind_evtchn_to_cpu(unsigned int chn, unsigned int cpu)
334 int irq = get_evtchn_to_irq(chn);
335 struct irq_info *info = info_for_irq(irq);
339 cpumask_copy(irq_to_desc(irq)->irq_data.affinity, 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 struct irq_desc *desc = irq_to_desc(irq);
370 /* By default all event channels notify CPU#0. */
371 cpumask_copy(desc->irq_data.affinity, cpumask_of(0));
374 info = kzalloc(sizeof(*info), GFP_KERNEL);
376 panic("Unable to allocate metadata for IRQ%d\n", irq);
378 info->type = IRQT_UNBOUND;
381 irq_set_handler_data(irq, info);
383 list_add_tail(&info->list, &xen_irq_list_head);
386 static int __must_check xen_allocate_irq_dynamic(void)
391 #ifdef CONFIG_X86_IO_APIC
393 * For an HVM guest or domain 0 which see "real" (emulated or
394 * actual respectively) GSIs we allocate dynamic IRQs
395 * e.g. those corresponding to event channels or MSIs
396 * etc. from the range above those "real" GSIs to avoid
399 if (xen_initial_domain() || xen_hvm_domain())
400 first = get_nr_irqs_gsi();
403 irq = irq_alloc_desc_from(first, -1);
411 static int __must_check 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);
435 static void xen_free_irq(unsigned irq)
437 struct irq_info *info = irq_get_handler_data(irq);
442 list_del(&info->list);
444 irq_set_handler_data(irq, NULL);
446 WARN_ON(info->refcnt > 0);
450 /* Legacy IRQ descriptors are managed by the arch. */
451 if (irq < NR_IRQS_LEGACY)
457 static void xen_evtchn_close(unsigned int port)
459 struct evtchn_close close;
462 if (HYPERVISOR_event_channel_op(EVTCHNOP_close, &close) != 0)
465 /* Closed ports are implicitly re-bound to VCPU0. */
466 bind_evtchn_to_cpu(port, 0);
469 static void pirq_query_unmask(int irq)
471 struct physdev_irq_status_query irq_status;
472 struct irq_info *info = info_for_irq(irq);
474 BUG_ON(info->type != IRQT_PIRQ);
476 irq_status.irq = pirq_from_irq(irq);
477 if (HYPERVISOR_physdev_op(PHYSDEVOP_irq_status_query, &irq_status))
478 irq_status.flags = 0;
480 info->u.pirq.flags &= ~PIRQ_NEEDS_EOI;
481 if (irq_status.flags & XENIRQSTAT_needs_eoi)
482 info->u.pirq.flags |= PIRQ_NEEDS_EOI;
485 static bool probing_irq(int irq)
487 struct irq_desc *desc = irq_to_desc(irq);
489 return desc && desc->action == NULL;
492 static void eoi_pirq(struct irq_data *data)
494 int evtchn = evtchn_from_irq(data->irq);
495 struct physdev_eoi eoi = { .irq = pirq_from_irq(data->irq) };
500 if (VALID_EVTCHN(evtchn))
501 clear_evtchn(evtchn);
503 if (pirq_needs_eoi(data->irq)) {
504 rc = HYPERVISOR_physdev_op(PHYSDEVOP_eoi, &eoi);
509 static void mask_ack_pirq(struct irq_data *data)
511 disable_dynirq(data);
515 static unsigned int __startup_pirq(unsigned int irq)
517 struct evtchn_bind_pirq bind_pirq;
518 struct irq_info *info = info_for_irq(irq);
519 int evtchn = evtchn_from_irq(irq);
522 BUG_ON(info->type != IRQT_PIRQ);
524 if (VALID_EVTCHN(evtchn))
527 bind_pirq.pirq = pirq_from_irq(irq);
528 /* NB. We are happy to share unless we are probing. */
529 bind_pirq.flags = info->u.pirq.flags & PIRQ_SHAREABLE ?
530 BIND_PIRQ__WILL_SHARE : 0;
531 rc = HYPERVISOR_event_channel_op(EVTCHNOP_bind_pirq, &bind_pirq);
533 if (!probing_irq(irq))
534 pr_info("Failed to obtain physical IRQ %d\n", irq);
537 evtchn = bind_pirq.port;
539 pirq_query_unmask(irq);
541 rc = set_evtchn_to_irq(evtchn, irq);
543 pr_err("irq%d: Failed to set port to irq mapping (%d)\n",
545 xen_evtchn_close(evtchn);
548 bind_evtchn_to_cpu(evtchn, 0);
549 info->evtchn = evtchn;
552 unmask_evtchn(evtchn);
553 eoi_pirq(irq_get_irq_data(irq));
558 static unsigned int startup_pirq(struct irq_data *data)
560 return __startup_pirq(data->irq);
563 static void shutdown_pirq(struct irq_data *data)
565 unsigned int irq = data->irq;
566 struct irq_info *info = info_for_irq(irq);
567 unsigned evtchn = evtchn_from_irq(irq);
569 BUG_ON(info->type != IRQT_PIRQ);
571 if (!VALID_EVTCHN(evtchn))
575 xen_evtchn_close(evtchn);
576 xen_irq_info_cleanup(info);
579 static void enable_pirq(struct irq_data *data)
584 static void disable_pirq(struct irq_data *data)
586 disable_dynirq(data);
589 int xen_irq_from_gsi(unsigned gsi)
591 struct irq_info *info;
593 list_for_each_entry(info, &xen_irq_list_head, list) {
594 if (info->type != IRQT_PIRQ)
597 if (info->u.pirq.gsi == gsi)
603 EXPORT_SYMBOL_GPL(xen_irq_from_gsi);
605 static void __unbind_from_irq(unsigned int irq)
607 int evtchn = evtchn_from_irq(irq);
608 struct irq_info *info = irq_get_handler_data(irq);
610 if (info->refcnt > 0) {
612 if (info->refcnt != 0)
616 if (VALID_EVTCHN(evtchn)) {
617 unsigned int cpu = cpu_from_irq(irq);
619 xen_evtchn_close(evtchn);
621 switch (type_from_irq(irq)) {
623 per_cpu(virq_to_irq, cpu)[virq_from_irq(irq)] = -1;
626 per_cpu(ipi_to_irq, cpu)[ipi_from_irq(irq)] = -1;
632 xen_irq_info_cleanup(info);
635 BUG_ON(info_for_irq(irq)->type == IRQT_UNBOUND);
641 * Do not make any assumptions regarding the relationship between the
642 * IRQ number returned here and the Xen pirq argument.
644 * Note: We don't assign an event channel until the irq actually started
645 * up. Return an existing irq if we've already got one for the gsi.
647 * Shareable implies level triggered, not shareable implies edge
650 int xen_bind_pirq_gsi_to_irq(unsigned gsi,
651 unsigned pirq, int shareable, char *name)
654 struct physdev_irq irq_op;
657 mutex_lock(&irq_mapping_update_lock);
659 irq = xen_irq_from_gsi(gsi);
661 pr_info("%s: returning irq %d for gsi %u\n",
666 irq = xen_allocate_irq_gsi(gsi);
673 /* Only the privileged domain can do this. For non-priv, the pcifront
674 * driver provides a PCI bus that does the call to do exactly
675 * this in the priv domain. */
676 if (xen_initial_domain() &&
677 HYPERVISOR_physdev_op(PHYSDEVOP_alloc_irq_vector, &irq_op)) {
683 ret = xen_irq_info_pirq_setup(irq, 0, pirq, gsi, DOMID_SELF,
684 shareable ? PIRQ_SHAREABLE : 0);
686 __unbind_from_irq(irq);
691 pirq_query_unmask(irq);
692 /* We try to use the handler with the appropriate semantic for the
693 * type of interrupt: if the interrupt is an edge triggered
694 * interrupt we use handle_edge_irq.
696 * On the other hand if the interrupt is level triggered we use
697 * handle_fasteoi_irq like the native code does for this kind of
700 * Depending on the Xen version, pirq_needs_eoi might return true
701 * not only for level triggered interrupts but for edge triggered
702 * interrupts too. In any case Xen always honors the eoi mechanism,
703 * not injecting any more pirqs of the same kind if the first one
704 * hasn't received an eoi yet. Therefore using the fasteoi handler
705 * is the right choice either way.
708 irq_set_chip_and_handler_name(irq, &xen_pirq_chip,
709 handle_fasteoi_irq, name);
711 irq_set_chip_and_handler_name(irq, &xen_pirq_chip,
712 handle_edge_irq, name);
715 mutex_unlock(&irq_mapping_update_lock);
720 #ifdef CONFIG_PCI_MSI
721 int xen_allocate_pirq_msi(struct pci_dev *dev, struct msi_desc *msidesc)
724 struct physdev_get_free_pirq op_get_free_pirq;
726 op_get_free_pirq.type = MAP_PIRQ_TYPE_MSI;
727 rc = HYPERVISOR_physdev_op(PHYSDEVOP_get_free_pirq, &op_get_free_pirq);
729 WARN_ONCE(rc == -ENOSYS,
730 "hypervisor does not support the PHYSDEVOP_get_free_pirq interface\n");
732 return rc ? -1 : op_get_free_pirq.pirq;
735 int xen_bind_pirq_msi_to_irq(struct pci_dev *dev, struct msi_desc *msidesc,
736 int pirq, const char *name, domid_t domid)
740 mutex_lock(&irq_mapping_update_lock);
742 irq = xen_allocate_irq_dynamic();
746 irq_set_chip_and_handler_name(irq, &xen_pirq_chip, handle_edge_irq,
749 ret = xen_irq_info_pirq_setup(irq, 0, pirq, 0, domid, 0);
752 ret = irq_set_msi_desc(irq, msidesc);
756 mutex_unlock(&irq_mapping_update_lock);
759 __unbind_from_irq(irq);
760 mutex_unlock(&irq_mapping_update_lock);
765 int xen_destroy_irq(int irq)
767 struct irq_desc *desc;
768 struct physdev_unmap_pirq unmap_irq;
769 struct irq_info *info = info_for_irq(irq);
772 mutex_lock(&irq_mapping_update_lock);
774 desc = irq_to_desc(irq);
778 if (xen_initial_domain()) {
779 unmap_irq.pirq = info->u.pirq.pirq;
780 unmap_irq.domid = info->u.pirq.domid;
781 rc = HYPERVISOR_physdev_op(PHYSDEVOP_unmap_pirq, &unmap_irq);
782 /* If another domain quits without making the pci_disable_msix
783 * call, the Xen hypervisor takes care of freeing the PIRQs
784 * (free_domain_pirqs).
786 if ((rc == -ESRCH && info->u.pirq.domid != DOMID_SELF))
787 pr_info("domain %d does not have %d anymore\n",
788 info->u.pirq.domid, info->u.pirq.pirq);
790 pr_warn("unmap irq failed %d\n", rc);
798 mutex_unlock(&irq_mapping_update_lock);
802 int xen_irq_from_pirq(unsigned pirq)
806 struct irq_info *info;
808 mutex_lock(&irq_mapping_update_lock);
810 list_for_each_entry(info, &xen_irq_list_head, list) {
811 if (info->type != IRQT_PIRQ)
814 if (info->u.pirq.pirq == pirq)
819 mutex_unlock(&irq_mapping_update_lock);
825 int xen_pirq_from_irq(unsigned irq)
827 return pirq_from_irq(irq);
829 EXPORT_SYMBOL_GPL(xen_pirq_from_irq);
831 int bind_evtchn_to_irq(unsigned int evtchn)
836 if (evtchn >= xen_evtchn_max_channels())
839 mutex_lock(&irq_mapping_update_lock);
841 irq = get_evtchn_to_irq(evtchn);
844 irq = xen_allocate_irq_dynamic();
848 irq_set_chip_and_handler_name(irq, &xen_dynamic_chip,
849 handle_edge_irq, "event");
851 ret = xen_irq_info_evtchn_setup(irq, evtchn);
853 __unbind_from_irq(irq);
858 struct irq_info *info = info_for_irq(irq);
859 WARN_ON(info == NULL || info->type != IRQT_EVTCHN);
863 mutex_unlock(&irq_mapping_update_lock);
867 EXPORT_SYMBOL_GPL(bind_evtchn_to_irq);
869 static int bind_ipi_to_irq(unsigned int ipi, unsigned int cpu)
871 struct evtchn_bind_ipi bind_ipi;
875 mutex_lock(&irq_mapping_update_lock);
877 irq = per_cpu(ipi_to_irq, cpu)[ipi];
880 irq = xen_allocate_irq_dynamic();
884 irq_set_chip_and_handler_name(irq, &xen_percpu_chip,
885 handle_percpu_irq, "ipi");
888 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_ipi,
891 evtchn = bind_ipi.port;
893 ret = xen_irq_info_ipi_setup(cpu, irq, evtchn, ipi);
895 __unbind_from_irq(irq);
899 bind_evtchn_to_cpu(evtchn, cpu);
901 struct irq_info *info = info_for_irq(irq);
902 WARN_ON(info == NULL || info->type != IRQT_IPI);
906 mutex_unlock(&irq_mapping_update_lock);
910 static int bind_interdomain_evtchn_to_irq(unsigned int remote_domain,
911 unsigned int remote_port)
913 struct evtchn_bind_interdomain bind_interdomain;
916 bind_interdomain.remote_dom = remote_domain;
917 bind_interdomain.remote_port = remote_port;
919 err = HYPERVISOR_event_channel_op(EVTCHNOP_bind_interdomain,
922 return err ? : bind_evtchn_to_irq(bind_interdomain.local_port);
925 static int find_virq(unsigned int virq, unsigned int cpu)
927 struct evtchn_status status;
928 int port, rc = -ENOENT;
930 memset(&status, 0, sizeof(status));
931 for (port = 0; port < xen_evtchn_max_channels(); port++) {
932 status.dom = DOMID_SELF;
934 rc = HYPERVISOR_event_channel_op(EVTCHNOP_status, &status);
937 if (status.status != EVTCHNSTAT_virq)
939 if (status.u.virq == virq && status.vcpu == cpu) {
947 int bind_virq_to_irq(unsigned int virq, unsigned int cpu)
949 struct evtchn_bind_virq bind_virq;
950 int evtchn, irq, ret;
952 mutex_lock(&irq_mapping_update_lock);
954 irq = per_cpu(virq_to_irq, cpu)[virq];
957 irq = xen_allocate_irq_dynamic();
961 irq_set_chip_and_handler_name(irq, &xen_percpu_chip,
962 handle_percpu_irq, "virq");
964 bind_virq.virq = virq;
965 bind_virq.vcpu = cpu;
966 ret = HYPERVISOR_event_channel_op(EVTCHNOP_bind_virq,
969 evtchn = bind_virq.port;
972 ret = find_virq(virq, cpu);
977 ret = xen_irq_info_virq_setup(cpu, irq, evtchn, virq);
979 __unbind_from_irq(irq);
984 bind_evtchn_to_cpu(evtchn, cpu);
986 struct irq_info *info = info_for_irq(irq);
987 WARN_ON(info == NULL || info->type != IRQT_VIRQ);
991 mutex_unlock(&irq_mapping_update_lock);
996 static void unbind_from_irq(unsigned int irq)
998 mutex_lock(&irq_mapping_update_lock);
999 __unbind_from_irq(irq);
1000 mutex_unlock(&irq_mapping_update_lock);
1003 int bind_evtchn_to_irqhandler(unsigned int evtchn,
1004 irq_handler_t handler,
1005 unsigned long irqflags,
1006 const char *devname, void *dev_id)
1010 irq = bind_evtchn_to_irq(evtchn);
1013 retval = request_irq(irq, handler, irqflags, devname, dev_id);
1015 unbind_from_irq(irq);
1021 EXPORT_SYMBOL_GPL(bind_evtchn_to_irqhandler);
1023 int bind_interdomain_evtchn_to_irqhandler(unsigned int remote_domain,
1024 unsigned int remote_port,
1025 irq_handler_t handler,
1026 unsigned long irqflags,
1027 const char *devname,
1032 irq = bind_interdomain_evtchn_to_irq(remote_domain, remote_port);
1036 retval = request_irq(irq, handler, irqflags, devname, dev_id);
1038 unbind_from_irq(irq);
1044 EXPORT_SYMBOL_GPL(bind_interdomain_evtchn_to_irqhandler);
1046 int bind_virq_to_irqhandler(unsigned int virq, unsigned int cpu,
1047 irq_handler_t handler,
1048 unsigned long irqflags, const char *devname, void *dev_id)
1052 irq = bind_virq_to_irq(virq, cpu);
1055 retval = request_irq(irq, handler, irqflags, devname, dev_id);
1057 unbind_from_irq(irq);
1063 EXPORT_SYMBOL_GPL(bind_virq_to_irqhandler);
1065 int bind_ipi_to_irqhandler(enum ipi_vector ipi,
1067 irq_handler_t handler,
1068 unsigned long irqflags,
1069 const char *devname,
1074 irq = bind_ipi_to_irq(ipi, cpu);
1078 irqflags |= IRQF_NO_SUSPEND | IRQF_FORCE_RESUME | IRQF_EARLY_RESUME;
1079 retval = request_irq(irq, handler, irqflags, devname, dev_id);
1081 unbind_from_irq(irq);
1088 void unbind_from_irqhandler(unsigned int irq, void *dev_id)
1090 struct irq_info *info = irq_get_handler_data(irq);
1094 free_irq(irq, dev_id);
1095 unbind_from_irq(irq);
1097 EXPORT_SYMBOL_GPL(unbind_from_irqhandler);
1099 int evtchn_make_refcounted(unsigned int evtchn)
1101 int irq = get_evtchn_to_irq(evtchn);
1102 struct irq_info *info;
1107 info = irq_get_handler_data(irq);
1112 WARN_ON(info->refcnt != -1);
1118 EXPORT_SYMBOL_GPL(evtchn_make_refcounted);
1120 int evtchn_get(unsigned int evtchn)
1123 struct irq_info *info;
1126 if (evtchn >= xen_evtchn_max_channels())
1129 mutex_lock(&irq_mapping_update_lock);
1131 irq = get_evtchn_to_irq(evtchn);
1135 info = irq_get_handler_data(irq);
1141 if (info->refcnt <= 0)
1147 mutex_unlock(&irq_mapping_update_lock);
1151 EXPORT_SYMBOL_GPL(evtchn_get);
1153 void evtchn_put(unsigned int evtchn)
1155 int irq = get_evtchn_to_irq(evtchn);
1156 if (WARN_ON(irq == -1))
1158 unbind_from_irq(irq);
1160 EXPORT_SYMBOL_GPL(evtchn_put);
1162 void xen_send_IPI_one(unsigned int cpu, enum ipi_vector vector)
1167 if (unlikely(vector == XEN_NMI_VECTOR)) {
1168 int rc = HYPERVISOR_vcpu_op(VCPUOP_send_nmi, cpu, NULL);
1170 printk(KERN_WARNING "Sending nmi to CPU%d failed (rc:%d)\n", cpu, rc);
1174 irq = per_cpu(ipi_to_irq, cpu)[vector];
1176 notify_remote_via_irq(irq);
1179 static DEFINE_PER_CPU(unsigned, xed_nesting_count);
1181 static void __xen_evtchn_do_upcall(void)
1183 struct vcpu_info *vcpu_info = __this_cpu_read(xen_vcpu);
1184 int cpu = get_cpu();
1188 vcpu_info->evtchn_upcall_pending = 0;
1190 if (__this_cpu_inc_return(xed_nesting_count) - 1)
1193 xen_evtchn_handle_events(cpu);
1195 BUG_ON(!irqs_disabled());
1197 count = __this_cpu_read(xed_nesting_count);
1198 __this_cpu_write(xed_nesting_count, 0);
1199 } while (count != 1 || vcpu_info->evtchn_upcall_pending);
1206 void xen_evtchn_do_upcall(struct pt_regs *regs)
1208 struct pt_regs *old_regs = set_irq_regs(regs);
1215 __xen_evtchn_do_upcall();
1218 set_irq_regs(old_regs);
1221 void xen_hvm_evtchn_do_upcall(void)
1223 __xen_evtchn_do_upcall();
1225 EXPORT_SYMBOL_GPL(xen_hvm_evtchn_do_upcall);
1227 /* Rebind a new event channel to an existing irq. */
1228 void rebind_evtchn_irq(int evtchn, int irq)
1230 struct irq_info *info = info_for_irq(irq);
1235 /* Make sure the irq is masked, since the new event channel
1236 will also be masked. */
1239 mutex_lock(&irq_mapping_update_lock);
1241 /* After resume the irq<->evtchn mappings are all cleared out */
1242 BUG_ON(get_evtchn_to_irq(evtchn) != -1);
1243 /* Expect irq to have been bound before,
1244 so there should be a proper type */
1245 BUG_ON(info->type == IRQT_UNBOUND);
1247 (void)xen_irq_info_evtchn_setup(irq, evtchn);
1249 mutex_unlock(&irq_mapping_update_lock);
1251 /* new event channels are always bound to cpu 0 */
1252 irq_set_affinity(irq, cpumask_of(0));
1254 /* Unmask the event channel. */
1258 /* Rebind an evtchn so that it gets delivered to a specific cpu */
1259 static int rebind_irq_to_cpu(unsigned irq, unsigned tcpu)
1261 struct evtchn_bind_vcpu bind_vcpu;
1262 int evtchn = evtchn_from_irq(irq);
1265 if (!VALID_EVTCHN(evtchn))
1269 * Events delivered via platform PCI interrupts are always
1270 * routed to vcpu 0 and hence cannot be rebound.
1272 if (xen_hvm_domain() && !xen_have_vector_callback)
1275 /* Send future instances of this interrupt to other vcpu. */
1276 bind_vcpu.port = evtchn;
1277 bind_vcpu.vcpu = tcpu;
1280 * Mask the event while changing the VCPU binding to prevent
1281 * it being delivered on an unexpected VCPU.
1283 masked = test_and_set_mask(evtchn);
1286 * If this fails, it usually just indicates that we're dealing with a
1287 * virq or IPI channel, which don't actually need to be rebound. Ignore
1288 * it, but don't do the xenlinux-level rebind in that case.
1290 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_vcpu, &bind_vcpu) >= 0)
1291 bind_evtchn_to_cpu(evtchn, tcpu);
1294 unmask_evtchn(evtchn);
1299 static int set_affinity_irq(struct irq_data *data, const struct cpumask *dest,
1302 unsigned tcpu = cpumask_first(dest);
1304 return rebind_irq_to_cpu(data->irq, tcpu);
1307 static int retrigger_evtchn(int evtchn)
1311 if (!VALID_EVTCHN(evtchn))
1314 masked = test_and_set_mask(evtchn);
1317 unmask_evtchn(evtchn);
1322 int resend_irq_on_evtchn(unsigned int irq)
1324 return retrigger_evtchn(evtchn_from_irq(irq));
1327 static void enable_dynirq(struct irq_data *data)
1329 int evtchn = evtchn_from_irq(data->irq);
1331 if (VALID_EVTCHN(evtchn))
1332 unmask_evtchn(evtchn);
1335 static void disable_dynirq(struct irq_data *data)
1337 int evtchn = evtchn_from_irq(data->irq);
1339 if (VALID_EVTCHN(evtchn))
1340 mask_evtchn(evtchn);
1343 static void ack_dynirq(struct irq_data *data)
1345 int evtchn = evtchn_from_irq(data->irq);
1349 if (VALID_EVTCHN(evtchn))
1350 clear_evtchn(evtchn);
1353 static void mask_ack_dynirq(struct irq_data *data)
1355 disable_dynirq(data);
1359 static int retrigger_dynirq(struct irq_data *data)
1361 return retrigger_evtchn(evtchn_from_irq(data->irq));
1364 static void restore_pirqs(void)
1366 int pirq, rc, irq, gsi;
1367 struct physdev_map_pirq map_irq;
1368 struct irq_info *info;
1370 list_for_each_entry(info, &xen_irq_list_head, list) {
1371 if (info->type != IRQT_PIRQ)
1374 pirq = info->u.pirq.pirq;
1375 gsi = info->u.pirq.gsi;
1378 /* save/restore of PT devices doesn't work, so at this point the
1379 * only devices present are GSI based emulated devices */
1383 map_irq.domid = DOMID_SELF;
1384 map_irq.type = MAP_PIRQ_TYPE_GSI;
1385 map_irq.index = gsi;
1386 map_irq.pirq = pirq;
1388 rc = HYPERVISOR_physdev_op(PHYSDEVOP_map_pirq, &map_irq);
1390 pr_warn("xen map irq failed gsi=%d irq=%d pirq=%d rc=%d\n",
1391 gsi, irq, pirq, rc);
1396 printk(KERN_DEBUG "xen: --> irq=%d, pirq=%d\n", irq, map_irq.pirq);
1398 __startup_pirq(irq);
1402 static void restore_cpu_virqs(unsigned int cpu)
1404 struct evtchn_bind_virq bind_virq;
1405 int virq, irq, evtchn;
1407 for (virq = 0; virq < NR_VIRQS; virq++) {
1408 if ((irq = per_cpu(virq_to_irq, cpu)[virq]) == -1)
1411 BUG_ON(virq_from_irq(irq) != virq);
1413 /* Get a new binding from Xen. */
1414 bind_virq.virq = virq;
1415 bind_virq.vcpu = cpu;
1416 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_virq,
1419 evtchn = bind_virq.port;
1421 /* Record the new mapping. */
1422 (void)xen_irq_info_virq_setup(cpu, irq, evtchn, virq);
1423 bind_evtchn_to_cpu(evtchn, cpu);
1427 static void restore_cpu_ipis(unsigned int cpu)
1429 struct evtchn_bind_ipi bind_ipi;
1430 int ipi, irq, evtchn;
1432 for (ipi = 0; ipi < XEN_NR_IPIS; ipi++) {
1433 if ((irq = per_cpu(ipi_to_irq, cpu)[ipi]) == -1)
1436 BUG_ON(ipi_from_irq(irq) != ipi);
1438 /* Get a new binding from Xen. */
1439 bind_ipi.vcpu = cpu;
1440 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_ipi,
1443 evtchn = bind_ipi.port;
1445 /* Record the new mapping. */
1446 (void)xen_irq_info_ipi_setup(cpu, irq, evtchn, ipi);
1447 bind_evtchn_to_cpu(evtchn, cpu);
1451 /* Clear an irq's pending state, in preparation for polling on it */
1452 void xen_clear_irq_pending(int irq)
1454 int evtchn = evtchn_from_irq(irq);
1456 if (VALID_EVTCHN(evtchn))
1457 clear_evtchn(evtchn);
1459 EXPORT_SYMBOL(xen_clear_irq_pending);
1460 void xen_set_irq_pending(int irq)
1462 int evtchn = evtchn_from_irq(irq);
1464 if (VALID_EVTCHN(evtchn))
1468 bool xen_test_irq_pending(int irq)
1470 int evtchn = evtchn_from_irq(irq);
1473 if (VALID_EVTCHN(evtchn))
1474 ret = test_evtchn(evtchn);
1479 /* Poll waiting for an irq to become pending with timeout. In the usual case,
1480 * the irq will be disabled so it won't deliver an interrupt. */
1481 void xen_poll_irq_timeout(int irq, u64 timeout)
1483 evtchn_port_t evtchn = evtchn_from_irq(irq);
1485 if (VALID_EVTCHN(evtchn)) {
1486 struct sched_poll poll;
1489 poll.timeout = timeout;
1490 set_xen_guest_handle(poll.ports, &evtchn);
1492 if (HYPERVISOR_sched_op(SCHEDOP_poll, &poll) != 0)
1496 EXPORT_SYMBOL(xen_poll_irq_timeout);
1497 /* Poll waiting for an irq to become pending. In the usual case, the
1498 * irq will be disabled so it won't deliver an interrupt. */
1499 void xen_poll_irq(int irq)
1501 xen_poll_irq_timeout(irq, 0 /* no timeout */);
1504 /* Check whether the IRQ line is shared with other guests. */
1505 int xen_test_irq_shared(int irq)
1507 struct irq_info *info = info_for_irq(irq);
1508 struct physdev_irq_status_query irq_status;
1513 irq_status.irq = info->u.pirq.pirq;
1515 if (HYPERVISOR_physdev_op(PHYSDEVOP_irq_status_query, &irq_status))
1517 return !(irq_status.flags & XENIRQSTAT_shared);
1519 EXPORT_SYMBOL_GPL(xen_test_irq_shared);
1521 void xen_irq_resume(void)
1523 unsigned int cpu, evtchn;
1524 struct irq_info *info;
1526 /* New event-channel space is not 'live' yet. */
1527 for (evtchn = 0; evtchn < xen_evtchn_nr_channels(); evtchn++)
1528 mask_evtchn(evtchn);
1530 /* No IRQ <-> event-channel mappings. */
1531 list_for_each_entry(info, &xen_irq_list_head, list)
1532 info->evtchn = 0; /* zap event-channel binding */
1534 clear_evtchn_to_irq_all();
1536 for_each_possible_cpu(cpu) {
1537 restore_cpu_virqs(cpu);
1538 restore_cpu_ipis(cpu);
1544 static struct irq_chip xen_dynamic_chip __read_mostly = {
1547 .irq_disable = disable_dynirq,
1548 .irq_mask = disable_dynirq,
1549 .irq_unmask = enable_dynirq,
1551 .irq_ack = ack_dynirq,
1552 .irq_mask_ack = mask_ack_dynirq,
1554 .irq_set_affinity = set_affinity_irq,
1555 .irq_retrigger = retrigger_dynirq,
1558 static struct irq_chip xen_pirq_chip __read_mostly = {
1561 .irq_startup = startup_pirq,
1562 .irq_shutdown = shutdown_pirq,
1563 .irq_enable = enable_pirq,
1564 .irq_disable = disable_pirq,
1566 .irq_mask = disable_dynirq,
1567 .irq_unmask = enable_dynirq,
1569 .irq_ack = eoi_pirq,
1570 .irq_eoi = eoi_pirq,
1571 .irq_mask_ack = mask_ack_pirq,
1573 .irq_set_affinity = set_affinity_irq,
1575 .irq_retrigger = retrigger_dynirq,
1578 static struct irq_chip xen_percpu_chip __read_mostly = {
1579 .name = "xen-percpu",
1581 .irq_disable = disable_dynirq,
1582 .irq_mask = disable_dynirq,
1583 .irq_unmask = enable_dynirq,
1585 .irq_ack = ack_dynirq,
1588 int xen_set_callback_via(uint64_t via)
1590 struct xen_hvm_param a;
1591 a.domid = DOMID_SELF;
1592 a.index = HVM_PARAM_CALLBACK_IRQ;
1594 return HYPERVISOR_hvm_op(HVMOP_set_param, &a);
1596 EXPORT_SYMBOL_GPL(xen_set_callback_via);
1598 #ifdef CONFIG_XEN_PVHVM
1599 /* Vector callbacks are better than PCI interrupts to receive event
1600 * channel notifications because we can receive vector callbacks on any
1601 * vcpu and we don't need PCI support or APIC interactions. */
1602 void xen_callback_vector(void)
1605 uint64_t callback_via;
1606 if (xen_have_vector_callback) {
1607 callback_via = HVM_CALLBACK_VECTOR(HYPERVISOR_CALLBACK_VECTOR);
1608 rc = xen_set_callback_via(callback_via);
1610 pr_err("Request for Xen HVM callback vector failed\n");
1611 xen_have_vector_callback = 0;
1614 pr_info("Xen HVM callback vector for event delivery is enabled\n");
1615 /* in the restore case the vector has already been allocated */
1616 if (!test_bit(HYPERVISOR_CALLBACK_VECTOR, used_vectors))
1617 alloc_intr_gate(HYPERVISOR_CALLBACK_VECTOR,
1618 xen_hvm_callback_vector);
1622 void xen_callback_vector(void) {}
1625 void __init xen_init_IRQ(void)
1629 xen_evtchn_2l_init();
1631 evtchn_to_irq = kcalloc(EVTCHN_ROW(xen_evtchn_max_channels()),
1632 sizeof(*evtchn_to_irq), GFP_KERNEL);
1633 BUG_ON(!evtchn_to_irq);
1635 /* No event channels are 'live' right now. */
1636 for (i = 0; i < xen_evtchn_nr_channels(); i++)
1639 pirq_needs_eoi = pirq_needs_eoi_flag;
1642 if (xen_hvm_domain()) {
1643 xen_callback_vector();
1645 /* pci_xen_hvm_init must be called after native_init_IRQ so that
1646 * __acpi_register_gsi can point at the right function */
1650 struct physdev_pirq_eoi_gmfn eoi_gmfn;
1652 irq_ctx_init(smp_processor_id());
1653 if (xen_initial_domain())
1654 pci_xen_initial_domain();
1656 pirq_eoi_map = (void *)__get_free_page(GFP_KERNEL|__GFP_ZERO);
1657 eoi_gmfn.gmfn = virt_to_mfn(pirq_eoi_map);
1658 rc = HYPERVISOR_physdev_op(PHYSDEVOP_pirq_eoi_gmfn_v2, &eoi_gmfn);
1660 free_page((unsigned long) pirq_eoi_map);
1661 pirq_eoi_map = NULL;
1663 pirq_needs_eoi = pirq_check_eoi_map;