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/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 >= 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_get_irq_data(irq)->affinity, cpumask_of(cpu));
341 xen_evtchn_port_bind_to_cpu(info, cpu);
346 static void xen_evtchn_mask_all(void)
350 for (evtchn = 0; evtchn < xen_evtchn_nr_channels(); evtchn++)
355 * notify_remote_via_irq - send event to remote end of event channel via irq
356 * @irq: irq of event channel to send event to
358 * Unlike notify_remote_via_evtchn(), this is safe to use across
359 * save/restore. Notifications on a broken connection are silently
362 void notify_remote_via_irq(int irq)
364 int evtchn = evtchn_from_irq(irq);
366 if (VALID_EVTCHN(evtchn))
367 notify_remote_via_evtchn(evtchn);
369 EXPORT_SYMBOL_GPL(notify_remote_via_irq);
371 static void xen_irq_init(unsigned irq)
373 struct irq_info *info;
375 /* By default all event channels notify CPU#0. */
376 cpumask_copy(irq_get_irq_data(irq)->affinity, cpumask_of(0));
379 info = kzalloc(sizeof(*info), GFP_KERNEL);
381 panic("Unable to allocate metadata for IRQ%d\n", irq);
383 info->type = IRQT_UNBOUND;
386 irq_set_handler_data(irq, info);
388 list_add_tail(&info->list, &xen_irq_list_head);
391 static int __must_check xen_allocate_irqs_dynamic(int nvec)
393 int i, irq = irq_alloc_descs(-1, 0, nvec, -1);
396 for (i = 0; i < nvec; i++)
397 xen_irq_init(irq + i);
403 static inline int __must_check xen_allocate_irq_dynamic(void)
406 return xen_allocate_irqs_dynamic(1);
409 static int __must_check xen_allocate_irq_gsi(unsigned gsi)
414 * A PV guest has no concept of a GSI (since it has no ACPI
415 * nor access to/knowledge of the physical APICs). Therefore
416 * all IRQs are dynamically allocated from the entire IRQ
419 if (xen_pv_domain() && !xen_initial_domain())
420 return xen_allocate_irq_dynamic();
422 /* Legacy IRQ descriptors are already allocated by the arch. */
423 if (gsi < NR_IRQS_LEGACY)
426 irq = irq_alloc_desc_at(gsi, -1);
433 static void xen_free_irq(unsigned irq)
435 struct irq_info *info = irq_get_handler_data(irq);
440 list_del(&info->list);
442 irq_set_handler_data(irq, NULL);
444 WARN_ON(info->refcnt > 0);
448 /* Legacy IRQ descriptors are managed by the arch. */
449 if (irq < NR_IRQS_LEGACY)
455 static void xen_evtchn_close(unsigned int port)
457 struct evtchn_close close;
460 if (HYPERVISOR_event_channel_op(EVTCHNOP_close, &close) != 0)
464 static void pirq_query_unmask(int irq)
466 struct physdev_irq_status_query irq_status;
467 struct irq_info *info = info_for_irq(irq);
469 BUG_ON(info->type != IRQT_PIRQ);
471 irq_status.irq = pirq_from_irq(irq);
472 if (HYPERVISOR_physdev_op(PHYSDEVOP_irq_status_query, &irq_status))
473 irq_status.flags = 0;
475 info->u.pirq.flags &= ~PIRQ_NEEDS_EOI;
476 if (irq_status.flags & XENIRQSTAT_needs_eoi)
477 info->u.pirq.flags |= PIRQ_NEEDS_EOI;
480 static void eoi_pirq(struct irq_data *data)
482 int evtchn = evtchn_from_irq(data->irq);
483 struct physdev_eoi eoi = { .irq = pirq_from_irq(data->irq) };
488 if (VALID_EVTCHN(evtchn))
489 clear_evtchn(evtchn);
491 if (pirq_needs_eoi(data->irq)) {
492 rc = HYPERVISOR_physdev_op(PHYSDEVOP_eoi, &eoi);
497 static void mask_ack_pirq(struct irq_data *data)
499 disable_dynirq(data);
503 static unsigned int __startup_pirq(unsigned int irq)
505 struct evtchn_bind_pirq bind_pirq;
506 struct irq_info *info = info_for_irq(irq);
507 int evtchn = evtchn_from_irq(irq);
510 BUG_ON(info->type != IRQT_PIRQ);
512 if (VALID_EVTCHN(evtchn))
515 bind_pirq.pirq = pirq_from_irq(irq);
516 /* NB. We are happy to share unless we are probing. */
517 bind_pirq.flags = info->u.pirq.flags & PIRQ_SHAREABLE ?
518 BIND_PIRQ__WILL_SHARE : 0;
519 rc = HYPERVISOR_event_channel_op(EVTCHNOP_bind_pirq, &bind_pirq);
521 pr_warn("Failed to obtain physical IRQ %d\n", irq);
524 evtchn = bind_pirq.port;
526 pirq_query_unmask(irq);
528 rc = set_evtchn_to_irq(evtchn, irq);
532 info->evtchn = evtchn;
533 bind_evtchn_to_cpu(evtchn, 0);
535 rc = xen_evtchn_port_setup(info);
540 unmask_evtchn(evtchn);
541 eoi_pirq(irq_get_irq_data(irq));
546 pr_err("irq%d: Failed to set port to irq mapping (%d)\n", irq, rc);
547 xen_evtchn_close(evtchn);
551 static unsigned int startup_pirq(struct irq_data *data)
553 return __startup_pirq(data->irq);
556 static void shutdown_pirq(struct irq_data *data)
558 unsigned int irq = data->irq;
559 struct irq_info *info = info_for_irq(irq);
560 unsigned evtchn = evtchn_from_irq(irq);
562 BUG_ON(info->type != IRQT_PIRQ);
564 if (!VALID_EVTCHN(evtchn))
568 xen_evtchn_close(evtchn);
569 xen_irq_info_cleanup(info);
572 static void enable_pirq(struct irq_data *data)
577 static void disable_pirq(struct irq_data *data)
579 disable_dynirq(data);
582 int xen_irq_from_gsi(unsigned gsi)
584 struct irq_info *info;
586 list_for_each_entry(info, &xen_irq_list_head, list) {
587 if (info->type != IRQT_PIRQ)
590 if (info->u.pirq.gsi == gsi)
596 EXPORT_SYMBOL_GPL(xen_irq_from_gsi);
598 static void __unbind_from_irq(unsigned int irq)
600 int evtchn = evtchn_from_irq(irq);
601 struct irq_info *info = irq_get_handler_data(irq);
603 if (info->refcnt > 0) {
605 if (info->refcnt != 0)
609 if (VALID_EVTCHN(evtchn)) {
610 unsigned int cpu = cpu_from_irq(irq);
612 xen_evtchn_close(evtchn);
614 switch (type_from_irq(irq)) {
616 per_cpu(virq_to_irq, cpu)[virq_from_irq(irq)] = -1;
619 per_cpu(ipi_to_irq, cpu)[ipi_from_irq(irq)] = -1;
625 xen_irq_info_cleanup(info);
628 BUG_ON(info_for_irq(irq)->type == IRQT_UNBOUND);
634 * Do not make any assumptions regarding the relationship between the
635 * IRQ number returned here and the Xen pirq argument.
637 * Note: We don't assign an event channel until the irq actually started
638 * up. Return an existing irq if we've already got one for the gsi.
640 * Shareable implies level triggered, not shareable implies edge
643 int xen_bind_pirq_gsi_to_irq(unsigned gsi,
644 unsigned pirq, int shareable, char *name)
647 struct physdev_irq irq_op;
650 mutex_lock(&irq_mapping_update_lock);
652 irq = xen_irq_from_gsi(gsi);
654 pr_info("%s: returning irq %d for gsi %u\n",
659 irq = xen_allocate_irq_gsi(gsi);
666 /* Only the privileged domain can do this. For non-priv, the pcifront
667 * driver provides a PCI bus that does the call to do exactly
668 * this in the priv domain. */
669 if (xen_initial_domain() &&
670 HYPERVISOR_physdev_op(PHYSDEVOP_alloc_irq_vector, &irq_op)) {
676 ret = xen_irq_info_pirq_setup(irq, 0, pirq, gsi, DOMID_SELF,
677 shareable ? PIRQ_SHAREABLE : 0);
679 __unbind_from_irq(irq);
684 pirq_query_unmask(irq);
685 /* We try to use the handler with the appropriate semantic for the
686 * type of interrupt: if the interrupt is an edge triggered
687 * interrupt we use handle_edge_irq.
689 * On the other hand if the interrupt is level triggered we use
690 * handle_fasteoi_irq like the native code does for this kind of
693 * Depending on the Xen version, pirq_needs_eoi might return true
694 * not only for level triggered interrupts but for edge triggered
695 * interrupts too. In any case Xen always honors the eoi mechanism,
696 * not injecting any more pirqs of the same kind if the first one
697 * hasn't received an eoi yet. Therefore using the fasteoi handler
698 * is the right choice either way.
701 irq_set_chip_and_handler_name(irq, &xen_pirq_chip,
702 handle_fasteoi_irq, name);
704 irq_set_chip_and_handler_name(irq, &xen_pirq_chip,
705 handle_edge_irq, name);
708 mutex_unlock(&irq_mapping_update_lock);
713 #ifdef CONFIG_PCI_MSI
714 int xen_allocate_pirq_msi(struct pci_dev *dev, struct msi_desc *msidesc)
717 struct physdev_get_free_pirq op_get_free_pirq;
719 op_get_free_pirq.type = MAP_PIRQ_TYPE_MSI;
720 rc = HYPERVISOR_physdev_op(PHYSDEVOP_get_free_pirq, &op_get_free_pirq);
722 WARN_ONCE(rc == -ENOSYS,
723 "hypervisor does not support the PHYSDEVOP_get_free_pirq interface\n");
725 return rc ? -1 : op_get_free_pirq.pirq;
728 int xen_bind_pirq_msi_to_irq(struct pci_dev *dev, struct msi_desc *msidesc,
729 int pirq, int nvec, const char *name, domid_t domid)
733 mutex_lock(&irq_mapping_update_lock);
735 irq = xen_allocate_irqs_dynamic(nvec);
739 for (i = 0; i < nvec; i++) {
740 irq_set_chip_and_handler_name(irq + i, &xen_pirq_chip, handle_edge_irq, name);
742 ret = xen_irq_info_pirq_setup(irq + i, 0, pirq + i, 0, domid,
743 i == 0 ? 0 : PIRQ_MSI_GROUP);
748 ret = irq_set_msi_desc(irq, msidesc);
752 mutex_unlock(&irq_mapping_update_lock);
756 __unbind_from_irq(irq + i);
757 mutex_unlock(&irq_mapping_update_lock);
762 int xen_destroy_irq(int irq)
764 struct physdev_unmap_pirq unmap_irq;
765 struct irq_info *info = info_for_irq(irq);
768 mutex_lock(&irq_mapping_update_lock);
771 * If trying to remove a vector in a MSI group different
772 * than the first one skip the PIRQ unmap unless this vector
773 * is the first one in the group.
775 if (xen_initial_domain() && !(info->u.pirq.flags & PIRQ_MSI_GROUP)) {
776 unmap_irq.pirq = info->u.pirq.pirq;
777 unmap_irq.domid = info->u.pirq.domid;
778 rc = HYPERVISOR_physdev_op(PHYSDEVOP_unmap_pirq, &unmap_irq);
779 /* If another domain quits without making the pci_disable_msix
780 * call, the Xen hypervisor takes care of freeing the PIRQs
781 * (free_domain_pirqs).
783 if ((rc == -ESRCH && info->u.pirq.domid != DOMID_SELF))
784 pr_info("domain %d does not have %d anymore\n",
785 info->u.pirq.domid, info->u.pirq.pirq);
787 pr_warn("unmap irq failed %d\n", rc);
795 mutex_unlock(&irq_mapping_update_lock);
799 int xen_irq_from_pirq(unsigned pirq)
803 struct irq_info *info;
805 mutex_lock(&irq_mapping_update_lock);
807 list_for_each_entry(info, &xen_irq_list_head, list) {
808 if (info->type != IRQT_PIRQ)
811 if (info->u.pirq.pirq == pirq)
816 mutex_unlock(&irq_mapping_update_lock);
822 int xen_pirq_from_irq(unsigned irq)
824 return pirq_from_irq(irq);
826 EXPORT_SYMBOL_GPL(xen_pirq_from_irq);
828 int bind_evtchn_to_irq(unsigned int evtchn)
833 if (evtchn >= xen_evtchn_max_channels())
836 mutex_lock(&irq_mapping_update_lock);
838 irq = get_evtchn_to_irq(evtchn);
841 irq = xen_allocate_irq_dynamic();
845 irq_set_chip_and_handler_name(irq, &xen_dynamic_chip,
846 handle_edge_irq, "event");
848 ret = xen_irq_info_evtchn_setup(irq, evtchn);
850 __unbind_from_irq(irq);
854 /* New interdomain events are bound to VCPU 0. */
855 bind_evtchn_to_cpu(evtchn, 0);
857 struct irq_info *info = info_for_irq(irq);
858 WARN_ON(info == NULL || info->type != IRQT_EVTCHN);
862 mutex_unlock(&irq_mapping_update_lock);
866 EXPORT_SYMBOL_GPL(bind_evtchn_to_irq);
868 static int bind_ipi_to_irq(unsigned int ipi, unsigned int cpu)
870 struct evtchn_bind_ipi bind_ipi;
874 mutex_lock(&irq_mapping_update_lock);
876 irq = per_cpu(ipi_to_irq, cpu)[ipi];
879 irq = xen_allocate_irq_dynamic();
883 irq_set_chip_and_handler_name(irq, &xen_percpu_chip,
884 handle_percpu_irq, "ipi");
887 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_ipi,
890 evtchn = bind_ipi.port;
892 ret = xen_irq_info_ipi_setup(cpu, irq, evtchn, ipi);
894 __unbind_from_irq(irq);
898 bind_evtchn_to_cpu(evtchn, cpu);
900 struct irq_info *info = info_for_irq(irq);
901 WARN_ON(info == NULL || info->type != IRQT_IPI);
905 mutex_unlock(&irq_mapping_update_lock);
909 int bind_interdomain_evtchn_to_irq(unsigned int remote_domain,
910 unsigned int remote_port)
912 struct evtchn_bind_interdomain bind_interdomain;
915 bind_interdomain.remote_dom = remote_domain;
916 bind_interdomain.remote_port = remote_port;
918 err = HYPERVISOR_event_channel_op(EVTCHNOP_bind_interdomain,
921 return err ? : bind_evtchn_to_irq(bind_interdomain.local_port);
923 EXPORT_SYMBOL_GPL(bind_interdomain_evtchn_to_irq);
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) {
948 * xen_evtchn_nr_channels - number of usable event channel ports
950 * This may be less than the maximum supported by the current
951 * hypervisor ABI. Use xen_evtchn_max_channels() for the maximum
954 unsigned xen_evtchn_nr_channels(void)
956 return evtchn_ops->nr_channels();
958 EXPORT_SYMBOL_GPL(xen_evtchn_nr_channels);
960 int bind_virq_to_irq(unsigned int virq, unsigned int cpu, bool percpu)
962 struct evtchn_bind_virq bind_virq;
963 int evtchn, irq, ret;
965 mutex_lock(&irq_mapping_update_lock);
967 irq = per_cpu(virq_to_irq, cpu)[virq];
970 irq = xen_allocate_irq_dynamic();
975 irq_set_chip_and_handler_name(irq, &xen_percpu_chip,
976 handle_percpu_irq, "virq");
978 irq_set_chip_and_handler_name(irq, &xen_dynamic_chip,
979 handle_edge_irq, "virq");
981 bind_virq.virq = virq;
982 bind_virq.vcpu = cpu;
983 ret = HYPERVISOR_event_channel_op(EVTCHNOP_bind_virq,
986 evtchn = bind_virq.port;
989 ret = find_virq(virq, cpu);
994 ret = xen_irq_info_virq_setup(cpu, irq, evtchn, virq);
996 __unbind_from_irq(irq);
1001 bind_evtchn_to_cpu(evtchn, cpu);
1003 struct irq_info *info = info_for_irq(irq);
1004 WARN_ON(info == NULL || info->type != IRQT_VIRQ);
1008 mutex_unlock(&irq_mapping_update_lock);
1013 static void unbind_from_irq(unsigned int irq)
1015 mutex_lock(&irq_mapping_update_lock);
1016 __unbind_from_irq(irq);
1017 mutex_unlock(&irq_mapping_update_lock);
1020 int bind_evtchn_to_irqhandler(unsigned int evtchn,
1021 irq_handler_t handler,
1022 unsigned long irqflags,
1023 const char *devname, void *dev_id)
1027 irq = bind_evtchn_to_irq(evtchn);
1030 retval = request_irq(irq, handler, irqflags, devname, dev_id);
1032 unbind_from_irq(irq);
1038 EXPORT_SYMBOL_GPL(bind_evtchn_to_irqhandler);
1040 int bind_interdomain_evtchn_to_irqhandler(unsigned int remote_domain,
1041 unsigned int remote_port,
1042 irq_handler_t handler,
1043 unsigned long irqflags,
1044 const char *devname,
1049 irq = bind_interdomain_evtchn_to_irq(remote_domain, remote_port);
1053 retval = request_irq(irq, handler, irqflags, devname, dev_id);
1055 unbind_from_irq(irq);
1061 EXPORT_SYMBOL_GPL(bind_interdomain_evtchn_to_irqhandler);
1063 int bind_virq_to_irqhandler(unsigned int virq, unsigned int cpu,
1064 irq_handler_t handler,
1065 unsigned long irqflags, const char *devname, void *dev_id)
1069 irq = bind_virq_to_irq(virq, cpu, irqflags & IRQF_PERCPU);
1072 retval = request_irq(irq, handler, irqflags, devname, dev_id);
1074 unbind_from_irq(irq);
1080 EXPORT_SYMBOL_GPL(bind_virq_to_irqhandler);
1082 int bind_ipi_to_irqhandler(enum ipi_vector ipi,
1084 irq_handler_t handler,
1085 unsigned long irqflags,
1086 const char *devname,
1091 irq = bind_ipi_to_irq(ipi, cpu);
1095 irqflags |= IRQF_NO_SUSPEND | IRQF_FORCE_RESUME | IRQF_EARLY_RESUME;
1096 retval = request_irq(irq, handler, irqflags, devname, dev_id);
1098 unbind_from_irq(irq);
1105 void unbind_from_irqhandler(unsigned int irq, void *dev_id)
1107 struct irq_info *info = irq_get_handler_data(irq);
1111 free_irq(irq, dev_id);
1112 unbind_from_irq(irq);
1114 EXPORT_SYMBOL_GPL(unbind_from_irqhandler);
1117 * xen_set_irq_priority() - set an event channel priority.
1118 * @irq:irq bound to an event channel.
1119 * @priority: priority between XEN_IRQ_PRIORITY_MAX and XEN_IRQ_PRIORITY_MIN.
1121 int xen_set_irq_priority(unsigned irq, unsigned priority)
1123 struct evtchn_set_priority set_priority;
1125 set_priority.port = evtchn_from_irq(irq);
1126 set_priority.priority = priority;
1128 return HYPERVISOR_event_channel_op(EVTCHNOP_set_priority,
1131 EXPORT_SYMBOL_GPL(xen_set_irq_priority);
1133 int evtchn_make_refcounted(unsigned int evtchn)
1135 int irq = get_evtchn_to_irq(evtchn);
1136 struct irq_info *info;
1141 info = irq_get_handler_data(irq);
1146 WARN_ON(info->refcnt != -1);
1152 EXPORT_SYMBOL_GPL(evtchn_make_refcounted);
1154 int evtchn_get(unsigned int evtchn)
1157 struct irq_info *info;
1160 if (evtchn >= xen_evtchn_max_channels())
1163 mutex_lock(&irq_mapping_update_lock);
1165 irq = get_evtchn_to_irq(evtchn);
1169 info = irq_get_handler_data(irq);
1175 if (info->refcnt <= 0)
1181 mutex_unlock(&irq_mapping_update_lock);
1185 EXPORT_SYMBOL_GPL(evtchn_get);
1187 void evtchn_put(unsigned int evtchn)
1189 int irq = get_evtchn_to_irq(evtchn);
1190 if (WARN_ON(irq == -1))
1192 unbind_from_irq(irq);
1194 EXPORT_SYMBOL_GPL(evtchn_put);
1196 void xen_send_IPI_one(unsigned int cpu, enum ipi_vector vector)
1201 if (unlikely(vector == XEN_NMI_VECTOR)) {
1202 int rc = HYPERVISOR_vcpu_op(VCPUOP_send_nmi, cpu, NULL);
1204 printk(KERN_WARNING "Sending nmi to CPU%d failed (rc:%d)\n", cpu, rc);
1208 irq = per_cpu(ipi_to_irq, cpu)[vector];
1210 notify_remote_via_irq(irq);
1213 static DEFINE_PER_CPU(unsigned, xed_nesting_count);
1215 static void __xen_evtchn_do_upcall(void)
1217 struct vcpu_info *vcpu_info = __this_cpu_read(xen_vcpu);
1218 int cpu = get_cpu();
1222 vcpu_info->evtchn_upcall_pending = 0;
1224 if (__this_cpu_inc_return(xed_nesting_count) - 1)
1227 xen_evtchn_handle_events(cpu);
1229 BUG_ON(!irqs_disabled());
1231 count = __this_cpu_read(xed_nesting_count);
1232 __this_cpu_write(xed_nesting_count, 0);
1233 } while (count != 1 || vcpu_info->evtchn_upcall_pending);
1240 void xen_evtchn_do_upcall(struct pt_regs *regs)
1242 struct pt_regs *old_regs = set_irq_regs(regs);
1247 inc_irq_stat(irq_hv_callback_count);
1250 __xen_evtchn_do_upcall();
1253 set_irq_regs(old_regs);
1256 void xen_hvm_evtchn_do_upcall(void)
1258 __xen_evtchn_do_upcall();
1260 EXPORT_SYMBOL_GPL(xen_hvm_evtchn_do_upcall);
1262 /* Rebind a new event channel to an existing irq. */
1263 void rebind_evtchn_irq(int evtchn, int irq)
1265 struct irq_info *info = info_for_irq(irq);
1270 /* Make sure the irq is masked, since the new event channel
1271 will also be masked. */
1274 mutex_lock(&irq_mapping_update_lock);
1276 /* After resume the irq<->evtchn mappings are all cleared out */
1277 BUG_ON(get_evtchn_to_irq(evtchn) != -1);
1278 /* Expect irq to have been bound before,
1279 so there should be a proper type */
1280 BUG_ON(info->type == IRQT_UNBOUND);
1282 (void)xen_irq_info_evtchn_setup(irq, evtchn);
1284 mutex_unlock(&irq_mapping_update_lock);
1286 bind_evtchn_to_cpu(evtchn, info->cpu);
1287 /* This will be deferred until interrupt is processed */
1288 irq_set_affinity(irq, cpumask_of(info->cpu));
1290 /* Unmask the event channel. */
1294 /* Rebind an evtchn so that it gets delivered to a specific cpu */
1295 static int rebind_irq_to_cpu(unsigned irq, unsigned tcpu)
1297 struct evtchn_bind_vcpu bind_vcpu;
1298 int evtchn = evtchn_from_irq(irq);
1301 if (!VALID_EVTCHN(evtchn))
1305 * Events delivered via platform PCI interrupts are always
1306 * routed to vcpu 0 and hence cannot be rebound.
1308 if (xen_hvm_domain() && !xen_have_vector_callback)
1311 /* Send future instances of this interrupt to other vcpu. */
1312 bind_vcpu.port = evtchn;
1313 bind_vcpu.vcpu = tcpu;
1316 * Mask the event while changing the VCPU binding to prevent
1317 * it being delivered on an unexpected VCPU.
1319 masked = test_and_set_mask(evtchn);
1322 * If this fails, it usually just indicates that we're dealing with a
1323 * virq or IPI channel, which don't actually need to be rebound. Ignore
1324 * it, but don't do the xenlinux-level rebind in that case.
1326 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_vcpu, &bind_vcpu) >= 0)
1327 bind_evtchn_to_cpu(evtchn, tcpu);
1330 unmask_evtchn(evtchn);
1335 static int set_affinity_irq(struct irq_data *data, const struct cpumask *dest,
1338 unsigned tcpu = cpumask_first_and(dest, cpu_online_mask);
1340 return rebind_irq_to_cpu(data->irq, tcpu);
1343 static void enable_dynirq(struct irq_data *data)
1345 int evtchn = evtchn_from_irq(data->irq);
1347 if (VALID_EVTCHN(evtchn))
1348 unmask_evtchn(evtchn);
1351 static void disable_dynirq(struct irq_data *data)
1353 int evtchn = evtchn_from_irq(data->irq);
1355 if (VALID_EVTCHN(evtchn))
1356 mask_evtchn(evtchn);
1359 static void ack_dynirq(struct irq_data *data)
1361 int evtchn = evtchn_from_irq(data->irq);
1365 if (VALID_EVTCHN(evtchn))
1366 clear_evtchn(evtchn);
1369 static void mask_ack_dynirq(struct irq_data *data)
1371 disable_dynirq(data);
1375 static int retrigger_dynirq(struct irq_data *data)
1377 unsigned int evtchn = evtchn_from_irq(data->irq);
1380 if (!VALID_EVTCHN(evtchn))
1383 masked = test_and_set_mask(evtchn);
1386 unmask_evtchn(evtchn);
1391 static void restore_pirqs(void)
1393 int pirq, rc, irq, gsi;
1394 struct physdev_map_pirq map_irq;
1395 struct irq_info *info;
1397 list_for_each_entry(info, &xen_irq_list_head, list) {
1398 if (info->type != IRQT_PIRQ)
1401 pirq = info->u.pirq.pirq;
1402 gsi = info->u.pirq.gsi;
1405 /* save/restore of PT devices doesn't work, so at this point the
1406 * only devices present are GSI based emulated devices */
1410 map_irq.domid = DOMID_SELF;
1411 map_irq.type = MAP_PIRQ_TYPE_GSI;
1412 map_irq.index = gsi;
1413 map_irq.pirq = pirq;
1415 rc = HYPERVISOR_physdev_op(PHYSDEVOP_map_pirq, &map_irq);
1417 pr_warn("xen map irq failed gsi=%d irq=%d pirq=%d rc=%d\n",
1418 gsi, irq, pirq, rc);
1423 printk(KERN_DEBUG "xen: --> irq=%d, pirq=%d\n", irq, map_irq.pirq);
1425 __startup_pirq(irq);
1429 static void restore_cpu_virqs(unsigned int cpu)
1431 struct evtchn_bind_virq bind_virq;
1432 int virq, irq, evtchn;
1434 for (virq = 0; virq < NR_VIRQS; virq++) {
1435 if ((irq = per_cpu(virq_to_irq, cpu)[virq]) == -1)
1438 BUG_ON(virq_from_irq(irq) != virq);
1440 /* Get a new binding from Xen. */
1441 bind_virq.virq = virq;
1442 bind_virq.vcpu = cpu;
1443 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_virq,
1446 evtchn = bind_virq.port;
1448 /* Record the new mapping. */
1449 (void)xen_irq_info_virq_setup(cpu, irq, evtchn, virq);
1450 bind_evtchn_to_cpu(evtchn, cpu);
1454 static void restore_cpu_ipis(unsigned int cpu)
1456 struct evtchn_bind_ipi bind_ipi;
1457 int ipi, irq, evtchn;
1459 for (ipi = 0; ipi < XEN_NR_IPIS; ipi++) {
1460 if ((irq = per_cpu(ipi_to_irq, cpu)[ipi]) == -1)
1463 BUG_ON(ipi_from_irq(irq) != ipi);
1465 /* Get a new binding from Xen. */
1466 bind_ipi.vcpu = cpu;
1467 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_ipi,
1470 evtchn = bind_ipi.port;
1472 /* Record the new mapping. */
1473 (void)xen_irq_info_ipi_setup(cpu, irq, evtchn, ipi);
1474 bind_evtchn_to_cpu(evtchn, cpu);
1478 /* Clear an irq's pending state, in preparation for polling on it */
1479 void xen_clear_irq_pending(int irq)
1481 int evtchn = evtchn_from_irq(irq);
1483 if (VALID_EVTCHN(evtchn))
1484 clear_evtchn(evtchn);
1486 EXPORT_SYMBOL(xen_clear_irq_pending);
1487 void xen_set_irq_pending(int irq)
1489 int evtchn = evtchn_from_irq(irq);
1491 if (VALID_EVTCHN(evtchn))
1495 bool xen_test_irq_pending(int irq)
1497 int evtchn = evtchn_from_irq(irq);
1500 if (VALID_EVTCHN(evtchn))
1501 ret = test_evtchn(evtchn);
1506 /* Poll waiting for an irq to become pending with timeout. In the usual case,
1507 * the irq will be disabled so it won't deliver an interrupt. */
1508 void xen_poll_irq_timeout(int irq, u64 timeout)
1510 evtchn_port_t evtchn = evtchn_from_irq(irq);
1512 if (VALID_EVTCHN(evtchn)) {
1513 struct sched_poll poll;
1516 poll.timeout = timeout;
1517 set_xen_guest_handle(poll.ports, &evtchn);
1519 if (HYPERVISOR_sched_op(SCHEDOP_poll, &poll) != 0)
1523 EXPORT_SYMBOL(xen_poll_irq_timeout);
1524 /* Poll waiting for an irq to become pending. In the usual case, the
1525 * irq will be disabled so it won't deliver an interrupt. */
1526 void xen_poll_irq(int irq)
1528 xen_poll_irq_timeout(irq, 0 /* no timeout */);
1531 /* Check whether the IRQ line is shared with other guests. */
1532 int xen_test_irq_shared(int irq)
1534 struct irq_info *info = info_for_irq(irq);
1535 struct physdev_irq_status_query irq_status;
1540 irq_status.irq = info->u.pirq.pirq;
1542 if (HYPERVISOR_physdev_op(PHYSDEVOP_irq_status_query, &irq_status))
1544 return !(irq_status.flags & XENIRQSTAT_shared);
1546 EXPORT_SYMBOL_GPL(xen_test_irq_shared);
1548 void xen_irq_resume(void)
1551 struct irq_info *info;
1553 /* New event-channel space is not 'live' yet. */
1554 xen_evtchn_mask_all();
1555 xen_evtchn_resume();
1557 /* No IRQ <-> event-channel mappings. */
1558 list_for_each_entry(info, &xen_irq_list_head, list)
1559 info->evtchn = 0; /* zap event-channel binding */
1561 clear_evtchn_to_irq_all();
1563 for_each_possible_cpu(cpu) {
1564 restore_cpu_virqs(cpu);
1565 restore_cpu_ipis(cpu);
1571 static struct irq_chip xen_dynamic_chip __read_mostly = {
1574 .irq_disable = disable_dynirq,
1575 .irq_mask = disable_dynirq,
1576 .irq_unmask = enable_dynirq,
1578 .irq_ack = ack_dynirq,
1579 .irq_mask_ack = mask_ack_dynirq,
1581 .irq_set_affinity = set_affinity_irq,
1582 .irq_retrigger = retrigger_dynirq,
1585 static struct irq_chip xen_pirq_chip __read_mostly = {
1588 .irq_startup = startup_pirq,
1589 .irq_shutdown = shutdown_pirq,
1590 .irq_enable = enable_pirq,
1591 .irq_disable = disable_pirq,
1593 .irq_mask = disable_dynirq,
1594 .irq_unmask = enable_dynirq,
1596 .irq_ack = eoi_pirq,
1597 .irq_eoi = eoi_pirq,
1598 .irq_mask_ack = mask_ack_pirq,
1600 .irq_set_affinity = set_affinity_irq,
1602 .irq_retrigger = retrigger_dynirq,
1605 static struct irq_chip xen_percpu_chip __read_mostly = {
1606 .name = "xen-percpu",
1608 .irq_disable = disable_dynirq,
1609 .irq_mask = disable_dynirq,
1610 .irq_unmask = enable_dynirq,
1612 .irq_ack = ack_dynirq,
1615 int xen_set_callback_via(uint64_t via)
1617 struct xen_hvm_param a;
1618 a.domid = DOMID_SELF;
1619 a.index = HVM_PARAM_CALLBACK_IRQ;
1621 return HYPERVISOR_hvm_op(HVMOP_set_param, &a);
1623 EXPORT_SYMBOL_GPL(xen_set_callback_via);
1625 #ifdef CONFIG_XEN_PVHVM
1626 /* Vector callbacks are better than PCI interrupts to receive event
1627 * channel notifications because we can receive vector callbacks on any
1628 * vcpu and we don't need PCI support or APIC interactions. */
1629 void xen_callback_vector(void)
1632 uint64_t callback_via;
1633 if (xen_have_vector_callback) {
1634 callback_via = HVM_CALLBACK_VECTOR(HYPERVISOR_CALLBACK_VECTOR);
1635 rc = xen_set_callback_via(callback_via);
1637 pr_err("Request for Xen HVM callback vector failed\n");
1638 xen_have_vector_callback = 0;
1641 pr_info("Xen HVM callback vector for event delivery is enabled\n");
1642 /* in the restore case the vector has already been allocated */
1643 if (!test_bit(HYPERVISOR_CALLBACK_VECTOR, used_vectors))
1644 alloc_intr_gate(HYPERVISOR_CALLBACK_VECTOR,
1645 xen_hvm_callback_vector);
1649 void xen_callback_vector(void) {}
1652 #undef MODULE_PARAM_PREFIX
1653 #define MODULE_PARAM_PREFIX "xen."
1655 static bool fifo_events = true;
1656 module_param(fifo_events, bool, 0);
1658 void __init xen_init_IRQ(void)
1663 ret = xen_evtchn_fifo_init();
1665 xen_evtchn_2l_init();
1667 evtchn_to_irq = kcalloc(EVTCHN_ROW(xen_evtchn_max_channels()),
1668 sizeof(*evtchn_to_irq), GFP_KERNEL);
1669 BUG_ON(!evtchn_to_irq);
1671 /* No event channels are 'live' right now. */
1672 xen_evtchn_mask_all();
1674 pirq_needs_eoi = pirq_needs_eoi_flag;
1677 if (xen_pv_domain()) {
1678 irq_ctx_init(smp_processor_id());
1679 if (xen_initial_domain())
1680 pci_xen_initial_domain();
1682 if (xen_feature(XENFEAT_hvm_callback_vector))
1683 xen_callback_vector();
1685 if (xen_hvm_domain()) {
1687 /* pci_xen_hvm_init must be called after native_init_IRQ so that
1688 * __acpi_register_gsi can point at the right function */
1692 struct physdev_pirq_eoi_gmfn eoi_gmfn;
1694 pirq_eoi_map = (void *)__get_free_page(GFP_KERNEL|__GFP_ZERO);
1695 eoi_gmfn.gmfn = virt_to_mfn(pirq_eoi_map);
1696 rc = HYPERVISOR_physdev_op(PHYSDEVOP_pirq_eoi_gmfn_v2, &eoi_gmfn);
1697 /* TODO: No PVH support for PIRQ EOI */
1699 free_page((unsigned long) pirq_eoi_map);
1700 pirq_eoi_map = NULL;
1702 pirq_needs_eoi = pirq_check_eoi_map;