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/moduleparam.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>
40 #include <asm/io_apic.h>
41 #include <asm/i8259.h>
42 #include <asm/xen/pci.h>
44 #include <asm/sync_bitops.h>
45 #include <asm/xen/hypercall.h>
46 #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_affinity_mask(irq), cpumask_of(cpu));
341 xen_evtchn_port_bind_to_cpu(info, cpu);
347 * notify_remote_via_irq - send event to remote end of event channel via irq
348 * @irq: irq of event channel to send event to
350 * Unlike notify_remote_via_evtchn(), this is safe to use across
351 * save/restore. Notifications on a broken connection are silently
354 void notify_remote_via_irq(int irq)
356 int evtchn = evtchn_from_irq(irq);
358 if (VALID_EVTCHN(evtchn))
359 notify_remote_via_evtchn(evtchn);
361 EXPORT_SYMBOL_GPL(notify_remote_via_irq);
363 static void xen_irq_init(unsigned irq)
365 struct irq_info *info;
367 /* By default all event channels notify CPU#0. */
368 cpumask_copy(irq_get_affinity_mask(irq), cpumask_of(0));
371 info = kzalloc(sizeof(*info), GFP_KERNEL);
373 panic("Unable to allocate metadata for IRQ%d\n", irq);
375 info->type = IRQT_UNBOUND;
378 irq_set_handler_data(irq, info);
380 list_add_tail(&info->list, &xen_irq_list_head);
383 static int __must_check xen_allocate_irqs_dynamic(int nvec)
385 int i, irq = irq_alloc_descs(-1, 0, nvec, -1);
388 for (i = 0; i < nvec; i++)
389 xen_irq_init(irq + i);
395 static inline int __must_check xen_allocate_irq_dynamic(void)
398 return xen_allocate_irqs_dynamic(1);
401 static int __must_check xen_allocate_irq_gsi(unsigned gsi)
406 * A PV guest has no concept of a GSI (since it has no ACPI
407 * nor access to/knowledge of the physical APICs). Therefore
408 * all IRQs are dynamically allocated from the entire IRQ
411 if (xen_pv_domain() && !xen_initial_domain())
412 return xen_allocate_irq_dynamic();
414 /* Legacy IRQ descriptors are already allocated by the arch. */
415 if (gsi < nr_legacy_irqs())
418 irq = irq_alloc_desc_at(gsi, -1);
425 static void xen_free_irq(unsigned irq)
427 struct irq_info *info = irq_get_handler_data(irq);
432 list_del(&info->list);
434 irq_set_handler_data(irq, NULL);
436 WARN_ON(info->refcnt > 0);
440 /* Legacy IRQ descriptors are managed by the arch. */
441 if (irq < nr_legacy_irqs())
447 static void xen_evtchn_close(unsigned int port)
449 struct evtchn_close close;
452 if (HYPERVISOR_event_channel_op(EVTCHNOP_close, &close) != 0)
456 static void pirq_query_unmask(int irq)
458 struct physdev_irq_status_query irq_status;
459 struct irq_info *info = info_for_irq(irq);
461 BUG_ON(info->type != IRQT_PIRQ);
463 irq_status.irq = pirq_from_irq(irq);
464 if (HYPERVISOR_physdev_op(PHYSDEVOP_irq_status_query, &irq_status))
465 irq_status.flags = 0;
467 info->u.pirq.flags &= ~PIRQ_NEEDS_EOI;
468 if (irq_status.flags & XENIRQSTAT_needs_eoi)
469 info->u.pirq.flags |= PIRQ_NEEDS_EOI;
472 static void eoi_pirq(struct irq_data *data)
474 int evtchn = evtchn_from_irq(data->irq);
475 struct physdev_eoi eoi = { .irq = pirq_from_irq(data->irq) };
478 if (!VALID_EVTCHN(evtchn))
481 if (unlikely(irqd_is_setaffinity_pending(data)) &&
482 likely(!irqd_irq_disabled(data))) {
483 int masked = test_and_set_mask(evtchn);
485 clear_evtchn(evtchn);
487 irq_move_masked_irq(data);
490 unmask_evtchn(evtchn);
492 clear_evtchn(evtchn);
494 if (pirq_needs_eoi(data->irq)) {
495 rc = HYPERVISOR_physdev_op(PHYSDEVOP_eoi, &eoi);
500 static void mask_ack_pirq(struct irq_data *data)
502 disable_dynirq(data);
506 static unsigned int __startup_pirq(unsigned int irq)
508 struct evtchn_bind_pirq bind_pirq;
509 struct irq_info *info = info_for_irq(irq);
510 int evtchn = evtchn_from_irq(irq);
513 BUG_ON(info->type != IRQT_PIRQ);
515 if (VALID_EVTCHN(evtchn))
518 bind_pirq.pirq = pirq_from_irq(irq);
519 /* NB. We are happy to share unless we are probing. */
520 bind_pirq.flags = info->u.pirq.flags & PIRQ_SHAREABLE ?
521 BIND_PIRQ__WILL_SHARE : 0;
522 rc = HYPERVISOR_event_channel_op(EVTCHNOP_bind_pirq, &bind_pirq);
524 pr_warn("Failed to obtain physical IRQ %d\n", irq);
527 evtchn = bind_pirq.port;
529 pirq_query_unmask(irq);
531 rc = set_evtchn_to_irq(evtchn, irq);
535 info->evtchn = evtchn;
536 bind_evtchn_to_cpu(evtchn, 0);
538 rc = xen_evtchn_port_setup(info);
543 unmask_evtchn(evtchn);
544 eoi_pirq(irq_get_irq_data(irq));
549 pr_err("irq%d: Failed to set port to irq mapping (%d)\n", irq, rc);
550 xen_evtchn_close(evtchn);
554 static unsigned int startup_pirq(struct irq_data *data)
556 return __startup_pirq(data->irq);
559 static void shutdown_pirq(struct irq_data *data)
561 unsigned int irq = data->irq;
562 struct irq_info *info = info_for_irq(irq);
563 unsigned evtchn = evtchn_from_irq(irq);
565 BUG_ON(info->type != IRQT_PIRQ);
567 if (!VALID_EVTCHN(evtchn))
571 xen_evtchn_close(evtchn);
572 xen_irq_info_cleanup(info);
575 static void enable_pirq(struct irq_data *data)
580 static void disable_pirq(struct irq_data *data)
582 disable_dynirq(data);
585 int xen_irq_from_gsi(unsigned gsi)
587 struct irq_info *info;
589 list_for_each_entry(info, &xen_irq_list_head, list) {
590 if (info->type != IRQT_PIRQ)
593 if (info->u.pirq.gsi == gsi)
599 EXPORT_SYMBOL_GPL(xen_irq_from_gsi);
601 static void __unbind_from_irq(unsigned int irq)
603 int evtchn = evtchn_from_irq(irq);
604 struct irq_info *info = irq_get_handler_data(irq);
606 if (info->refcnt > 0) {
608 if (info->refcnt != 0)
612 if (VALID_EVTCHN(evtchn)) {
613 unsigned int cpu = cpu_from_irq(irq);
615 xen_evtchn_close(evtchn);
617 switch (type_from_irq(irq)) {
619 per_cpu(virq_to_irq, cpu)[virq_from_irq(irq)] = -1;
622 per_cpu(ipi_to_irq, cpu)[ipi_from_irq(irq)] = -1;
628 xen_irq_info_cleanup(info);
631 BUG_ON(info_for_irq(irq)->type == IRQT_UNBOUND);
637 * Do not make any assumptions regarding the relationship between the
638 * IRQ number returned here and the Xen pirq argument.
640 * Note: We don't assign an event channel until the irq actually started
641 * up. Return an existing irq if we've already got one for the gsi.
643 * Shareable implies level triggered, not shareable implies edge
646 int xen_bind_pirq_gsi_to_irq(unsigned gsi,
647 unsigned pirq, int shareable, char *name)
650 struct physdev_irq irq_op;
653 mutex_lock(&irq_mapping_update_lock);
655 irq = xen_irq_from_gsi(gsi);
657 pr_info("%s: returning irq %d for gsi %u\n",
662 irq = xen_allocate_irq_gsi(gsi);
669 /* Only the privileged domain can do this. For non-priv, the pcifront
670 * driver provides a PCI bus that does the call to do exactly
671 * this in the priv domain. */
672 if (xen_initial_domain() &&
673 HYPERVISOR_physdev_op(PHYSDEVOP_alloc_irq_vector, &irq_op)) {
679 ret = xen_irq_info_pirq_setup(irq, 0, pirq, gsi, DOMID_SELF,
680 shareable ? PIRQ_SHAREABLE : 0);
682 __unbind_from_irq(irq);
687 pirq_query_unmask(irq);
688 /* We try to use the handler with the appropriate semantic for the
689 * type of interrupt: if the interrupt is an edge triggered
690 * interrupt we use handle_edge_irq.
692 * On the other hand if the interrupt is level triggered we use
693 * handle_fasteoi_irq like the native code does for this kind of
696 * Depending on the Xen version, pirq_needs_eoi might return true
697 * not only for level triggered interrupts but for edge triggered
698 * interrupts too. In any case Xen always honors the eoi mechanism,
699 * not injecting any more pirqs of the same kind if the first one
700 * hasn't received an eoi yet. Therefore using the fasteoi handler
701 * is the right choice either way.
704 irq_set_chip_and_handler_name(irq, &xen_pirq_chip,
705 handle_fasteoi_irq, name);
707 irq_set_chip_and_handler_name(irq, &xen_pirq_chip,
708 handle_edge_irq, name);
711 mutex_unlock(&irq_mapping_update_lock);
716 #ifdef CONFIG_PCI_MSI
717 int xen_allocate_pirq_msi(struct pci_dev *dev, struct msi_desc *msidesc)
720 struct physdev_get_free_pirq op_get_free_pirq;
722 op_get_free_pirq.type = MAP_PIRQ_TYPE_MSI;
723 rc = HYPERVISOR_physdev_op(PHYSDEVOP_get_free_pirq, &op_get_free_pirq);
725 WARN_ONCE(rc == -ENOSYS,
726 "hypervisor does not support the PHYSDEVOP_get_free_pirq interface\n");
728 return rc ? -1 : op_get_free_pirq.pirq;
731 int xen_bind_pirq_msi_to_irq(struct pci_dev *dev, struct msi_desc *msidesc,
732 int pirq, int nvec, const char *name, domid_t domid)
736 mutex_lock(&irq_mapping_update_lock);
738 irq = xen_allocate_irqs_dynamic(nvec);
742 for (i = 0; i < nvec; i++) {
743 irq_set_chip_and_handler_name(irq + i, &xen_pirq_chip, handle_edge_irq, name);
745 ret = xen_irq_info_pirq_setup(irq + i, 0, pirq + i, 0, domid,
746 i == 0 ? 0 : PIRQ_MSI_GROUP);
751 ret = irq_set_msi_desc(irq, msidesc);
755 mutex_unlock(&irq_mapping_update_lock);
759 __unbind_from_irq(irq + nvec);
760 mutex_unlock(&irq_mapping_update_lock);
765 int xen_destroy_irq(int irq)
767 struct physdev_unmap_pirq unmap_irq;
768 struct irq_info *info = info_for_irq(irq);
771 mutex_lock(&irq_mapping_update_lock);
774 * If trying to remove a vector in a MSI group different
775 * than the first one skip the PIRQ unmap unless this vector
776 * is the first one in the group.
778 if (xen_initial_domain() && !(info->u.pirq.flags & PIRQ_MSI_GROUP)) {
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);
857 /* New interdomain events are bound to VCPU 0. */
858 bind_evtchn_to_cpu(evtchn, 0);
860 struct irq_info *info = info_for_irq(irq);
861 WARN_ON(info == NULL || info->type != IRQT_EVTCHN);
865 mutex_unlock(&irq_mapping_update_lock);
869 EXPORT_SYMBOL_GPL(bind_evtchn_to_irq);
871 static int bind_ipi_to_irq(unsigned int ipi, unsigned int cpu)
873 struct evtchn_bind_ipi bind_ipi;
877 mutex_lock(&irq_mapping_update_lock);
879 irq = per_cpu(ipi_to_irq, cpu)[ipi];
882 irq = xen_allocate_irq_dynamic();
886 irq_set_chip_and_handler_name(irq, &xen_percpu_chip,
887 handle_percpu_irq, "ipi");
889 bind_ipi.vcpu = xen_vcpu_nr(cpu);
890 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_ipi,
893 evtchn = bind_ipi.port;
895 ret = xen_irq_info_ipi_setup(cpu, irq, evtchn, ipi);
897 __unbind_from_irq(irq);
901 bind_evtchn_to_cpu(evtchn, cpu);
903 struct irq_info *info = info_for_irq(irq);
904 WARN_ON(info == NULL || info->type != IRQT_IPI);
908 mutex_unlock(&irq_mapping_update_lock);
912 int bind_interdomain_evtchn_to_irq(unsigned int remote_domain,
913 unsigned int remote_port)
915 struct evtchn_bind_interdomain bind_interdomain;
918 bind_interdomain.remote_dom = remote_domain;
919 bind_interdomain.remote_port = remote_port;
921 err = HYPERVISOR_event_channel_op(EVTCHNOP_bind_interdomain,
924 return err ? : bind_evtchn_to_irq(bind_interdomain.local_port);
926 EXPORT_SYMBOL_GPL(bind_interdomain_evtchn_to_irq);
928 static int find_virq(unsigned int virq, unsigned int cpu)
930 struct evtchn_status status;
931 int port, rc = -ENOENT;
933 memset(&status, 0, sizeof(status));
934 for (port = 0; port < xen_evtchn_max_channels(); port++) {
935 status.dom = DOMID_SELF;
937 rc = HYPERVISOR_event_channel_op(EVTCHNOP_status, &status);
940 if (status.status != EVTCHNSTAT_virq)
942 if (status.u.virq == virq && status.vcpu == xen_vcpu_nr(cpu)) {
951 * xen_evtchn_nr_channels - number of usable event channel ports
953 * This may be less than the maximum supported by the current
954 * hypervisor ABI. Use xen_evtchn_max_channels() for the maximum
957 unsigned xen_evtchn_nr_channels(void)
959 return evtchn_ops->nr_channels();
961 EXPORT_SYMBOL_GPL(xen_evtchn_nr_channels);
963 int bind_virq_to_irq(unsigned int virq, unsigned int cpu, bool percpu)
965 struct evtchn_bind_virq bind_virq;
966 int evtchn, irq, ret;
968 mutex_lock(&irq_mapping_update_lock);
970 irq = per_cpu(virq_to_irq, cpu)[virq];
973 irq = xen_allocate_irq_dynamic();
978 irq_set_chip_and_handler_name(irq, &xen_percpu_chip,
979 handle_percpu_irq, "virq");
981 irq_set_chip_and_handler_name(irq, &xen_dynamic_chip,
982 handle_edge_irq, "virq");
984 bind_virq.virq = virq;
985 bind_virq.vcpu = xen_vcpu_nr(cpu);
986 ret = HYPERVISOR_event_channel_op(EVTCHNOP_bind_virq,
989 evtchn = bind_virq.port;
992 ret = find_virq(virq, cpu);
997 ret = xen_irq_info_virq_setup(cpu, irq, evtchn, virq);
999 __unbind_from_irq(irq);
1004 bind_evtchn_to_cpu(evtchn, cpu);
1006 struct irq_info *info = info_for_irq(irq);
1007 WARN_ON(info == NULL || info->type != IRQT_VIRQ);
1011 mutex_unlock(&irq_mapping_update_lock);
1016 static void unbind_from_irq(unsigned int irq)
1018 mutex_lock(&irq_mapping_update_lock);
1019 __unbind_from_irq(irq);
1020 mutex_unlock(&irq_mapping_update_lock);
1023 int bind_evtchn_to_irqhandler(unsigned int evtchn,
1024 irq_handler_t handler,
1025 unsigned long irqflags,
1026 const char *devname, void *dev_id)
1030 irq = bind_evtchn_to_irq(evtchn);
1033 retval = request_irq(irq, handler, irqflags, devname, dev_id);
1035 unbind_from_irq(irq);
1041 EXPORT_SYMBOL_GPL(bind_evtchn_to_irqhandler);
1043 int bind_interdomain_evtchn_to_irqhandler(unsigned int remote_domain,
1044 unsigned int remote_port,
1045 irq_handler_t handler,
1046 unsigned long irqflags,
1047 const char *devname,
1052 irq = bind_interdomain_evtchn_to_irq(remote_domain, remote_port);
1056 retval = request_irq(irq, handler, irqflags, devname, dev_id);
1058 unbind_from_irq(irq);
1064 EXPORT_SYMBOL_GPL(bind_interdomain_evtchn_to_irqhandler);
1066 int bind_virq_to_irqhandler(unsigned int virq, unsigned int cpu,
1067 irq_handler_t handler,
1068 unsigned long irqflags, const char *devname, void *dev_id)
1072 irq = bind_virq_to_irq(virq, cpu, irqflags & IRQF_PERCPU);
1075 retval = request_irq(irq, handler, irqflags, devname, dev_id);
1077 unbind_from_irq(irq);
1083 EXPORT_SYMBOL_GPL(bind_virq_to_irqhandler);
1085 int bind_ipi_to_irqhandler(enum ipi_vector ipi,
1087 irq_handler_t handler,
1088 unsigned long irqflags,
1089 const char *devname,
1094 irq = bind_ipi_to_irq(ipi, cpu);
1098 irqflags |= IRQF_NO_SUSPEND | IRQF_FORCE_RESUME | IRQF_EARLY_RESUME;
1099 retval = request_irq(irq, handler, irqflags, devname, dev_id);
1101 unbind_from_irq(irq);
1108 void unbind_from_irqhandler(unsigned int irq, void *dev_id)
1110 struct irq_info *info = irq_get_handler_data(irq);
1114 free_irq(irq, dev_id);
1115 unbind_from_irq(irq);
1117 EXPORT_SYMBOL_GPL(unbind_from_irqhandler);
1120 * xen_set_irq_priority() - set an event channel priority.
1121 * @irq:irq bound to an event channel.
1122 * @priority: priority between XEN_IRQ_PRIORITY_MAX and XEN_IRQ_PRIORITY_MIN.
1124 int xen_set_irq_priority(unsigned irq, unsigned priority)
1126 struct evtchn_set_priority set_priority;
1128 set_priority.port = evtchn_from_irq(irq);
1129 set_priority.priority = priority;
1131 return HYPERVISOR_event_channel_op(EVTCHNOP_set_priority,
1134 EXPORT_SYMBOL_GPL(xen_set_irq_priority);
1136 int evtchn_make_refcounted(unsigned int evtchn)
1138 int irq = get_evtchn_to_irq(evtchn);
1139 struct irq_info *info;
1144 info = irq_get_handler_data(irq);
1149 WARN_ON(info->refcnt != -1);
1155 EXPORT_SYMBOL_GPL(evtchn_make_refcounted);
1157 int evtchn_get(unsigned int evtchn)
1160 struct irq_info *info;
1163 if (evtchn >= xen_evtchn_max_channels())
1166 mutex_lock(&irq_mapping_update_lock);
1168 irq = get_evtchn_to_irq(evtchn);
1172 info = irq_get_handler_data(irq);
1178 if (info->refcnt <= 0)
1184 mutex_unlock(&irq_mapping_update_lock);
1188 EXPORT_SYMBOL_GPL(evtchn_get);
1190 void evtchn_put(unsigned int evtchn)
1192 int irq = get_evtchn_to_irq(evtchn);
1193 if (WARN_ON(irq == -1))
1195 unbind_from_irq(irq);
1197 EXPORT_SYMBOL_GPL(evtchn_put);
1199 void xen_send_IPI_one(unsigned int cpu, enum ipi_vector vector)
1204 if (unlikely(vector == XEN_NMI_VECTOR)) {
1205 int rc = HYPERVISOR_vcpu_op(VCPUOP_send_nmi, xen_vcpu_nr(cpu),
1208 printk(KERN_WARNING "Sending nmi to CPU%d failed (rc:%d)\n", cpu, rc);
1212 irq = per_cpu(ipi_to_irq, cpu)[vector];
1214 notify_remote_via_irq(irq);
1217 static DEFINE_PER_CPU(unsigned, xed_nesting_count);
1219 static void __xen_evtchn_do_upcall(void)
1221 struct vcpu_info *vcpu_info = __this_cpu_read(xen_vcpu);
1222 int cpu = get_cpu();
1226 vcpu_info->evtchn_upcall_pending = 0;
1228 if (__this_cpu_inc_return(xed_nesting_count) - 1)
1231 xen_evtchn_handle_events(cpu);
1233 BUG_ON(!irqs_disabled());
1235 count = __this_cpu_read(xed_nesting_count);
1236 __this_cpu_write(xed_nesting_count, 0);
1237 } while (count != 1 || vcpu_info->evtchn_upcall_pending);
1244 void xen_evtchn_do_upcall(struct pt_regs *regs)
1246 struct pt_regs *old_regs = set_irq_regs(regs);
1250 inc_irq_stat(irq_hv_callback_count);
1253 __xen_evtchn_do_upcall();
1256 set_irq_regs(old_regs);
1259 void xen_hvm_evtchn_do_upcall(void)
1261 __xen_evtchn_do_upcall();
1263 EXPORT_SYMBOL_GPL(xen_hvm_evtchn_do_upcall);
1265 /* Rebind a new event channel to an existing irq. */
1266 void rebind_evtchn_irq(int evtchn, int irq)
1268 struct irq_info *info = info_for_irq(irq);
1273 /* Make sure the irq is masked, since the new event channel
1274 will also be masked. */
1277 mutex_lock(&irq_mapping_update_lock);
1279 /* After resume the irq<->evtchn mappings are all cleared out */
1280 BUG_ON(get_evtchn_to_irq(evtchn) != -1);
1281 /* Expect irq to have been bound before,
1282 so there should be a proper type */
1283 BUG_ON(info->type == IRQT_UNBOUND);
1285 (void)xen_irq_info_evtchn_setup(irq, evtchn);
1287 mutex_unlock(&irq_mapping_update_lock);
1289 bind_evtchn_to_cpu(evtchn, info->cpu);
1290 /* This will be deferred until interrupt is processed */
1291 irq_set_affinity(irq, cpumask_of(info->cpu));
1293 /* Unmask the event channel. */
1297 /* Rebind an evtchn so that it gets delivered to a specific cpu */
1298 int xen_rebind_evtchn_to_cpu(int evtchn, unsigned tcpu)
1300 struct evtchn_bind_vcpu bind_vcpu;
1303 if (!VALID_EVTCHN(evtchn))
1306 if (!xen_support_evtchn_rebind())
1309 /* Send future instances of this interrupt to other vcpu. */
1310 bind_vcpu.port = evtchn;
1311 bind_vcpu.vcpu = xen_vcpu_nr(tcpu);
1314 * Mask the event while changing the VCPU binding to prevent
1315 * it being delivered on an unexpected VCPU.
1317 masked = test_and_set_mask(evtchn);
1320 * If this fails, it usually just indicates that we're dealing with a
1321 * virq or IPI channel, which don't actually need to be rebound. Ignore
1322 * it, but don't do the xenlinux-level rebind in that case.
1324 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_vcpu, &bind_vcpu) >= 0)
1325 bind_evtchn_to_cpu(evtchn, tcpu);
1328 unmask_evtchn(evtchn);
1332 EXPORT_SYMBOL_GPL(xen_rebind_evtchn_to_cpu);
1334 static int set_affinity_irq(struct irq_data *data, const struct cpumask *dest,
1337 unsigned tcpu = cpumask_first_and(dest, cpu_online_mask);
1338 int ret = xen_rebind_evtchn_to_cpu(evtchn_from_irq(data->irq), tcpu);
1341 irq_data_update_effective_affinity(data, cpumask_of(tcpu));
1346 static void enable_dynirq(struct irq_data *data)
1348 int evtchn = evtchn_from_irq(data->irq);
1350 if (VALID_EVTCHN(evtchn))
1351 unmask_evtchn(evtchn);
1354 static void disable_dynirq(struct irq_data *data)
1356 int evtchn = evtchn_from_irq(data->irq);
1358 if (VALID_EVTCHN(evtchn))
1359 mask_evtchn(evtchn);
1362 static void ack_dynirq(struct irq_data *data)
1364 int evtchn = evtchn_from_irq(data->irq);
1366 if (!VALID_EVTCHN(evtchn))
1369 if (unlikely(irqd_is_setaffinity_pending(data)) &&
1370 likely(!irqd_irq_disabled(data))) {
1371 int masked = test_and_set_mask(evtchn);
1373 clear_evtchn(evtchn);
1375 irq_move_masked_irq(data);
1378 unmask_evtchn(evtchn);
1380 clear_evtchn(evtchn);
1383 static void mask_ack_dynirq(struct irq_data *data)
1385 disable_dynirq(data);
1389 static int retrigger_dynirq(struct irq_data *data)
1391 unsigned int evtchn = evtchn_from_irq(data->irq);
1394 if (!VALID_EVTCHN(evtchn))
1397 masked = test_and_set_mask(evtchn);
1400 unmask_evtchn(evtchn);
1405 static void restore_pirqs(void)
1407 int pirq, rc, irq, gsi;
1408 struct physdev_map_pirq map_irq;
1409 struct irq_info *info;
1411 list_for_each_entry(info, &xen_irq_list_head, list) {
1412 if (info->type != IRQT_PIRQ)
1415 pirq = info->u.pirq.pirq;
1416 gsi = info->u.pirq.gsi;
1419 /* save/restore of PT devices doesn't work, so at this point the
1420 * only devices present are GSI based emulated devices */
1424 map_irq.domid = DOMID_SELF;
1425 map_irq.type = MAP_PIRQ_TYPE_GSI;
1426 map_irq.index = gsi;
1427 map_irq.pirq = pirq;
1429 rc = HYPERVISOR_physdev_op(PHYSDEVOP_map_pirq, &map_irq);
1431 pr_warn("xen map irq failed gsi=%d irq=%d pirq=%d rc=%d\n",
1432 gsi, irq, pirq, rc);
1437 printk(KERN_DEBUG "xen: --> irq=%d, pirq=%d\n", irq, map_irq.pirq);
1439 __startup_pirq(irq);
1443 static void restore_cpu_virqs(unsigned int cpu)
1445 struct evtchn_bind_virq bind_virq;
1446 int virq, irq, evtchn;
1448 for (virq = 0; virq < NR_VIRQS; virq++) {
1449 if ((irq = per_cpu(virq_to_irq, cpu)[virq]) == -1)
1452 BUG_ON(virq_from_irq(irq) != virq);
1454 /* Get a new binding from Xen. */
1455 bind_virq.virq = virq;
1456 bind_virq.vcpu = xen_vcpu_nr(cpu);
1457 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_virq,
1460 evtchn = bind_virq.port;
1462 /* Record the new mapping. */
1463 (void)xen_irq_info_virq_setup(cpu, irq, evtchn, virq);
1464 bind_evtchn_to_cpu(evtchn, cpu);
1468 static void restore_cpu_ipis(unsigned int cpu)
1470 struct evtchn_bind_ipi bind_ipi;
1471 int ipi, irq, evtchn;
1473 for (ipi = 0; ipi < XEN_NR_IPIS; ipi++) {
1474 if ((irq = per_cpu(ipi_to_irq, cpu)[ipi]) == -1)
1477 BUG_ON(ipi_from_irq(irq) != ipi);
1479 /* Get a new binding from Xen. */
1480 bind_ipi.vcpu = xen_vcpu_nr(cpu);
1481 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_ipi,
1484 evtchn = bind_ipi.port;
1486 /* Record the new mapping. */
1487 (void)xen_irq_info_ipi_setup(cpu, irq, evtchn, ipi);
1488 bind_evtchn_to_cpu(evtchn, cpu);
1492 /* Clear an irq's pending state, in preparation for polling on it */
1493 void xen_clear_irq_pending(int irq)
1495 int evtchn = evtchn_from_irq(irq);
1497 if (VALID_EVTCHN(evtchn))
1498 clear_evtchn(evtchn);
1500 EXPORT_SYMBOL(xen_clear_irq_pending);
1501 void xen_set_irq_pending(int irq)
1503 int evtchn = evtchn_from_irq(irq);
1505 if (VALID_EVTCHN(evtchn))
1509 bool xen_test_irq_pending(int irq)
1511 int evtchn = evtchn_from_irq(irq);
1514 if (VALID_EVTCHN(evtchn))
1515 ret = test_evtchn(evtchn);
1520 /* Poll waiting for an irq to become pending with timeout. In the usual case,
1521 * the irq will be disabled so it won't deliver an interrupt. */
1522 void xen_poll_irq_timeout(int irq, u64 timeout)
1524 evtchn_port_t evtchn = evtchn_from_irq(irq);
1526 if (VALID_EVTCHN(evtchn)) {
1527 struct sched_poll poll;
1530 poll.timeout = timeout;
1531 set_xen_guest_handle(poll.ports, &evtchn);
1533 if (HYPERVISOR_sched_op(SCHEDOP_poll, &poll) != 0)
1537 EXPORT_SYMBOL(xen_poll_irq_timeout);
1538 /* Poll waiting for an irq to become pending. In the usual case, the
1539 * irq will be disabled so it won't deliver an interrupt. */
1540 void xen_poll_irq(int irq)
1542 xen_poll_irq_timeout(irq, 0 /* no timeout */);
1545 /* Check whether the IRQ line is shared with other guests. */
1546 int xen_test_irq_shared(int irq)
1548 struct irq_info *info = info_for_irq(irq);
1549 struct physdev_irq_status_query irq_status;
1554 irq_status.irq = info->u.pirq.pirq;
1556 if (HYPERVISOR_physdev_op(PHYSDEVOP_irq_status_query, &irq_status))
1558 return !(irq_status.flags & XENIRQSTAT_shared);
1560 EXPORT_SYMBOL_GPL(xen_test_irq_shared);
1562 void xen_irq_resume(void)
1565 struct irq_info *info;
1567 /* New event-channel space is not 'live' yet. */
1568 xen_evtchn_resume();
1570 /* No IRQ <-> event-channel mappings. */
1571 list_for_each_entry(info, &xen_irq_list_head, list)
1572 info->evtchn = 0; /* zap event-channel binding */
1574 clear_evtchn_to_irq_all();
1576 for_each_possible_cpu(cpu) {
1577 restore_cpu_virqs(cpu);
1578 restore_cpu_ipis(cpu);
1584 static struct irq_chip xen_dynamic_chip __read_mostly = {
1587 .irq_disable = disable_dynirq,
1588 .irq_mask = disable_dynirq,
1589 .irq_unmask = enable_dynirq,
1591 .irq_ack = ack_dynirq,
1592 .irq_mask_ack = mask_ack_dynirq,
1594 .irq_set_affinity = set_affinity_irq,
1595 .irq_retrigger = retrigger_dynirq,
1598 static struct irq_chip xen_pirq_chip __read_mostly = {
1601 .irq_startup = startup_pirq,
1602 .irq_shutdown = shutdown_pirq,
1603 .irq_enable = enable_pirq,
1604 .irq_disable = disable_pirq,
1606 .irq_mask = disable_dynirq,
1607 .irq_unmask = enable_dynirq,
1609 .irq_ack = eoi_pirq,
1610 .irq_eoi = eoi_pirq,
1611 .irq_mask_ack = mask_ack_pirq,
1613 .irq_set_affinity = set_affinity_irq,
1615 .irq_retrigger = retrigger_dynirq,
1618 static struct irq_chip xen_percpu_chip __read_mostly = {
1619 .name = "xen-percpu",
1621 .irq_disable = disable_dynirq,
1622 .irq_mask = disable_dynirq,
1623 .irq_unmask = enable_dynirq,
1625 .irq_ack = ack_dynirq,
1628 int xen_set_callback_via(uint64_t via)
1630 struct xen_hvm_param a;
1631 a.domid = DOMID_SELF;
1632 a.index = HVM_PARAM_CALLBACK_IRQ;
1634 return HYPERVISOR_hvm_op(HVMOP_set_param, &a);
1636 EXPORT_SYMBOL_GPL(xen_set_callback_via);
1638 #ifdef CONFIG_XEN_PVHVM
1639 /* Vector callbacks are better than PCI interrupts to receive event
1640 * channel notifications because we can receive vector callbacks on any
1641 * vcpu and we don't need PCI support or APIC interactions. */
1642 void xen_callback_vector(void)
1645 uint64_t callback_via;
1647 if (xen_have_vector_callback) {
1648 callback_via = HVM_CALLBACK_VECTOR(HYPERVISOR_CALLBACK_VECTOR);
1649 rc = xen_set_callback_via(callback_via);
1651 pr_err("Request for Xen HVM callback vector failed\n");
1652 xen_have_vector_callback = 0;
1655 pr_info("Xen HVM callback vector for event delivery is enabled\n");
1656 alloc_intr_gate(HYPERVISOR_CALLBACK_VECTOR,
1657 xen_hvm_callback_vector);
1661 void xen_callback_vector(void) {}
1664 #undef MODULE_PARAM_PREFIX
1665 #define MODULE_PARAM_PREFIX "xen."
1667 static bool fifo_events = true;
1668 module_param(fifo_events, bool, 0);
1670 void __init xen_init_IRQ(void)
1673 unsigned int evtchn;
1676 ret = xen_evtchn_fifo_init();
1678 xen_evtchn_2l_init();
1680 evtchn_to_irq = kcalloc(EVTCHN_ROW(xen_evtchn_max_channels()),
1681 sizeof(*evtchn_to_irq), GFP_KERNEL);
1682 BUG_ON(!evtchn_to_irq);
1684 /* No event channels are 'live' right now. */
1685 for (evtchn = 0; evtchn < xen_evtchn_nr_channels(); evtchn++)
1686 mask_evtchn(evtchn);
1688 pirq_needs_eoi = pirq_needs_eoi_flag;
1691 if (xen_pv_domain()) {
1692 irq_ctx_init(smp_processor_id());
1693 if (xen_initial_domain())
1694 pci_xen_initial_domain();
1696 if (xen_feature(XENFEAT_hvm_callback_vector))
1697 xen_callback_vector();
1699 if (xen_hvm_domain()) {
1701 /* pci_xen_hvm_init must be called after native_init_IRQ so that
1702 * __acpi_register_gsi can point at the right function */
1706 struct physdev_pirq_eoi_gmfn eoi_gmfn;
1708 pirq_eoi_map = (void *)__get_free_page(GFP_KERNEL|__GFP_ZERO);
1709 eoi_gmfn.gmfn = virt_to_gfn(pirq_eoi_map);
1710 rc = HYPERVISOR_physdev_op(PHYSDEVOP_pirq_eoi_gmfn_v2, &eoi_gmfn);
1712 free_page((unsigned long) pirq_eoi_map);
1713 pirq_eoi_map = NULL;
1715 pirq_needs_eoi = pirq_check_eoi_map;