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_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)
396 #ifdef CONFIG_X86_IO_APIC
398 * For an HVM guest or domain 0 which see "real" (emulated or
399 * actual respectively) GSIs we allocate dynamic IRQs
400 * e.g. those corresponding to event channels or MSIs
401 * etc. from the range above those "real" GSIs to avoid
404 if (xen_initial_domain() || xen_hvm_domain())
405 first = get_nr_irqs_gsi();
408 irq = irq_alloc_descs_from(first, nvec, -1);
411 for (i = 0; i < nvec; i++)
412 xen_irq_init(irq + i);
418 static inline int __must_check xen_allocate_irq_dynamic(void)
421 return xen_allocate_irqs_dynamic(1);
424 static int __must_check xen_allocate_irq_gsi(unsigned gsi)
429 * A PV guest has no concept of a GSI (since it has no ACPI
430 * nor access to/knowledge of the physical APICs). Therefore
431 * all IRQs are dynamically allocated from the entire IRQ
434 if (xen_pv_domain() && !xen_initial_domain())
435 return xen_allocate_irq_dynamic();
437 /* Legacy IRQ descriptors are already allocated by the arch. */
438 if (gsi < NR_IRQS_LEGACY)
441 irq = irq_alloc_desc_at(gsi, -1);
448 static void xen_free_irq(unsigned irq)
450 struct irq_info *info = irq_get_handler_data(irq);
455 list_del(&info->list);
457 irq_set_handler_data(irq, NULL);
459 WARN_ON(info->refcnt > 0);
463 /* Legacy IRQ descriptors are managed by the arch. */
464 if (irq < NR_IRQS_LEGACY)
470 static void xen_evtchn_close(unsigned int port)
472 struct evtchn_close close;
475 if (HYPERVISOR_event_channel_op(EVTCHNOP_close, &close) != 0)
479 static void pirq_query_unmask(int irq)
481 struct physdev_irq_status_query irq_status;
482 struct irq_info *info = info_for_irq(irq);
484 BUG_ON(info->type != IRQT_PIRQ);
486 irq_status.irq = pirq_from_irq(irq);
487 if (HYPERVISOR_physdev_op(PHYSDEVOP_irq_status_query, &irq_status))
488 irq_status.flags = 0;
490 info->u.pirq.flags &= ~PIRQ_NEEDS_EOI;
491 if (irq_status.flags & XENIRQSTAT_needs_eoi)
492 info->u.pirq.flags |= PIRQ_NEEDS_EOI;
495 static void eoi_pirq(struct irq_data *data)
497 int evtchn = evtchn_from_irq(data->irq);
498 struct physdev_eoi eoi = { .irq = pirq_from_irq(data->irq) };
503 if (VALID_EVTCHN(evtchn))
504 clear_evtchn(evtchn);
506 if (pirq_needs_eoi(data->irq)) {
507 rc = HYPERVISOR_physdev_op(PHYSDEVOP_eoi, &eoi);
512 static void mask_ack_pirq(struct irq_data *data)
514 disable_dynirq(data);
518 static unsigned int __startup_pirq(unsigned int irq)
520 struct evtchn_bind_pirq bind_pirq;
521 struct irq_info *info = info_for_irq(irq);
522 int evtchn = evtchn_from_irq(irq);
525 BUG_ON(info->type != IRQT_PIRQ);
527 if (VALID_EVTCHN(evtchn))
530 bind_pirq.pirq = pirq_from_irq(irq);
531 /* NB. We are happy to share unless we are probing. */
532 bind_pirq.flags = info->u.pirq.flags & PIRQ_SHAREABLE ?
533 BIND_PIRQ__WILL_SHARE : 0;
534 rc = HYPERVISOR_event_channel_op(EVTCHNOP_bind_pirq, &bind_pirq);
536 pr_warn("Failed to obtain physical IRQ %d\n", irq);
539 evtchn = bind_pirq.port;
541 pirq_query_unmask(irq);
543 rc = set_evtchn_to_irq(evtchn, irq);
545 pr_err("irq%d: Failed to set port to irq mapping (%d)\n",
547 xen_evtchn_close(evtchn);
550 bind_evtchn_to_cpu(evtchn, 0);
551 info->evtchn = evtchn;
554 unmask_evtchn(evtchn);
555 eoi_pirq(irq_get_irq_data(irq));
560 static unsigned int startup_pirq(struct irq_data *data)
562 return __startup_pirq(data->irq);
565 static void shutdown_pirq(struct irq_data *data)
567 unsigned int irq = data->irq;
568 struct irq_info *info = info_for_irq(irq);
569 unsigned evtchn = evtchn_from_irq(irq);
571 BUG_ON(info->type != IRQT_PIRQ);
573 if (!VALID_EVTCHN(evtchn))
577 xen_evtchn_close(evtchn);
578 xen_irq_info_cleanup(info);
581 static void enable_pirq(struct irq_data *data)
586 static void disable_pirq(struct irq_data *data)
588 disable_dynirq(data);
591 int xen_irq_from_gsi(unsigned gsi)
593 struct irq_info *info;
595 list_for_each_entry(info, &xen_irq_list_head, list) {
596 if (info->type != IRQT_PIRQ)
599 if (info->u.pirq.gsi == gsi)
605 EXPORT_SYMBOL_GPL(xen_irq_from_gsi);
607 static void __unbind_from_irq(unsigned int irq)
609 int evtchn = evtchn_from_irq(irq);
610 struct irq_info *info = irq_get_handler_data(irq);
612 if (info->refcnt > 0) {
614 if (info->refcnt != 0)
618 if (VALID_EVTCHN(evtchn)) {
619 unsigned int cpu = cpu_from_irq(irq);
621 xen_evtchn_close(evtchn);
623 switch (type_from_irq(irq)) {
625 per_cpu(virq_to_irq, cpu)[virq_from_irq(irq)] = -1;
628 per_cpu(ipi_to_irq, cpu)[ipi_from_irq(irq)] = -1;
634 xen_irq_info_cleanup(info);
637 BUG_ON(info_for_irq(irq)->type == IRQT_UNBOUND);
643 * Do not make any assumptions regarding the relationship between the
644 * IRQ number returned here and the Xen pirq argument.
646 * Note: We don't assign an event channel until the irq actually started
647 * up. Return an existing irq if we've already got one for the gsi.
649 * Shareable implies level triggered, not shareable implies edge
652 int xen_bind_pirq_gsi_to_irq(unsigned gsi,
653 unsigned pirq, int shareable, char *name)
656 struct physdev_irq irq_op;
659 mutex_lock(&irq_mapping_update_lock);
661 irq = xen_irq_from_gsi(gsi);
663 pr_info("%s: returning irq %d for gsi %u\n",
668 irq = xen_allocate_irq_gsi(gsi);
675 /* Only the privileged domain can do this. For non-priv, the pcifront
676 * driver provides a PCI bus that does the call to do exactly
677 * this in the priv domain. */
678 if (xen_initial_domain() &&
679 HYPERVISOR_physdev_op(PHYSDEVOP_alloc_irq_vector, &irq_op)) {
685 ret = xen_irq_info_pirq_setup(irq, 0, pirq, gsi, DOMID_SELF,
686 shareable ? PIRQ_SHAREABLE : 0);
688 __unbind_from_irq(irq);
693 pirq_query_unmask(irq);
694 /* We try to use the handler with the appropriate semantic for the
695 * type of interrupt: if the interrupt is an edge triggered
696 * interrupt we use handle_edge_irq.
698 * On the other hand if the interrupt is level triggered we use
699 * handle_fasteoi_irq like the native code does for this kind of
702 * Depending on the Xen version, pirq_needs_eoi might return true
703 * not only for level triggered interrupts but for edge triggered
704 * interrupts too. In any case Xen always honors the eoi mechanism,
705 * not injecting any more pirqs of the same kind if the first one
706 * hasn't received an eoi yet. Therefore using the fasteoi handler
707 * is the right choice either way.
710 irq_set_chip_and_handler_name(irq, &xen_pirq_chip,
711 handle_fasteoi_irq, name);
713 irq_set_chip_and_handler_name(irq, &xen_pirq_chip,
714 handle_edge_irq, name);
717 mutex_unlock(&irq_mapping_update_lock);
722 #ifdef CONFIG_PCI_MSI
723 int xen_allocate_pirq_msi(struct pci_dev *dev, struct msi_desc *msidesc)
726 struct physdev_get_free_pirq op_get_free_pirq;
728 op_get_free_pirq.type = MAP_PIRQ_TYPE_MSI;
729 rc = HYPERVISOR_physdev_op(PHYSDEVOP_get_free_pirq, &op_get_free_pirq);
731 WARN_ONCE(rc == -ENOSYS,
732 "hypervisor does not support the PHYSDEVOP_get_free_pirq interface\n");
734 return rc ? -1 : op_get_free_pirq.pirq;
737 int xen_bind_pirq_msi_to_irq(struct pci_dev *dev, struct msi_desc *msidesc,
738 int pirq, int nvec, const char *name, domid_t domid)
742 mutex_lock(&irq_mapping_update_lock);
744 irq = xen_allocate_irqs_dynamic(nvec);
748 for (i = 0; i < nvec; i++) {
749 irq_set_chip_and_handler_name(irq + i, &xen_pirq_chip, handle_edge_irq, name);
751 ret = xen_irq_info_pirq_setup(irq + i, 0, pirq + i, 0, domid,
752 i == 0 ? 0 : PIRQ_MSI_GROUP);
757 ret = irq_set_msi_desc(irq, msidesc);
761 mutex_unlock(&irq_mapping_update_lock);
765 __unbind_from_irq(irq + i);
766 mutex_unlock(&irq_mapping_update_lock);
771 int xen_destroy_irq(int irq)
773 struct physdev_unmap_pirq unmap_irq;
774 struct irq_info *info = info_for_irq(irq);
777 mutex_lock(&irq_mapping_update_lock);
780 * If trying to remove a vector in a MSI group different
781 * than the first one skip the PIRQ unmap unless this vector
782 * is the first one in the group.
784 if (xen_initial_domain() && !(info->u.pirq.flags & PIRQ_MSI_GROUP)) {
785 unmap_irq.pirq = info->u.pirq.pirq;
786 unmap_irq.domid = info->u.pirq.domid;
787 rc = HYPERVISOR_physdev_op(PHYSDEVOP_unmap_pirq, &unmap_irq);
788 /* If another domain quits without making the pci_disable_msix
789 * call, the Xen hypervisor takes care of freeing the PIRQs
790 * (free_domain_pirqs).
792 if ((rc == -ESRCH && info->u.pirq.domid != DOMID_SELF))
793 pr_info("domain %d does not have %d anymore\n",
794 info->u.pirq.domid, info->u.pirq.pirq);
796 pr_warn("unmap irq failed %d\n", rc);
804 mutex_unlock(&irq_mapping_update_lock);
808 int xen_irq_from_pirq(unsigned pirq)
812 struct irq_info *info;
814 mutex_lock(&irq_mapping_update_lock);
816 list_for_each_entry(info, &xen_irq_list_head, list) {
817 if (info->type != IRQT_PIRQ)
820 if (info->u.pirq.pirq == pirq)
825 mutex_unlock(&irq_mapping_update_lock);
831 int xen_pirq_from_irq(unsigned irq)
833 return pirq_from_irq(irq);
835 EXPORT_SYMBOL_GPL(xen_pirq_from_irq);
837 int bind_evtchn_to_irq(unsigned int evtchn)
842 if (evtchn >= xen_evtchn_max_channels())
845 mutex_lock(&irq_mapping_update_lock);
847 irq = get_evtchn_to_irq(evtchn);
850 irq = xen_allocate_irq_dynamic();
854 irq_set_chip_and_handler_name(irq, &xen_dynamic_chip,
855 handle_edge_irq, "event");
857 ret = xen_irq_info_evtchn_setup(irq, evtchn);
859 __unbind_from_irq(irq);
863 /* New interdomain events are bound to VCPU 0. */
864 bind_evtchn_to_cpu(evtchn, 0);
866 struct irq_info *info = info_for_irq(irq);
867 WARN_ON(info == NULL || info->type != IRQT_EVTCHN);
871 mutex_unlock(&irq_mapping_update_lock);
875 EXPORT_SYMBOL_GPL(bind_evtchn_to_irq);
877 static int bind_ipi_to_irq(unsigned int ipi, unsigned int cpu)
879 struct evtchn_bind_ipi bind_ipi;
883 mutex_lock(&irq_mapping_update_lock);
885 irq = per_cpu(ipi_to_irq, cpu)[ipi];
888 irq = xen_allocate_irq_dynamic();
892 irq_set_chip_and_handler_name(irq, &xen_percpu_chip,
893 handle_percpu_irq, "ipi");
896 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_ipi,
899 evtchn = bind_ipi.port;
901 ret = xen_irq_info_ipi_setup(cpu, irq, evtchn, ipi);
903 __unbind_from_irq(irq);
907 bind_evtchn_to_cpu(evtchn, cpu);
909 struct irq_info *info = info_for_irq(irq);
910 WARN_ON(info == NULL || info->type != IRQT_IPI);
914 mutex_unlock(&irq_mapping_update_lock);
918 static int bind_interdomain_evtchn_to_irq(unsigned int remote_domain,
919 unsigned int remote_port)
921 struct evtchn_bind_interdomain bind_interdomain;
924 bind_interdomain.remote_dom = remote_domain;
925 bind_interdomain.remote_port = remote_port;
927 err = HYPERVISOR_event_channel_op(EVTCHNOP_bind_interdomain,
930 return err ? : bind_evtchn_to_irq(bind_interdomain.local_port);
933 static int find_virq(unsigned int virq, unsigned int cpu)
935 struct evtchn_status status;
936 int port, rc = -ENOENT;
938 memset(&status, 0, sizeof(status));
939 for (port = 0; port < xen_evtchn_max_channels(); port++) {
940 status.dom = DOMID_SELF;
942 rc = HYPERVISOR_event_channel_op(EVTCHNOP_status, &status);
945 if (status.status != EVTCHNSTAT_virq)
947 if (status.u.virq == virq && status.vcpu == cpu) {
956 * xen_evtchn_nr_channels - number of usable event channel ports
958 * This may be less than the maximum supported by the current
959 * hypervisor ABI. Use xen_evtchn_max_channels() for the maximum
962 unsigned xen_evtchn_nr_channels(void)
964 return evtchn_ops->nr_channels();
966 EXPORT_SYMBOL_GPL(xen_evtchn_nr_channels);
968 int bind_virq_to_irq(unsigned int virq, unsigned int cpu)
970 struct evtchn_bind_virq bind_virq;
971 int evtchn, irq, ret;
973 mutex_lock(&irq_mapping_update_lock);
975 irq = per_cpu(virq_to_irq, cpu)[virq];
978 irq = xen_allocate_irq_dynamic();
982 irq_set_chip_and_handler_name(irq, &xen_percpu_chip,
983 handle_percpu_irq, "virq");
985 bind_virq.virq = virq;
986 bind_virq.vcpu = cpu;
987 ret = HYPERVISOR_event_channel_op(EVTCHNOP_bind_virq,
990 evtchn = bind_virq.port;
993 ret = find_virq(virq, cpu);
998 ret = xen_irq_info_virq_setup(cpu, irq, evtchn, virq);
1000 __unbind_from_irq(irq);
1005 bind_evtchn_to_cpu(evtchn, cpu);
1007 struct irq_info *info = info_for_irq(irq);
1008 WARN_ON(info == NULL || info->type != IRQT_VIRQ);
1012 mutex_unlock(&irq_mapping_update_lock);
1017 static void unbind_from_irq(unsigned int irq)
1019 mutex_lock(&irq_mapping_update_lock);
1020 __unbind_from_irq(irq);
1021 mutex_unlock(&irq_mapping_update_lock);
1024 int bind_evtchn_to_irqhandler(unsigned int evtchn,
1025 irq_handler_t handler,
1026 unsigned long irqflags,
1027 const char *devname, void *dev_id)
1031 irq = bind_evtchn_to_irq(evtchn);
1034 retval = request_irq(irq, handler, irqflags, devname, dev_id);
1036 unbind_from_irq(irq);
1042 EXPORT_SYMBOL_GPL(bind_evtchn_to_irqhandler);
1044 int bind_interdomain_evtchn_to_irqhandler(unsigned int remote_domain,
1045 unsigned int remote_port,
1046 irq_handler_t handler,
1047 unsigned long irqflags,
1048 const char *devname,
1053 irq = bind_interdomain_evtchn_to_irq(remote_domain, remote_port);
1057 retval = request_irq(irq, handler, irqflags, devname, dev_id);
1059 unbind_from_irq(irq);
1065 EXPORT_SYMBOL_GPL(bind_interdomain_evtchn_to_irqhandler);
1067 int bind_virq_to_irqhandler(unsigned int virq, unsigned int cpu,
1068 irq_handler_t handler,
1069 unsigned long irqflags, const char *devname, void *dev_id)
1073 irq = bind_virq_to_irq(virq, cpu);
1076 retval = request_irq(irq, handler, irqflags, devname, dev_id);
1078 unbind_from_irq(irq);
1084 EXPORT_SYMBOL_GPL(bind_virq_to_irqhandler);
1086 int bind_ipi_to_irqhandler(enum ipi_vector ipi,
1088 irq_handler_t handler,
1089 unsigned long irqflags,
1090 const char *devname,
1095 irq = bind_ipi_to_irq(ipi, cpu);
1099 irqflags |= IRQF_NO_SUSPEND | IRQF_FORCE_RESUME | IRQF_EARLY_RESUME;
1100 retval = request_irq(irq, handler, irqflags, devname, dev_id);
1102 unbind_from_irq(irq);
1109 void unbind_from_irqhandler(unsigned int irq, void *dev_id)
1111 struct irq_info *info = irq_get_handler_data(irq);
1115 free_irq(irq, dev_id);
1116 unbind_from_irq(irq);
1118 EXPORT_SYMBOL_GPL(unbind_from_irqhandler);
1121 * xen_set_irq_priority() - set an event channel priority.
1122 * @irq:irq bound to an event channel.
1123 * @priority: priority between XEN_IRQ_PRIORITY_MAX and XEN_IRQ_PRIORITY_MIN.
1125 int xen_set_irq_priority(unsigned irq, unsigned priority)
1127 struct evtchn_set_priority set_priority;
1129 set_priority.port = evtchn_from_irq(irq);
1130 set_priority.priority = priority;
1132 return HYPERVISOR_event_channel_op(EVTCHNOP_set_priority,
1135 EXPORT_SYMBOL_GPL(xen_set_irq_priority);
1137 int evtchn_make_refcounted(unsigned int evtchn)
1139 int irq = get_evtchn_to_irq(evtchn);
1140 struct irq_info *info;
1145 info = irq_get_handler_data(irq);
1150 WARN_ON(info->refcnt != -1);
1156 EXPORT_SYMBOL_GPL(evtchn_make_refcounted);
1158 int evtchn_get(unsigned int evtchn)
1161 struct irq_info *info;
1164 if (evtchn >= xen_evtchn_max_channels())
1167 mutex_lock(&irq_mapping_update_lock);
1169 irq = get_evtchn_to_irq(evtchn);
1173 info = irq_get_handler_data(irq);
1179 if (info->refcnt <= 0)
1185 mutex_unlock(&irq_mapping_update_lock);
1189 EXPORT_SYMBOL_GPL(evtchn_get);
1191 void evtchn_put(unsigned int evtchn)
1193 int irq = get_evtchn_to_irq(evtchn);
1194 if (WARN_ON(irq == -1))
1196 unbind_from_irq(irq);
1198 EXPORT_SYMBOL_GPL(evtchn_put);
1200 void xen_send_IPI_one(unsigned int cpu, enum ipi_vector vector)
1205 if (unlikely(vector == XEN_NMI_VECTOR)) {
1206 int rc = HYPERVISOR_vcpu_op(VCPUOP_send_nmi, cpu, NULL);
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);
1251 inc_irq_stat(irq_hv_callback_count);
1254 __xen_evtchn_do_upcall();
1257 set_irq_regs(old_regs);
1260 void xen_hvm_evtchn_do_upcall(void)
1262 __xen_evtchn_do_upcall();
1264 EXPORT_SYMBOL_GPL(xen_hvm_evtchn_do_upcall);
1266 /* Rebind a new event channel to an existing irq. */
1267 void rebind_evtchn_irq(int evtchn, int irq)
1269 struct irq_info *info = info_for_irq(irq);
1274 /* Make sure the irq is masked, since the new event channel
1275 will also be masked. */
1278 mutex_lock(&irq_mapping_update_lock);
1280 /* After resume the irq<->evtchn mappings are all cleared out */
1281 BUG_ON(get_evtchn_to_irq(evtchn) != -1);
1282 /* Expect irq to have been bound before,
1283 so there should be a proper type */
1284 BUG_ON(info->type == IRQT_UNBOUND);
1286 (void)xen_irq_info_evtchn_setup(irq, evtchn);
1288 mutex_unlock(&irq_mapping_update_lock);
1290 /* new event channels are always bound to cpu 0 */
1291 irq_set_affinity(irq, cpumask_of(0));
1293 /* Unmask the event channel. */
1297 /* Rebind an evtchn so that it gets delivered to a specific cpu */
1298 static int rebind_irq_to_cpu(unsigned irq, unsigned tcpu)
1300 struct evtchn_bind_vcpu bind_vcpu;
1301 int evtchn = evtchn_from_irq(irq);
1304 if (!VALID_EVTCHN(evtchn))
1308 * Events delivered via platform PCI interrupts are always
1309 * routed to vcpu 0 and hence cannot be rebound.
1311 if (xen_hvm_domain() && !xen_have_vector_callback)
1314 /* Send future instances of this interrupt to other vcpu. */
1315 bind_vcpu.port = evtchn;
1316 bind_vcpu.vcpu = tcpu;
1319 * Mask the event while changing the VCPU binding to prevent
1320 * it being delivered on an unexpected VCPU.
1322 masked = test_and_set_mask(evtchn);
1325 * If this fails, it usually just indicates that we're dealing with a
1326 * virq or IPI channel, which don't actually need to be rebound. Ignore
1327 * it, but don't do the xenlinux-level rebind in that case.
1329 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_vcpu, &bind_vcpu) >= 0)
1330 bind_evtchn_to_cpu(evtchn, tcpu);
1333 unmask_evtchn(evtchn);
1338 static int set_affinity_irq(struct irq_data *data, const struct cpumask *dest,
1341 unsigned tcpu = cpumask_first_and(dest, cpu_online_mask);
1343 return rebind_irq_to_cpu(data->irq, 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);
1368 if (VALID_EVTCHN(evtchn))
1369 clear_evtchn(evtchn);
1372 static void mask_ack_dynirq(struct irq_data *data)
1374 disable_dynirq(data);
1378 static int retrigger_dynirq(struct irq_data *data)
1380 unsigned int evtchn = evtchn_from_irq(data->irq);
1383 if (!VALID_EVTCHN(evtchn))
1386 masked = test_and_set_mask(evtchn);
1389 unmask_evtchn(evtchn);
1394 static void restore_pirqs(void)
1396 int pirq, rc, irq, gsi;
1397 struct physdev_map_pirq map_irq;
1398 struct irq_info *info;
1400 list_for_each_entry(info, &xen_irq_list_head, list) {
1401 if (info->type != IRQT_PIRQ)
1404 pirq = info->u.pirq.pirq;
1405 gsi = info->u.pirq.gsi;
1408 /* save/restore of PT devices doesn't work, so at this point the
1409 * only devices present are GSI based emulated devices */
1413 map_irq.domid = DOMID_SELF;
1414 map_irq.type = MAP_PIRQ_TYPE_GSI;
1415 map_irq.index = gsi;
1416 map_irq.pirq = pirq;
1418 rc = HYPERVISOR_physdev_op(PHYSDEVOP_map_pirq, &map_irq);
1420 pr_warn("xen map irq failed gsi=%d irq=%d pirq=%d rc=%d\n",
1421 gsi, irq, pirq, rc);
1426 printk(KERN_DEBUG "xen: --> irq=%d, pirq=%d\n", irq, map_irq.pirq);
1428 __startup_pirq(irq);
1432 static void restore_cpu_virqs(unsigned int cpu)
1434 struct evtchn_bind_virq bind_virq;
1435 int virq, irq, evtchn;
1437 for (virq = 0; virq < NR_VIRQS; virq++) {
1438 if ((irq = per_cpu(virq_to_irq, cpu)[virq]) == -1)
1441 BUG_ON(virq_from_irq(irq) != virq);
1443 /* Get a new binding from Xen. */
1444 bind_virq.virq = virq;
1445 bind_virq.vcpu = cpu;
1446 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_virq,
1449 evtchn = bind_virq.port;
1451 /* Record the new mapping. */
1452 (void)xen_irq_info_virq_setup(cpu, irq, evtchn, virq);
1453 bind_evtchn_to_cpu(evtchn, cpu);
1457 static void restore_cpu_ipis(unsigned int cpu)
1459 struct evtchn_bind_ipi bind_ipi;
1460 int ipi, irq, evtchn;
1462 for (ipi = 0; ipi < XEN_NR_IPIS; ipi++) {
1463 if ((irq = per_cpu(ipi_to_irq, cpu)[ipi]) == -1)
1466 BUG_ON(ipi_from_irq(irq) != ipi);
1468 /* Get a new binding from Xen. */
1469 bind_ipi.vcpu = cpu;
1470 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_ipi,
1473 evtchn = bind_ipi.port;
1475 /* Record the new mapping. */
1476 (void)xen_irq_info_ipi_setup(cpu, irq, evtchn, ipi);
1477 bind_evtchn_to_cpu(evtchn, cpu);
1481 /* Clear an irq's pending state, in preparation for polling on it */
1482 void xen_clear_irq_pending(int irq)
1484 int evtchn = evtchn_from_irq(irq);
1486 if (VALID_EVTCHN(evtchn))
1487 clear_evtchn(evtchn);
1489 EXPORT_SYMBOL(xen_clear_irq_pending);
1490 void xen_set_irq_pending(int irq)
1492 int evtchn = evtchn_from_irq(irq);
1494 if (VALID_EVTCHN(evtchn))
1498 bool xen_test_irq_pending(int irq)
1500 int evtchn = evtchn_from_irq(irq);
1503 if (VALID_EVTCHN(evtchn))
1504 ret = test_evtchn(evtchn);
1509 /* Poll waiting for an irq to become pending with timeout. In the usual case,
1510 * the irq will be disabled so it won't deliver an interrupt. */
1511 void xen_poll_irq_timeout(int irq, u64 timeout)
1513 evtchn_port_t evtchn = evtchn_from_irq(irq);
1515 if (VALID_EVTCHN(evtchn)) {
1516 struct sched_poll poll;
1519 poll.timeout = timeout;
1520 set_xen_guest_handle(poll.ports, &evtchn);
1522 if (HYPERVISOR_sched_op(SCHEDOP_poll, &poll) != 0)
1526 EXPORT_SYMBOL(xen_poll_irq_timeout);
1527 /* Poll waiting for an irq to become pending. In the usual case, the
1528 * irq will be disabled so it won't deliver an interrupt. */
1529 void xen_poll_irq(int irq)
1531 xen_poll_irq_timeout(irq, 0 /* no timeout */);
1534 /* Check whether the IRQ line is shared with other guests. */
1535 int xen_test_irq_shared(int irq)
1537 struct irq_info *info = info_for_irq(irq);
1538 struct physdev_irq_status_query irq_status;
1543 irq_status.irq = info->u.pirq.pirq;
1545 if (HYPERVISOR_physdev_op(PHYSDEVOP_irq_status_query, &irq_status))
1547 return !(irq_status.flags & XENIRQSTAT_shared);
1549 EXPORT_SYMBOL_GPL(xen_test_irq_shared);
1551 void xen_irq_resume(void)
1554 struct irq_info *info;
1556 /* New event-channel space is not 'live' yet. */
1557 xen_evtchn_mask_all();
1558 xen_evtchn_resume();
1560 /* No IRQ <-> event-channel mappings. */
1561 list_for_each_entry(info, &xen_irq_list_head, list)
1562 info->evtchn = 0; /* zap event-channel binding */
1564 clear_evtchn_to_irq_all();
1566 for_each_possible_cpu(cpu) {
1567 restore_cpu_virqs(cpu);
1568 restore_cpu_ipis(cpu);
1574 static struct irq_chip xen_dynamic_chip __read_mostly = {
1577 .irq_disable = disable_dynirq,
1578 .irq_mask = disable_dynirq,
1579 .irq_unmask = enable_dynirq,
1581 .irq_ack = ack_dynirq,
1582 .irq_mask_ack = mask_ack_dynirq,
1584 .irq_set_affinity = set_affinity_irq,
1585 .irq_retrigger = retrigger_dynirq,
1588 static struct irq_chip xen_pirq_chip __read_mostly = {
1591 .irq_startup = startup_pirq,
1592 .irq_shutdown = shutdown_pirq,
1593 .irq_enable = enable_pirq,
1594 .irq_disable = disable_pirq,
1596 .irq_mask = disable_dynirq,
1597 .irq_unmask = enable_dynirq,
1599 .irq_ack = eoi_pirq,
1600 .irq_eoi = eoi_pirq,
1601 .irq_mask_ack = mask_ack_pirq,
1603 .irq_set_affinity = set_affinity_irq,
1605 .irq_retrigger = retrigger_dynirq,
1608 static struct irq_chip xen_percpu_chip __read_mostly = {
1609 .name = "xen-percpu",
1611 .irq_disable = disable_dynirq,
1612 .irq_mask = disable_dynirq,
1613 .irq_unmask = enable_dynirq,
1615 .irq_ack = ack_dynirq,
1618 int xen_set_callback_via(uint64_t via)
1620 struct xen_hvm_param a;
1621 a.domid = DOMID_SELF;
1622 a.index = HVM_PARAM_CALLBACK_IRQ;
1624 return HYPERVISOR_hvm_op(HVMOP_set_param, &a);
1626 EXPORT_SYMBOL_GPL(xen_set_callback_via);
1628 #ifdef CONFIG_XEN_PVHVM
1629 /* Vector callbacks are better than PCI interrupts to receive event
1630 * channel notifications because we can receive vector callbacks on any
1631 * vcpu and we don't need PCI support or APIC interactions. */
1632 void xen_callback_vector(void)
1635 uint64_t callback_via;
1636 if (xen_have_vector_callback) {
1637 callback_via = HVM_CALLBACK_VECTOR(HYPERVISOR_CALLBACK_VECTOR);
1638 rc = xen_set_callback_via(callback_via);
1640 pr_err("Request for Xen HVM callback vector failed\n");
1641 xen_have_vector_callback = 0;
1644 pr_info("Xen HVM callback vector for event delivery is enabled\n");
1645 /* in the restore case the vector has already been allocated */
1646 if (!test_bit(HYPERVISOR_CALLBACK_VECTOR, used_vectors))
1647 alloc_intr_gate(HYPERVISOR_CALLBACK_VECTOR,
1648 xen_hvm_callback_vector);
1652 void xen_callback_vector(void) {}
1655 #undef MODULE_PARAM_PREFIX
1656 #define MODULE_PARAM_PREFIX "xen."
1658 static bool fifo_events = true;
1659 module_param(fifo_events, bool, 0);
1661 void __init xen_init_IRQ(void)
1666 ret = xen_evtchn_fifo_init();
1668 xen_evtchn_2l_init();
1670 evtchn_to_irq = kcalloc(EVTCHN_ROW(xen_evtchn_max_channels()),
1671 sizeof(*evtchn_to_irq), GFP_KERNEL);
1672 BUG_ON(!evtchn_to_irq);
1674 /* No event channels are 'live' right now. */
1675 xen_evtchn_mask_all();
1677 pirq_needs_eoi = pirq_needs_eoi_flag;
1680 if (xen_pv_domain()) {
1681 irq_ctx_init(smp_processor_id());
1682 if (xen_initial_domain())
1683 pci_xen_initial_domain();
1685 if (xen_feature(XENFEAT_hvm_callback_vector))
1686 xen_callback_vector();
1688 if (xen_hvm_domain()) {
1690 /* pci_xen_hvm_init must be called after native_init_IRQ so that
1691 * __acpi_register_gsi can point at the right function */
1695 struct physdev_pirq_eoi_gmfn eoi_gmfn;
1697 pirq_eoi_map = (void *)__get_free_page(GFP_KERNEL|__GFP_ZERO);
1698 eoi_gmfn.gmfn = virt_to_mfn(pirq_eoi_map);
1699 rc = HYPERVISOR_physdev_op(PHYSDEVOP_pirq_eoi_gmfn_v2, &eoi_gmfn);
1700 /* TODO: No PVH support for PIRQ EOI */
1702 free_page((unsigned long) pirq_eoi_map);
1703 pirq_eoi_map = NULL;
1705 pirq_needs_eoi = pirq_check_eoi_map;