4 * Xen models interrupts with abstract event channels. Because each
5 * domain gets 1024 event channels, but NR_IRQ is not that large, we
6 * must dynamically map irqs<->event channels. The event channels
7 * interface with the rest of the kernel by defining a xen interrupt
8 * chip. When an event is received, it is mapped to an irq and sent
9 * through the normal interrupt processing path.
11 * There are four kinds of events which can be mapped to an event
14 * 1. Inter-domain notifications. This includes all the virtual
15 * device events, since they're driven by front-ends in another domain
17 * 2. VIRQs, typically used for timers. These are per-cpu events.
19 * 4. PIRQs - Hardware interrupts.
21 * Jeremy Fitzhardinge <jeremy@xensource.com>, XenSource Inc, 2007
24 #define pr_fmt(fmt) "xen:" KBUILD_MODNAME ": " fmt
26 #include <linux/linkage.h>
27 #include <linux/interrupt.h>
28 #include <linux/irq.h>
29 #include <linux/module.h>
30 #include <linux/string.h>
31 #include <linux/bootmem.h>
32 #include <linux/slab.h>
33 #include <linux/irqnr.h>
34 #include <linux/pci.h>
38 #include <asm/ptrace.h>
41 #include <asm/io_apic.h>
42 #include <asm/xen/page.h>
43 #include <asm/xen/pci.h>
45 #include <asm/sync_bitops.h>
46 #include <asm/xen/hypercall.h>
47 #include <asm/xen/hypervisor.h>
51 #include <xen/xen-ops.h>
52 #include <xen/events.h>
53 #include <xen/interface/xen.h>
54 #include <xen/interface/event_channel.h>
55 #include <xen/interface/hvm/hvm_op.h>
56 #include <xen/interface/hvm/params.h>
57 #include <xen/interface/physdev.h>
58 #include <xen/interface/sched.h>
59 #include <xen/interface/vcpu.h>
60 #include <asm/hw_irq.h>
62 #include "events_internal.h"
64 const struct evtchn_ops *evtchn_ops;
67 * This lock protects updates to the following mapping and reference-count
68 * arrays. The lock does not need to be acquired to read the mapping tables.
70 static DEFINE_MUTEX(irq_mapping_update_lock);
72 static LIST_HEAD(xen_irq_list_head);
74 /* IRQ <-> VIRQ mapping. */
75 static DEFINE_PER_CPU(int [NR_VIRQS], virq_to_irq) = {[0 ... NR_VIRQS-1] = -1};
77 /* IRQ <-> IPI mapping */
78 static DEFINE_PER_CPU(int [XEN_NR_IPIS], ipi_to_irq) = {[0 ... XEN_NR_IPIS-1] = -1};
82 static unsigned long *pirq_eoi_map;
84 static bool (*pirq_needs_eoi)(unsigned irq);
86 #define EVTCHN_ROW(e) (e / (PAGE_SIZE/sizeof(**evtchn_to_irq)))
87 #define EVTCHN_COL(e) (e % (PAGE_SIZE/sizeof(**evtchn_to_irq)))
88 #define EVTCHN_PER_ROW (PAGE_SIZE / sizeof(**evtchn_to_irq))
90 /* Xen will never allocate port zero for any purpose. */
91 #define VALID_EVTCHN(chn) ((chn) != 0)
93 static struct irq_chip xen_dynamic_chip;
94 static struct irq_chip xen_percpu_chip;
95 static struct irq_chip xen_pirq_chip;
96 static void enable_dynirq(struct irq_data *data);
97 static void disable_dynirq(struct irq_data *data);
99 static void clear_evtchn_to_irq_row(unsigned row)
103 for (col = 0; col < EVTCHN_PER_ROW; col++)
104 evtchn_to_irq[row][col] = -1;
107 static void clear_evtchn_to_irq_all(void)
111 for (row = 0; row < EVTCHN_ROW(xen_evtchn_max_channels()); row++) {
112 if (evtchn_to_irq[row] == NULL)
114 clear_evtchn_to_irq_row(row);
118 static int set_evtchn_to_irq(unsigned evtchn, unsigned irq)
123 if (evtchn >= xen_evtchn_max_channels())
126 row = EVTCHN_ROW(evtchn);
127 col = EVTCHN_COL(evtchn);
129 if (evtchn_to_irq[row] == NULL) {
130 /* Unallocated irq entries return -1 anyway */
134 evtchn_to_irq[row] = (int *)get_zeroed_page(GFP_KERNEL);
135 if (evtchn_to_irq[row] == NULL)
138 clear_evtchn_to_irq_row(row);
141 evtchn_to_irq[EVTCHN_ROW(evtchn)][EVTCHN_COL(evtchn)] = irq;
145 int get_evtchn_to_irq(unsigned evtchn)
147 if (evtchn >= xen_evtchn_max_channels())
149 if (evtchn_to_irq[EVTCHN_ROW(evtchn)] == NULL)
151 return evtchn_to_irq[EVTCHN_ROW(evtchn)][EVTCHN_COL(evtchn)];
154 /* Get info for IRQ */
155 struct irq_info *info_for_irq(unsigned irq)
157 return irq_get_handler_data(irq);
160 /* Constructors for packed IRQ information. */
161 static int xen_irq_info_common_setup(struct irq_info *info,
163 enum xen_irq_type type,
169 BUG_ON(info->type != IRQT_UNBOUND && info->type != type);
173 info->evtchn = evtchn;
176 ret = set_evtchn_to_irq(evtchn, irq);
180 irq_clear_status_flags(irq, IRQ_NOREQUEST|IRQ_NOAUTOEN);
182 return xen_evtchn_port_setup(info);
185 static int xen_irq_info_evtchn_setup(unsigned irq,
188 struct irq_info *info = info_for_irq(irq);
190 return xen_irq_info_common_setup(info, irq, IRQT_EVTCHN, evtchn, 0);
193 static int xen_irq_info_ipi_setup(unsigned cpu,
198 struct irq_info *info = info_for_irq(irq);
202 per_cpu(ipi_to_irq, cpu)[ipi] = irq;
204 return xen_irq_info_common_setup(info, irq, IRQT_IPI, evtchn, 0);
207 static int xen_irq_info_virq_setup(unsigned cpu,
212 struct irq_info *info = info_for_irq(irq);
216 per_cpu(virq_to_irq, cpu)[virq] = irq;
218 return xen_irq_info_common_setup(info, irq, IRQT_VIRQ, evtchn, 0);
221 static int xen_irq_info_pirq_setup(unsigned irq,
228 struct irq_info *info = info_for_irq(irq);
230 info->u.pirq.pirq = pirq;
231 info->u.pirq.gsi = gsi;
232 info->u.pirq.domid = domid;
233 info->u.pirq.flags = flags;
235 return xen_irq_info_common_setup(info, irq, IRQT_PIRQ, evtchn, 0);
238 static void xen_irq_info_cleanup(struct irq_info *info)
240 set_evtchn_to_irq(info->evtchn, -1);
245 * Accessors for packed IRQ information.
247 unsigned int evtchn_from_irq(unsigned irq)
249 if (unlikely(WARN(irq < 0 || irq >= nr_irqs, "Invalid irq %d!\n", irq)))
252 return info_for_irq(irq)->evtchn;
255 unsigned irq_from_evtchn(unsigned int evtchn)
257 return get_evtchn_to_irq(evtchn);
259 EXPORT_SYMBOL_GPL(irq_from_evtchn);
261 int irq_from_virq(unsigned int cpu, unsigned int virq)
263 return per_cpu(virq_to_irq, cpu)[virq];
266 static enum ipi_vector ipi_from_irq(unsigned irq)
268 struct irq_info *info = info_for_irq(irq);
270 BUG_ON(info == NULL);
271 BUG_ON(info->type != IRQT_IPI);
276 static unsigned virq_from_irq(unsigned irq)
278 struct irq_info *info = info_for_irq(irq);
280 BUG_ON(info == NULL);
281 BUG_ON(info->type != IRQT_VIRQ);
286 static unsigned pirq_from_irq(unsigned irq)
288 struct irq_info *info = info_for_irq(irq);
290 BUG_ON(info == NULL);
291 BUG_ON(info->type != IRQT_PIRQ);
293 return info->u.pirq.pirq;
296 static enum xen_irq_type type_from_irq(unsigned irq)
298 return info_for_irq(irq)->type;
301 unsigned cpu_from_irq(unsigned irq)
303 return info_for_irq(irq)->cpu;
306 unsigned int cpu_from_evtchn(unsigned int evtchn)
308 int irq = get_evtchn_to_irq(evtchn);
312 ret = cpu_from_irq(irq);
318 static bool pirq_check_eoi_map(unsigned irq)
320 return test_bit(pirq_from_irq(irq), pirq_eoi_map);
324 static bool pirq_needs_eoi_flag(unsigned irq)
326 struct irq_info *info = info_for_irq(irq);
327 BUG_ON(info->type != IRQT_PIRQ);
329 return info->u.pirq.flags & PIRQ_NEEDS_EOI;
332 static void bind_evtchn_to_cpu(unsigned int chn, unsigned int cpu)
334 int irq = get_evtchn_to_irq(chn);
335 struct irq_info *info = info_for_irq(irq);
339 cpumask_copy(irq_to_desc(irq)->irq_data.affinity, cpumask_of(cpu));
342 xen_evtchn_port_bind_to_cpu(info, cpu);
347 static void xen_evtchn_mask_all(void)
351 for (evtchn = 0; evtchn < xen_evtchn_nr_channels(); evtchn++)
356 * notify_remote_via_irq - send event to remote end of event channel via irq
357 * @irq: irq of event channel to send event to
359 * Unlike notify_remote_via_evtchn(), this is safe to use across
360 * save/restore. Notifications on a broken connection are silently
363 void notify_remote_via_irq(int irq)
365 int evtchn = evtchn_from_irq(irq);
367 if (VALID_EVTCHN(evtchn))
368 notify_remote_via_evtchn(evtchn);
370 EXPORT_SYMBOL_GPL(notify_remote_via_irq);
372 static void xen_irq_init(unsigned irq)
374 struct irq_info *info;
376 struct irq_desc *desc = irq_to_desc(irq);
378 /* By default all event channels notify CPU#0. */
379 cpumask_copy(desc->irq_data.affinity, cpumask_of(0));
382 info = kzalloc(sizeof(*info), GFP_KERNEL);
384 panic("Unable to allocate metadata for IRQ%d\n", irq);
386 info->type = IRQT_UNBOUND;
389 irq_set_handler_data(irq, info);
391 list_add_tail(&info->list, &xen_irq_list_head);
394 static int __must_check xen_allocate_irqs_dynamic(int nvec)
399 #ifdef CONFIG_X86_IO_APIC
401 * For an HVM guest or domain 0 which see "real" (emulated or
402 * actual respectively) GSIs we allocate dynamic IRQs
403 * e.g. those corresponding to event channels or MSIs
404 * etc. from the range above those "real" GSIs to avoid
407 if (xen_initial_domain() || xen_hvm_domain())
408 first = get_nr_irqs_gsi();
411 irq = irq_alloc_descs_from(first, nvec, -1);
414 for (i = 0; i < nvec; i++)
415 xen_irq_init(irq + i);
421 static inline int __must_check xen_allocate_irq_dynamic(void)
424 return xen_allocate_irqs_dynamic(1);
427 static int __must_check xen_allocate_irq_gsi(unsigned gsi)
432 * A PV guest has no concept of a GSI (since it has no ACPI
433 * nor access to/knowledge of the physical APICs). Therefore
434 * all IRQs are dynamically allocated from the entire IRQ
437 if (xen_pv_domain() && !xen_initial_domain())
438 return xen_allocate_irq_dynamic();
440 /* Legacy IRQ descriptors are already allocated by the arch. */
441 if (gsi < NR_IRQS_LEGACY)
444 irq = irq_alloc_desc_at(gsi, -1);
451 static void xen_free_irq(unsigned irq)
453 struct irq_info *info = irq_get_handler_data(irq);
458 list_del(&info->list);
460 irq_set_handler_data(irq, NULL);
462 WARN_ON(info->refcnt > 0);
466 /* Legacy IRQ descriptors are managed by the arch. */
467 if (irq < NR_IRQS_LEGACY)
473 static void xen_evtchn_close(unsigned int port)
475 struct evtchn_close close;
478 if (HYPERVISOR_event_channel_op(EVTCHNOP_close, &close) != 0)
482 static void pirq_query_unmask(int irq)
484 struct physdev_irq_status_query irq_status;
485 struct irq_info *info = info_for_irq(irq);
487 BUG_ON(info->type != IRQT_PIRQ);
489 irq_status.irq = pirq_from_irq(irq);
490 if (HYPERVISOR_physdev_op(PHYSDEVOP_irq_status_query, &irq_status))
491 irq_status.flags = 0;
493 info->u.pirq.flags &= ~PIRQ_NEEDS_EOI;
494 if (irq_status.flags & XENIRQSTAT_needs_eoi)
495 info->u.pirq.flags |= PIRQ_NEEDS_EOI;
498 static bool probing_irq(int irq)
500 struct irq_desc *desc = irq_to_desc(irq);
502 return desc && desc->action == NULL;
505 static void eoi_pirq(struct irq_data *data)
507 int evtchn = evtchn_from_irq(data->irq);
508 struct physdev_eoi eoi = { .irq = pirq_from_irq(data->irq) };
513 if (VALID_EVTCHN(evtchn))
514 clear_evtchn(evtchn);
516 if (pirq_needs_eoi(data->irq)) {
517 rc = HYPERVISOR_physdev_op(PHYSDEVOP_eoi, &eoi);
522 static void mask_ack_pirq(struct irq_data *data)
524 disable_dynirq(data);
528 static unsigned int __startup_pirq(unsigned int irq)
530 struct evtchn_bind_pirq bind_pirq;
531 struct irq_info *info = info_for_irq(irq);
532 int evtchn = evtchn_from_irq(irq);
535 BUG_ON(info->type != IRQT_PIRQ);
537 if (VALID_EVTCHN(evtchn))
540 bind_pirq.pirq = pirq_from_irq(irq);
541 /* NB. We are happy to share unless we are probing. */
542 bind_pirq.flags = info->u.pirq.flags & PIRQ_SHAREABLE ?
543 BIND_PIRQ__WILL_SHARE : 0;
544 rc = HYPERVISOR_event_channel_op(EVTCHNOP_bind_pirq, &bind_pirq);
546 if (!probing_irq(irq))
547 pr_info("Failed to obtain physical IRQ %d\n", irq);
550 evtchn = bind_pirq.port;
552 pirq_query_unmask(irq);
554 rc = set_evtchn_to_irq(evtchn, irq);
556 pr_err("irq%d: Failed to set port to irq mapping (%d)\n",
558 xen_evtchn_close(evtchn);
561 bind_evtchn_to_cpu(evtchn, 0);
562 info->evtchn = evtchn;
565 unmask_evtchn(evtchn);
566 eoi_pirq(irq_get_irq_data(irq));
571 static unsigned int startup_pirq(struct irq_data *data)
573 return __startup_pirq(data->irq);
576 static void shutdown_pirq(struct irq_data *data)
578 unsigned int irq = data->irq;
579 struct irq_info *info = info_for_irq(irq);
580 unsigned evtchn = evtchn_from_irq(irq);
582 BUG_ON(info->type != IRQT_PIRQ);
584 if (!VALID_EVTCHN(evtchn))
588 xen_evtchn_close(evtchn);
589 xen_irq_info_cleanup(info);
592 static void enable_pirq(struct irq_data *data)
597 static void disable_pirq(struct irq_data *data)
599 disable_dynirq(data);
602 int xen_irq_from_gsi(unsigned gsi)
604 struct irq_info *info;
606 list_for_each_entry(info, &xen_irq_list_head, list) {
607 if (info->type != IRQT_PIRQ)
610 if (info->u.pirq.gsi == gsi)
616 EXPORT_SYMBOL_GPL(xen_irq_from_gsi);
618 static void __unbind_from_irq(unsigned int irq)
620 int evtchn = evtchn_from_irq(irq);
621 struct irq_info *info = irq_get_handler_data(irq);
623 if (info->refcnt > 0) {
625 if (info->refcnt != 0)
629 if (VALID_EVTCHN(evtchn)) {
630 unsigned int cpu = cpu_from_irq(irq);
632 xen_evtchn_close(evtchn);
634 switch (type_from_irq(irq)) {
636 per_cpu(virq_to_irq, cpu)[virq_from_irq(irq)] = -1;
639 per_cpu(ipi_to_irq, cpu)[ipi_from_irq(irq)] = -1;
645 xen_irq_info_cleanup(info);
648 BUG_ON(info_for_irq(irq)->type == IRQT_UNBOUND);
654 * Do not make any assumptions regarding the relationship between the
655 * IRQ number returned here and the Xen pirq argument.
657 * Note: We don't assign an event channel until the irq actually started
658 * up. Return an existing irq if we've already got one for the gsi.
660 * Shareable implies level triggered, not shareable implies edge
663 int xen_bind_pirq_gsi_to_irq(unsigned gsi,
664 unsigned pirq, int shareable, char *name)
667 struct physdev_irq irq_op;
670 mutex_lock(&irq_mapping_update_lock);
672 irq = xen_irq_from_gsi(gsi);
674 pr_info("%s: returning irq %d for gsi %u\n",
679 irq = xen_allocate_irq_gsi(gsi);
686 /* Only the privileged domain can do this. For non-priv, the pcifront
687 * driver provides a PCI bus that does the call to do exactly
688 * this in the priv domain. */
689 if (xen_initial_domain() &&
690 HYPERVISOR_physdev_op(PHYSDEVOP_alloc_irq_vector, &irq_op)) {
696 ret = xen_irq_info_pirq_setup(irq, 0, pirq, gsi, DOMID_SELF,
697 shareable ? PIRQ_SHAREABLE : 0);
699 __unbind_from_irq(irq);
704 pirq_query_unmask(irq);
705 /* We try to use the handler with the appropriate semantic for the
706 * type of interrupt: if the interrupt is an edge triggered
707 * interrupt we use handle_edge_irq.
709 * On the other hand if the interrupt is level triggered we use
710 * handle_fasteoi_irq like the native code does for this kind of
713 * Depending on the Xen version, pirq_needs_eoi might return true
714 * not only for level triggered interrupts but for edge triggered
715 * interrupts too. In any case Xen always honors the eoi mechanism,
716 * not injecting any more pirqs of the same kind if the first one
717 * hasn't received an eoi yet. Therefore using the fasteoi handler
718 * is the right choice either way.
721 irq_set_chip_and_handler_name(irq, &xen_pirq_chip,
722 handle_fasteoi_irq, name);
724 irq_set_chip_and_handler_name(irq, &xen_pirq_chip,
725 handle_edge_irq, name);
728 mutex_unlock(&irq_mapping_update_lock);
733 #ifdef CONFIG_PCI_MSI
734 int xen_allocate_pirq_msi(struct pci_dev *dev, struct msi_desc *msidesc)
737 struct physdev_get_free_pirq op_get_free_pirq;
739 op_get_free_pirq.type = MAP_PIRQ_TYPE_MSI;
740 rc = HYPERVISOR_physdev_op(PHYSDEVOP_get_free_pirq, &op_get_free_pirq);
742 WARN_ONCE(rc == -ENOSYS,
743 "hypervisor does not support the PHYSDEVOP_get_free_pirq interface\n");
745 return rc ? -1 : op_get_free_pirq.pirq;
748 int xen_bind_pirq_msi_to_irq(struct pci_dev *dev, struct msi_desc *msidesc,
749 int pirq, int nvec, const char *name, domid_t domid)
753 mutex_lock(&irq_mapping_update_lock);
755 irq = xen_allocate_irqs_dynamic(nvec);
759 for (i = 0; i < nvec; i++) {
760 irq_set_chip_and_handler_name(irq + i, &xen_pirq_chip, handle_edge_irq, name);
762 ret = xen_irq_info_pirq_setup(irq + i, 0, pirq + i, 0, domid,
763 i == 0 ? 0 : PIRQ_MSI_GROUP);
768 ret = irq_set_msi_desc(irq, msidesc);
772 mutex_unlock(&irq_mapping_update_lock);
776 __unbind_from_irq(irq + i);
777 mutex_unlock(&irq_mapping_update_lock);
782 int xen_destroy_irq(int irq)
784 struct irq_desc *desc;
785 struct physdev_unmap_pirq unmap_irq;
786 struct irq_info *info = info_for_irq(irq);
789 mutex_lock(&irq_mapping_update_lock);
791 desc = irq_to_desc(irq);
796 * If trying to remove a vector in a MSI group different
797 * than the first one skip the PIRQ unmap unless this vector
798 * is the first one in the group.
800 if (xen_initial_domain() && !(info->u.pirq.flags & PIRQ_MSI_GROUP)) {
801 unmap_irq.pirq = info->u.pirq.pirq;
802 unmap_irq.domid = info->u.pirq.domid;
803 rc = HYPERVISOR_physdev_op(PHYSDEVOP_unmap_pirq, &unmap_irq);
804 /* If another domain quits without making the pci_disable_msix
805 * call, the Xen hypervisor takes care of freeing the PIRQs
806 * (free_domain_pirqs).
808 if ((rc == -ESRCH && info->u.pirq.domid != DOMID_SELF))
809 pr_info("domain %d does not have %d anymore\n",
810 info->u.pirq.domid, info->u.pirq.pirq);
812 pr_warn("unmap irq failed %d\n", rc);
820 mutex_unlock(&irq_mapping_update_lock);
824 int xen_irq_from_pirq(unsigned pirq)
828 struct irq_info *info;
830 mutex_lock(&irq_mapping_update_lock);
832 list_for_each_entry(info, &xen_irq_list_head, list) {
833 if (info->type != IRQT_PIRQ)
836 if (info->u.pirq.pirq == pirq)
841 mutex_unlock(&irq_mapping_update_lock);
847 int xen_pirq_from_irq(unsigned irq)
849 return pirq_from_irq(irq);
851 EXPORT_SYMBOL_GPL(xen_pirq_from_irq);
853 int bind_evtchn_to_irq(unsigned int evtchn)
858 if (evtchn >= xen_evtchn_max_channels())
861 mutex_lock(&irq_mapping_update_lock);
863 irq = get_evtchn_to_irq(evtchn);
866 irq = xen_allocate_irq_dynamic();
870 irq_set_chip_and_handler_name(irq, &xen_dynamic_chip,
871 handle_edge_irq, "event");
873 ret = xen_irq_info_evtchn_setup(irq, evtchn);
875 __unbind_from_irq(irq);
879 /* New interdomain events are bound to VCPU 0. */
880 bind_evtchn_to_cpu(evtchn, 0);
882 struct irq_info *info = info_for_irq(irq);
883 WARN_ON(info == NULL || info->type != IRQT_EVTCHN);
887 mutex_unlock(&irq_mapping_update_lock);
891 EXPORT_SYMBOL_GPL(bind_evtchn_to_irq);
893 static int bind_ipi_to_irq(unsigned int ipi, unsigned int cpu)
895 struct evtchn_bind_ipi bind_ipi;
899 mutex_lock(&irq_mapping_update_lock);
901 irq = per_cpu(ipi_to_irq, cpu)[ipi];
904 irq = xen_allocate_irq_dynamic();
908 irq_set_chip_and_handler_name(irq, &xen_percpu_chip,
909 handle_percpu_irq, "ipi");
912 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_ipi,
915 evtchn = bind_ipi.port;
917 ret = xen_irq_info_ipi_setup(cpu, irq, evtchn, ipi);
919 __unbind_from_irq(irq);
923 bind_evtchn_to_cpu(evtchn, cpu);
925 struct irq_info *info = info_for_irq(irq);
926 WARN_ON(info == NULL || info->type != IRQT_IPI);
930 mutex_unlock(&irq_mapping_update_lock);
934 static int bind_interdomain_evtchn_to_irq(unsigned int remote_domain,
935 unsigned int remote_port)
937 struct evtchn_bind_interdomain bind_interdomain;
940 bind_interdomain.remote_dom = remote_domain;
941 bind_interdomain.remote_port = remote_port;
943 err = HYPERVISOR_event_channel_op(EVTCHNOP_bind_interdomain,
946 return err ? : bind_evtchn_to_irq(bind_interdomain.local_port);
949 static int find_virq(unsigned int virq, unsigned int cpu)
951 struct evtchn_status status;
952 int port, rc = -ENOENT;
954 memset(&status, 0, sizeof(status));
955 for (port = 0; port < xen_evtchn_max_channels(); port++) {
956 status.dom = DOMID_SELF;
958 rc = HYPERVISOR_event_channel_op(EVTCHNOP_status, &status);
961 if (status.status != EVTCHNSTAT_virq)
963 if (status.u.virq == virq && status.vcpu == cpu) {
972 * xen_evtchn_nr_channels - number of usable event channel ports
974 * This may be less than the maximum supported by the current
975 * hypervisor ABI. Use xen_evtchn_max_channels() for the maximum
978 unsigned xen_evtchn_nr_channels(void)
980 return evtchn_ops->nr_channels();
982 EXPORT_SYMBOL_GPL(xen_evtchn_nr_channels);
984 int bind_virq_to_irq(unsigned int virq, unsigned int cpu)
986 struct evtchn_bind_virq bind_virq;
987 int evtchn, irq, ret;
989 mutex_lock(&irq_mapping_update_lock);
991 irq = per_cpu(virq_to_irq, cpu)[virq];
994 irq = xen_allocate_irq_dynamic();
998 irq_set_chip_and_handler_name(irq, &xen_percpu_chip,
999 handle_percpu_irq, "virq");
1001 bind_virq.virq = virq;
1002 bind_virq.vcpu = cpu;
1003 ret = HYPERVISOR_event_channel_op(EVTCHNOP_bind_virq,
1006 evtchn = bind_virq.port;
1009 ret = find_virq(virq, cpu);
1014 ret = xen_irq_info_virq_setup(cpu, irq, evtchn, virq);
1016 __unbind_from_irq(irq);
1021 bind_evtchn_to_cpu(evtchn, cpu);
1023 struct irq_info *info = info_for_irq(irq);
1024 WARN_ON(info == NULL || info->type != IRQT_VIRQ);
1028 mutex_unlock(&irq_mapping_update_lock);
1033 static void unbind_from_irq(unsigned int irq)
1035 mutex_lock(&irq_mapping_update_lock);
1036 __unbind_from_irq(irq);
1037 mutex_unlock(&irq_mapping_update_lock);
1040 int bind_evtchn_to_irqhandler(unsigned int evtchn,
1041 irq_handler_t handler,
1042 unsigned long irqflags,
1043 const char *devname, void *dev_id)
1047 irq = bind_evtchn_to_irq(evtchn);
1050 retval = request_irq(irq, handler, irqflags, devname, dev_id);
1052 unbind_from_irq(irq);
1058 EXPORT_SYMBOL_GPL(bind_evtchn_to_irqhandler);
1060 int bind_interdomain_evtchn_to_irqhandler(unsigned int remote_domain,
1061 unsigned int remote_port,
1062 irq_handler_t handler,
1063 unsigned long irqflags,
1064 const char *devname,
1069 irq = bind_interdomain_evtchn_to_irq(remote_domain, remote_port);
1073 retval = request_irq(irq, handler, irqflags, devname, dev_id);
1075 unbind_from_irq(irq);
1081 EXPORT_SYMBOL_GPL(bind_interdomain_evtchn_to_irqhandler);
1083 int bind_virq_to_irqhandler(unsigned int virq, unsigned int cpu,
1084 irq_handler_t handler,
1085 unsigned long irqflags, const char *devname, void *dev_id)
1089 irq = bind_virq_to_irq(virq, cpu);
1092 retval = request_irq(irq, handler, irqflags, devname, dev_id);
1094 unbind_from_irq(irq);
1100 EXPORT_SYMBOL_GPL(bind_virq_to_irqhandler);
1102 int bind_ipi_to_irqhandler(enum ipi_vector ipi,
1104 irq_handler_t handler,
1105 unsigned long irqflags,
1106 const char *devname,
1111 irq = bind_ipi_to_irq(ipi, cpu);
1115 irqflags |= IRQF_NO_SUSPEND | IRQF_FORCE_RESUME | IRQF_EARLY_RESUME;
1116 retval = request_irq(irq, handler, irqflags, devname, dev_id);
1118 unbind_from_irq(irq);
1125 void unbind_from_irqhandler(unsigned int irq, void *dev_id)
1127 struct irq_info *info = irq_get_handler_data(irq);
1131 free_irq(irq, dev_id);
1132 unbind_from_irq(irq);
1134 EXPORT_SYMBOL_GPL(unbind_from_irqhandler);
1137 * xen_set_irq_priority() - set an event channel priority.
1138 * @irq:irq bound to an event channel.
1139 * @priority: priority between XEN_IRQ_PRIORITY_MAX and XEN_IRQ_PRIORITY_MIN.
1141 int xen_set_irq_priority(unsigned irq, unsigned priority)
1143 struct evtchn_set_priority set_priority;
1145 set_priority.port = evtchn_from_irq(irq);
1146 set_priority.priority = priority;
1148 return HYPERVISOR_event_channel_op(EVTCHNOP_set_priority,
1151 EXPORT_SYMBOL_GPL(xen_set_irq_priority);
1153 int evtchn_make_refcounted(unsigned int evtchn)
1155 int irq = get_evtchn_to_irq(evtchn);
1156 struct irq_info *info;
1161 info = irq_get_handler_data(irq);
1166 WARN_ON(info->refcnt != -1);
1172 EXPORT_SYMBOL_GPL(evtchn_make_refcounted);
1174 int evtchn_get(unsigned int evtchn)
1177 struct irq_info *info;
1180 if (evtchn >= xen_evtchn_max_channels())
1183 mutex_lock(&irq_mapping_update_lock);
1185 irq = get_evtchn_to_irq(evtchn);
1189 info = irq_get_handler_data(irq);
1195 if (info->refcnt <= 0)
1201 mutex_unlock(&irq_mapping_update_lock);
1205 EXPORT_SYMBOL_GPL(evtchn_get);
1207 void evtchn_put(unsigned int evtchn)
1209 int irq = get_evtchn_to_irq(evtchn);
1210 if (WARN_ON(irq == -1))
1212 unbind_from_irq(irq);
1214 EXPORT_SYMBOL_GPL(evtchn_put);
1216 void xen_send_IPI_one(unsigned int cpu, enum ipi_vector vector)
1221 if (unlikely(vector == XEN_NMI_VECTOR)) {
1222 int rc = HYPERVISOR_vcpu_op(VCPUOP_send_nmi, cpu, NULL);
1224 printk(KERN_WARNING "Sending nmi to CPU%d failed (rc:%d)\n", cpu, rc);
1228 irq = per_cpu(ipi_to_irq, cpu)[vector];
1230 notify_remote_via_irq(irq);
1233 static DEFINE_PER_CPU(unsigned, xed_nesting_count);
1235 static void __xen_evtchn_do_upcall(void)
1237 struct vcpu_info *vcpu_info = __this_cpu_read(xen_vcpu);
1238 int cpu = get_cpu();
1242 vcpu_info->evtchn_upcall_pending = 0;
1244 if (__this_cpu_inc_return(xed_nesting_count) - 1)
1247 xen_evtchn_handle_events(cpu);
1249 BUG_ON(!irqs_disabled());
1251 count = __this_cpu_read(xed_nesting_count);
1252 __this_cpu_write(xed_nesting_count, 0);
1253 } while (count != 1 || vcpu_info->evtchn_upcall_pending);
1260 void xen_evtchn_do_upcall(struct pt_regs *regs)
1262 struct pt_regs *old_regs = set_irq_regs(regs);
1269 __xen_evtchn_do_upcall();
1272 set_irq_regs(old_regs);
1275 void xen_hvm_evtchn_do_upcall(void)
1277 __xen_evtchn_do_upcall();
1279 EXPORT_SYMBOL_GPL(xen_hvm_evtchn_do_upcall);
1281 /* Rebind a new event channel to an existing irq. */
1282 void rebind_evtchn_irq(int evtchn, int irq)
1284 struct irq_info *info = info_for_irq(irq);
1289 /* Make sure the irq is masked, since the new event channel
1290 will also be masked. */
1293 mutex_lock(&irq_mapping_update_lock);
1295 /* After resume the irq<->evtchn mappings are all cleared out */
1296 BUG_ON(get_evtchn_to_irq(evtchn) != -1);
1297 /* Expect irq to have been bound before,
1298 so there should be a proper type */
1299 BUG_ON(info->type == IRQT_UNBOUND);
1301 (void)xen_irq_info_evtchn_setup(irq, evtchn);
1303 mutex_unlock(&irq_mapping_update_lock);
1305 /* new event channels are always bound to cpu 0 */
1306 irq_set_affinity(irq, cpumask_of(0));
1308 /* Unmask the event channel. */
1312 /* Rebind an evtchn so that it gets delivered to a specific cpu */
1313 static int rebind_irq_to_cpu(unsigned irq, unsigned tcpu)
1315 struct evtchn_bind_vcpu bind_vcpu;
1316 int evtchn = evtchn_from_irq(irq);
1319 if (!VALID_EVTCHN(evtchn))
1323 * Events delivered via platform PCI interrupts are always
1324 * routed to vcpu 0 and hence cannot be rebound.
1326 if (xen_hvm_domain() && !xen_have_vector_callback)
1329 /* Send future instances of this interrupt to other vcpu. */
1330 bind_vcpu.port = evtchn;
1331 bind_vcpu.vcpu = tcpu;
1334 * Mask the event while changing the VCPU binding to prevent
1335 * it being delivered on an unexpected VCPU.
1337 masked = test_and_set_mask(evtchn);
1340 * If this fails, it usually just indicates that we're dealing with a
1341 * virq or IPI channel, which don't actually need to be rebound. Ignore
1342 * it, but don't do the xenlinux-level rebind in that case.
1344 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_vcpu, &bind_vcpu) >= 0)
1345 bind_evtchn_to_cpu(evtchn, tcpu);
1348 unmask_evtchn(evtchn);
1353 static int set_affinity_irq(struct irq_data *data, const struct cpumask *dest,
1356 unsigned tcpu = cpumask_first(dest);
1358 return rebind_irq_to_cpu(data->irq, tcpu);
1361 static void enable_dynirq(struct irq_data *data)
1363 int evtchn = evtchn_from_irq(data->irq);
1365 if (VALID_EVTCHN(evtchn))
1366 unmask_evtchn(evtchn);
1369 static void disable_dynirq(struct irq_data *data)
1371 int evtchn = evtchn_from_irq(data->irq);
1373 if (VALID_EVTCHN(evtchn))
1374 mask_evtchn(evtchn);
1377 static void ack_dynirq(struct irq_data *data)
1379 int evtchn = evtchn_from_irq(data->irq);
1383 if (VALID_EVTCHN(evtchn))
1384 clear_evtchn(evtchn);
1387 static void mask_ack_dynirq(struct irq_data *data)
1389 disable_dynirq(data);
1393 static int retrigger_dynirq(struct irq_data *data)
1395 unsigned int evtchn = evtchn_from_irq(data->irq);
1398 if (!VALID_EVTCHN(evtchn))
1401 masked = test_and_set_mask(evtchn);
1404 unmask_evtchn(evtchn);
1409 static void restore_pirqs(void)
1411 int pirq, rc, irq, gsi;
1412 struct physdev_map_pirq map_irq;
1413 struct irq_info *info;
1415 list_for_each_entry(info, &xen_irq_list_head, list) {
1416 if (info->type != IRQT_PIRQ)
1419 pirq = info->u.pirq.pirq;
1420 gsi = info->u.pirq.gsi;
1423 /* save/restore of PT devices doesn't work, so at this point the
1424 * only devices present are GSI based emulated devices */
1428 map_irq.domid = DOMID_SELF;
1429 map_irq.type = MAP_PIRQ_TYPE_GSI;
1430 map_irq.index = gsi;
1431 map_irq.pirq = pirq;
1433 rc = HYPERVISOR_physdev_op(PHYSDEVOP_map_pirq, &map_irq);
1435 pr_warn("xen map irq failed gsi=%d irq=%d pirq=%d rc=%d\n",
1436 gsi, irq, pirq, rc);
1441 printk(KERN_DEBUG "xen: --> irq=%d, pirq=%d\n", irq, map_irq.pirq);
1443 __startup_pirq(irq);
1447 static void restore_cpu_virqs(unsigned int cpu)
1449 struct evtchn_bind_virq bind_virq;
1450 int virq, irq, evtchn;
1452 for (virq = 0; virq < NR_VIRQS; virq++) {
1453 if ((irq = per_cpu(virq_to_irq, cpu)[virq]) == -1)
1456 BUG_ON(virq_from_irq(irq) != virq);
1458 /* Get a new binding from Xen. */
1459 bind_virq.virq = virq;
1460 bind_virq.vcpu = cpu;
1461 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_virq,
1464 evtchn = bind_virq.port;
1466 /* Record the new mapping. */
1467 (void)xen_irq_info_virq_setup(cpu, irq, evtchn, virq);
1468 bind_evtchn_to_cpu(evtchn, cpu);
1472 static void restore_cpu_ipis(unsigned int cpu)
1474 struct evtchn_bind_ipi bind_ipi;
1475 int ipi, irq, evtchn;
1477 for (ipi = 0; ipi < XEN_NR_IPIS; ipi++) {
1478 if ((irq = per_cpu(ipi_to_irq, cpu)[ipi]) == -1)
1481 BUG_ON(ipi_from_irq(irq) != ipi);
1483 /* Get a new binding from Xen. */
1484 bind_ipi.vcpu = cpu;
1485 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_ipi,
1488 evtchn = bind_ipi.port;
1490 /* Record the new mapping. */
1491 (void)xen_irq_info_ipi_setup(cpu, irq, evtchn, ipi);
1492 bind_evtchn_to_cpu(evtchn, cpu);
1496 /* Clear an irq's pending state, in preparation for polling on it */
1497 void xen_clear_irq_pending(int irq)
1499 int evtchn = evtchn_from_irq(irq);
1501 if (VALID_EVTCHN(evtchn))
1502 clear_evtchn(evtchn);
1504 EXPORT_SYMBOL(xen_clear_irq_pending);
1505 void xen_set_irq_pending(int irq)
1507 int evtchn = evtchn_from_irq(irq);
1509 if (VALID_EVTCHN(evtchn))
1513 bool xen_test_irq_pending(int irq)
1515 int evtchn = evtchn_from_irq(irq);
1518 if (VALID_EVTCHN(evtchn))
1519 ret = test_evtchn(evtchn);
1524 /* Poll waiting for an irq to become pending with timeout. In the usual case,
1525 * the irq will be disabled so it won't deliver an interrupt. */
1526 void xen_poll_irq_timeout(int irq, u64 timeout)
1528 evtchn_port_t evtchn = evtchn_from_irq(irq);
1530 if (VALID_EVTCHN(evtchn)) {
1531 struct sched_poll poll;
1534 poll.timeout = timeout;
1535 set_xen_guest_handle(poll.ports, &evtchn);
1537 if (HYPERVISOR_sched_op(SCHEDOP_poll, &poll) != 0)
1541 EXPORT_SYMBOL(xen_poll_irq_timeout);
1542 /* Poll waiting for an irq to become pending. In the usual case, the
1543 * irq will be disabled so it won't deliver an interrupt. */
1544 void xen_poll_irq(int irq)
1546 xen_poll_irq_timeout(irq, 0 /* no timeout */);
1549 /* Check whether the IRQ line is shared with other guests. */
1550 int xen_test_irq_shared(int irq)
1552 struct irq_info *info = info_for_irq(irq);
1553 struct physdev_irq_status_query irq_status;
1558 irq_status.irq = info->u.pirq.pirq;
1560 if (HYPERVISOR_physdev_op(PHYSDEVOP_irq_status_query, &irq_status))
1562 return !(irq_status.flags & XENIRQSTAT_shared);
1564 EXPORT_SYMBOL_GPL(xen_test_irq_shared);
1566 void xen_irq_resume(void)
1569 struct irq_info *info;
1571 /* New event-channel space is not 'live' yet. */
1572 xen_evtchn_mask_all();
1573 xen_evtchn_resume();
1575 /* No IRQ <-> event-channel mappings. */
1576 list_for_each_entry(info, &xen_irq_list_head, list)
1577 info->evtchn = 0; /* zap event-channel binding */
1579 clear_evtchn_to_irq_all();
1581 for_each_possible_cpu(cpu) {
1582 restore_cpu_virqs(cpu);
1583 restore_cpu_ipis(cpu);
1589 static struct irq_chip xen_dynamic_chip __read_mostly = {
1592 .irq_disable = disable_dynirq,
1593 .irq_mask = disable_dynirq,
1594 .irq_unmask = enable_dynirq,
1596 .irq_ack = ack_dynirq,
1597 .irq_mask_ack = mask_ack_dynirq,
1599 .irq_set_affinity = set_affinity_irq,
1600 .irq_retrigger = retrigger_dynirq,
1603 static struct irq_chip xen_pirq_chip __read_mostly = {
1606 .irq_startup = startup_pirq,
1607 .irq_shutdown = shutdown_pirq,
1608 .irq_enable = enable_pirq,
1609 .irq_disable = disable_pirq,
1611 .irq_mask = disable_dynirq,
1612 .irq_unmask = enable_dynirq,
1614 .irq_ack = eoi_pirq,
1615 .irq_eoi = eoi_pirq,
1616 .irq_mask_ack = mask_ack_pirq,
1618 .irq_set_affinity = set_affinity_irq,
1620 .irq_retrigger = retrigger_dynirq,
1623 static struct irq_chip xen_percpu_chip __read_mostly = {
1624 .name = "xen-percpu",
1626 .irq_disable = disable_dynirq,
1627 .irq_mask = disable_dynirq,
1628 .irq_unmask = enable_dynirq,
1630 .irq_ack = ack_dynirq,
1633 int xen_set_callback_via(uint64_t via)
1635 struct xen_hvm_param a;
1636 a.domid = DOMID_SELF;
1637 a.index = HVM_PARAM_CALLBACK_IRQ;
1639 return HYPERVISOR_hvm_op(HVMOP_set_param, &a);
1641 EXPORT_SYMBOL_GPL(xen_set_callback_via);
1643 #ifdef CONFIG_XEN_PVHVM
1644 /* Vector callbacks are better than PCI interrupts to receive event
1645 * channel notifications because we can receive vector callbacks on any
1646 * vcpu and we don't need PCI support or APIC interactions. */
1647 void xen_callback_vector(void)
1650 uint64_t callback_via;
1651 if (xen_have_vector_callback) {
1652 callback_via = HVM_CALLBACK_VECTOR(HYPERVISOR_CALLBACK_VECTOR);
1653 rc = xen_set_callback_via(callback_via);
1655 pr_err("Request for Xen HVM callback vector failed\n");
1656 xen_have_vector_callback = 0;
1659 pr_info("Xen HVM callback vector for event delivery is enabled\n");
1660 /* in the restore case the vector has already been allocated */
1661 if (!test_bit(HYPERVISOR_CALLBACK_VECTOR, used_vectors))
1662 alloc_intr_gate(HYPERVISOR_CALLBACK_VECTOR,
1663 xen_hvm_callback_vector);
1667 void xen_callback_vector(void) {}
1670 #undef MODULE_PARAM_PREFIX
1671 #define MODULE_PARAM_PREFIX "xen."
1673 static bool fifo_events = true;
1674 module_param(fifo_events, bool, 0);
1676 void __init xen_init_IRQ(void)
1681 ret = xen_evtchn_fifo_init();
1683 xen_evtchn_2l_init();
1685 evtchn_to_irq = kcalloc(EVTCHN_ROW(xen_evtchn_max_channels()),
1686 sizeof(*evtchn_to_irq), GFP_KERNEL);
1687 BUG_ON(!evtchn_to_irq);
1689 /* No event channels are 'live' right now. */
1690 xen_evtchn_mask_all();
1692 pirq_needs_eoi = pirq_needs_eoi_flag;
1695 if (xen_pv_domain()) {
1696 irq_ctx_init(smp_processor_id());
1697 if (xen_initial_domain())
1698 pci_xen_initial_domain();
1700 if (xen_feature(XENFEAT_hvm_callback_vector))
1701 xen_callback_vector();
1703 if (xen_hvm_domain()) {
1705 /* pci_xen_hvm_init must be called after native_init_IRQ so that
1706 * __acpi_register_gsi can point at the right function */
1710 struct physdev_pirq_eoi_gmfn eoi_gmfn;
1712 pirq_eoi_map = (void *)__get_free_page(GFP_KERNEL|__GFP_ZERO);
1713 eoi_gmfn.gmfn = virt_to_mfn(pirq_eoi_map);
1714 rc = HYPERVISOR_physdev_op(PHYSDEVOP_pirq_eoi_gmfn_v2, &eoi_gmfn);
1715 /* TODO: No PVH support for PIRQ EOI */
1717 free_page((unsigned long) pirq_eoi_map);
1718 pirq_eoi_map = NULL;
1720 pirq_needs_eoi = pirq_check_eoi_map;