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_irq_dynamic(void)
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_desc_from(first, -1);
419 static int __must_check xen_allocate_irq_gsi(unsigned gsi)
424 * A PV guest has no concept of a GSI (since it has no ACPI
425 * nor access to/knowledge of the physical APICs). Therefore
426 * all IRQs are dynamically allocated from the entire IRQ
429 if (xen_pv_domain() && !xen_initial_domain())
430 return xen_allocate_irq_dynamic();
432 /* Legacy IRQ descriptors are already allocated by the arch. */
433 if (gsi < NR_IRQS_LEGACY)
436 irq = irq_alloc_desc_at(gsi, -1);
443 static void xen_free_irq(unsigned irq)
445 struct irq_info *info = irq_get_handler_data(irq);
450 list_del(&info->list);
452 irq_set_handler_data(irq, NULL);
454 WARN_ON(info->refcnt > 0);
458 /* Legacy IRQ descriptors are managed by the arch. */
459 if (irq < NR_IRQS_LEGACY)
465 static void xen_evtchn_close(unsigned int port)
467 struct evtchn_close close;
470 if (HYPERVISOR_event_channel_op(EVTCHNOP_close, &close) != 0)
473 /* Closed ports are implicitly re-bound to VCPU0. */
474 bind_evtchn_to_cpu(port, 0);
477 static void pirq_query_unmask(int irq)
479 struct physdev_irq_status_query irq_status;
480 struct irq_info *info = info_for_irq(irq);
482 BUG_ON(info->type != IRQT_PIRQ);
484 irq_status.irq = pirq_from_irq(irq);
485 if (HYPERVISOR_physdev_op(PHYSDEVOP_irq_status_query, &irq_status))
486 irq_status.flags = 0;
488 info->u.pirq.flags &= ~PIRQ_NEEDS_EOI;
489 if (irq_status.flags & XENIRQSTAT_needs_eoi)
490 info->u.pirq.flags |= PIRQ_NEEDS_EOI;
493 static bool probing_irq(int irq)
495 struct irq_desc *desc = irq_to_desc(irq);
497 return desc && desc->action == NULL;
500 static void eoi_pirq(struct irq_data *data)
502 int evtchn = evtchn_from_irq(data->irq);
503 struct physdev_eoi eoi = { .irq = pirq_from_irq(data->irq) };
508 if (VALID_EVTCHN(evtchn))
509 clear_evtchn(evtchn);
511 if (pirq_needs_eoi(data->irq)) {
512 rc = HYPERVISOR_physdev_op(PHYSDEVOP_eoi, &eoi);
517 static void mask_ack_pirq(struct irq_data *data)
519 disable_dynirq(data);
523 static unsigned int __startup_pirq(unsigned int irq)
525 struct evtchn_bind_pirq bind_pirq;
526 struct irq_info *info = info_for_irq(irq);
527 int evtchn = evtchn_from_irq(irq);
530 BUG_ON(info->type != IRQT_PIRQ);
532 if (VALID_EVTCHN(evtchn))
535 bind_pirq.pirq = pirq_from_irq(irq);
536 /* NB. We are happy to share unless we are probing. */
537 bind_pirq.flags = info->u.pirq.flags & PIRQ_SHAREABLE ?
538 BIND_PIRQ__WILL_SHARE : 0;
539 rc = HYPERVISOR_event_channel_op(EVTCHNOP_bind_pirq, &bind_pirq);
541 if (!probing_irq(irq))
542 pr_info("Failed to obtain physical IRQ %d\n", irq);
545 evtchn = bind_pirq.port;
547 pirq_query_unmask(irq);
549 rc = set_evtchn_to_irq(evtchn, irq);
551 pr_err("irq%d: Failed to set port to irq mapping (%d)\n",
553 xen_evtchn_close(evtchn);
556 bind_evtchn_to_cpu(evtchn, 0);
557 info->evtchn = evtchn;
560 unmask_evtchn(evtchn);
561 eoi_pirq(irq_get_irq_data(irq));
566 static unsigned int startup_pirq(struct irq_data *data)
568 return __startup_pirq(data->irq);
571 static void shutdown_pirq(struct irq_data *data)
573 unsigned int irq = data->irq;
574 struct irq_info *info = info_for_irq(irq);
575 unsigned evtchn = evtchn_from_irq(irq);
577 BUG_ON(info->type != IRQT_PIRQ);
579 if (!VALID_EVTCHN(evtchn))
583 xen_evtchn_close(evtchn);
584 xen_irq_info_cleanup(info);
587 static void enable_pirq(struct irq_data *data)
592 static void disable_pirq(struct irq_data *data)
594 disable_dynirq(data);
597 int xen_irq_from_gsi(unsigned gsi)
599 struct irq_info *info;
601 list_for_each_entry(info, &xen_irq_list_head, list) {
602 if (info->type != IRQT_PIRQ)
605 if (info->u.pirq.gsi == gsi)
611 EXPORT_SYMBOL_GPL(xen_irq_from_gsi);
613 static void __unbind_from_irq(unsigned int irq)
615 int evtchn = evtchn_from_irq(irq);
616 struct irq_info *info = irq_get_handler_data(irq);
618 if (info->refcnt > 0) {
620 if (info->refcnt != 0)
624 if (VALID_EVTCHN(evtchn)) {
625 unsigned int cpu = cpu_from_irq(irq);
627 xen_evtchn_close(evtchn);
629 switch (type_from_irq(irq)) {
631 per_cpu(virq_to_irq, cpu)[virq_from_irq(irq)] = -1;
634 per_cpu(ipi_to_irq, cpu)[ipi_from_irq(irq)] = -1;
640 xen_irq_info_cleanup(info);
643 BUG_ON(info_for_irq(irq)->type == IRQT_UNBOUND);
649 * Do not make any assumptions regarding the relationship between the
650 * IRQ number returned here and the Xen pirq argument.
652 * Note: We don't assign an event channel until the irq actually started
653 * up. Return an existing irq if we've already got one for the gsi.
655 * Shareable implies level triggered, not shareable implies edge
658 int xen_bind_pirq_gsi_to_irq(unsigned gsi,
659 unsigned pirq, int shareable, char *name)
662 struct physdev_irq irq_op;
665 mutex_lock(&irq_mapping_update_lock);
667 irq = xen_irq_from_gsi(gsi);
669 pr_info("%s: returning irq %d for gsi %u\n",
674 irq = xen_allocate_irq_gsi(gsi);
681 /* Only the privileged domain can do this. For non-priv, the pcifront
682 * driver provides a PCI bus that does the call to do exactly
683 * this in the priv domain. */
684 if (xen_initial_domain() &&
685 HYPERVISOR_physdev_op(PHYSDEVOP_alloc_irq_vector, &irq_op)) {
691 ret = xen_irq_info_pirq_setup(irq, 0, pirq, gsi, DOMID_SELF,
692 shareable ? PIRQ_SHAREABLE : 0);
694 __unbind_from_irq(irq);
699 pirq_query_unmask(irq);
700 /* We try to use the handler with the appropriate semantic for the
701 * type of interrupt: if the interrupt is an edge triggered
702 * interrupt we use handle_edge_irq.
704 * On the other hand if the interrupt is level triggered we use
705 * handle_fasteoi_irq like the native code does for this kind of
708 * Depending on the Xen version, pirq_needs_eoi might return true
709 * not only for level triggered interrupts but for edge triggered
710 * interrupts too. In any case Xen always honors the eoi mechanism,
711 * not injecting any more pirqs of the same kind if the first one
712 * hasn't received an eoi yet. Therefore using the fasteoi handler
713 * is the right choice either way.
716 irq_set_chip_and_handler_name(irq, &xen_pirq_chip,
717 handle_fasteoi_irq, name);
719 irq_set_chip_and_handler_name(irq, &xen_pirq_chip,
720 handle_edge_irq, name);
723 mutex_unlock(&irq_mapping_update_lock);
728 #ifdef CONFIG_PCI_MSI
729 int xen_allocate_pirq_msi(struct pci_dev *dev, struct msi_desc *msidesc)
732 struct physdev_get_free_pirq op_get_free_pirq;
734 op_get_free_pirq.type = MAP_PIRQ_TYPE_MSI;
735 rc = HYPERVISOR_physdev_op(PHYSDEVOP_get_free_pirq, &op_get_free_pirq);
737 WARN_ONCE(rc == -ENOSYS,
738 "hypervisor does not support the PHYSDEVOP_get_free_pirq interface\n");
740 return rc ? -1 : op_get_free_pirq.pirq;
743 int xen_bind_pirq_msi_to_irq(struct pci_dev *dev, struct msi_desc *msidesc,
744 int pirq, const char *name, domid_t domid)
748 mutex_lock(&irq_mapping_update_lock);
750 irq = xen_allocate_irq_dynamic();
754 irq_set_chip_and_handler_name(irq, &xen_pirq_chip, handle_edge_irq,
757 ret = xen_irq_info_pirq_setup(irq, 0, pirq, 0, domid, 0);
760 ret = irq_set_msi_desc(irq, msidesc);
764 mutex_unlock(&irq_mapping_update_lock);
767 __unbind_from_irq(irq);
768 mutex_unlock(&irq_mapping_update_lock);
773 int xen_destroy_irq(int irq)
775 struct irq_desc *desc;
776 struct physdev_unmap_pirq unmap_irq;
777 struct irq_info *info = info_for_irq(irq);
780 mutex_lock(&irq_mapping_update_lock);
782 desc = irq_to_desc(irq);
786 if (xen_initial_domain()) {
787 unmap_irq.pirq = info->u.pirq.pirq;
788 unmap_irq.domid = info->u.pirq.domid;
789 rc = HYPERVISOR_physdev_op(PHYSDEVOP_unmap_pirq, &unmap_irq);
790 /* If another domain quits without making the pci_disable_msix
791 * call, the Xen hypervisor takes care of freeing the PIRQs
792 * (free_domain_pirqs).
794 if ((rc == -ESRCH && info->u.pirq.domid != DOMID_SELF))
795 pr_info("domain %d does not have %d anymore\n",
796 info->u.pirq.domid, info->u.pirq.pirq);
798 pr_warn("unmap irq failed %d\n", rc);
806 mutex_unlock(&irq_mapping_update_lock);
810 int xen_irq_from_pirq(unsigned pirq)
814 struct irq_info *info;
816 mutex_lock(&irq_mapping_update_lock);
818 list_for_each_entry(info, &xen_irq_list_head, list) {
819 if (info->type != IRQT_PIRQ)
822 if (info->u.pirq.pirq == pirq)
827 mutex_unlock(&irq_mapping_update_lock);
833 int xen_pirq_from_irq(unsigned irq)
835 return pirq_from_irq(irq);
837 EXPORT_SYMBOL_GPL(xen_pirq_from_irq);
839 int bind_evtchn_to_irq(unsigned int evtchn)
844 if (evtchn >= xen_evtchn_max_channels())
847 mutex_lock(&irq_mapping_update_lock);
849 irq = get_evtchn_to_irq(evtchn);
852 irq = xen_allocate_irq_dynamic();
856 irq_set_chip_and_handler_name(irq, &xen_dynamic_chip,
857 handle_edge_irq, "event");
859 ret = xen_irq_info_evtchn_setup(irq, evtchn);
861 __unbind_from_irq(irq);
865 /* New interdomain events are bound to VCPU 0. */
866 bind_evtchn_to_cpu(evtchn, 0);
868 struct irq_info *info = info_for_irq(irq);
869 WARN_ON(info == NULL || info->type != IRQT_EVTCHN);
873 mutex_unlock(&irq_mapping_update_lock);
877 EXPORT_SYMBOL_GPL(bind_evtchn_to_irq);
879 static int bind_ipi_to_irq(unsigned int ipi, unsigned int cpu)
881 struct evtchn_bind_ipi bind_ipi;
885 mutex_lock(&irq_mapping_update_lock);
887 irq = per_cpu(ipi_to_irq, cpu)[ipi];
890 irq = xen_allocate_irq_dynamic();
894 irq_set_chip_and_handler_name(irq, &xen_percpu_chip,
895 handle_percpu_irq, "ipi");
898 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_ipi,
901 evtchn = bind_ipi.port;
903 ret = xen_irq_info_ipi_setup(cpu, irq, evtchn, ipi);
905 __unbind_from_irq(irq);
909 bind_evtchn_to_cpu(evtchn, cpu);
911 struct irq_info *info = info_for_irq(irq);
912 WARN_ON(info == NULL || info->type != IRQT_IPI);
916 mutex_unlock(&irq_mapping_update_lock);
920 static int bind_interdomain_evtchn_to_irq(unsigned int remote_domain,
921 unsigned int remote_port)
923 struct evtchn_bind_interdomain bind_interdomain;
926 bind_interdomain.remote_dom = remote_domain;
927 bind_interdomain.remote_port = remote_port;
929 err = HYPERVISOR_event_channel_op(EVTCHNOP_bind_interdomain,
932 return err ? : bind_evtchn_to_irq(bind_interdomain.local_port);
935 static int find_virq(unsigned int virq, unsigned int cpu)
937 struct evtchn_status status;
938 int port, rc = -ENOENT;
940 memset(&status, 0, sizeof(status));
941 for (port = 0; port < xen_evtchn_max_channels(); port++) {
942 status.dom = DOMID_SELF;
944 rc = HYPERVISOR_event_channel_op(EVTCHNOP_status, &status);
947 if (status.status != EVTCHNSTAT_virq)
949 if (status.u.virq == virq && status.vcpu == cpu) {
958 * xen_evtchn_nr_channels - number of usable event channel ports
960 * This may be less than the maximum supported by the current
961 * hypervisor ABI. Use xen_evtchn_max_channels() for the maximum
964 unsigned xen_evtchn_nr_channels(void)
966 return evtchn_ops->nr_channels();
968 EXPORT_SYMBOL_GPL(xen_evtchn_nr_channels);
970 int bind_virq_to_irq(unsigned int virq, unsigned int cpu)
972 struct evtchn_bind_virq bind_virq;
973 int evtchn, irq, ret;
975 mutex_lock(&irq_mapping_update_lock);
977 irq = per_cpu(virq_to_irq, cpu)[virq];
980 irq = xen_allocate_irq_dynamic();
984 irq_set_chip_and_handler_name(irq, &xen_percpu_chip,
985 handle_percpu_irq, "virq");
987 bind_virq.virq = virq;
988 bind_virq.vcpu = cpu;
989 ret = HYPERVISOR_event_channel_op(EVTCHNOP_bind_virq,
992 evtchn = bind_virq.port;
995 ret = find_virq(virq, cpu);
1000 ret = xen_irq_info_virq_setup(cpu, irq, evtchn, virq);
1002 __unbind_from_irq(irq);
1007 bind_evtchn_to_cpu(evtchn, cpu);
1009 struct irq_info *info = info_for_irq(irq);
1010 WARN_ON(info == NULL || info->type != IRQT_VIRQ);
1014 mutex_unlock(&irq_mapping_update_lock);
1019 static void unbind_from_irq(unsigned int irq)
1021 mutex_lock(&irq_mapping_update_lock);
1022 __unbind_from_irq(irq);
1023 mutex_unlock(&irq_mapping_update_lock);
1026 int bind_evtchn_to_irqhandler(unsigned int evtchn,
1027 irq_handler_t handler,
1028 unsigned long irqflags,
1029 const char *devname, void *dev_id)
1033 irq = bind_evtchn_to_irq(evtchn);
1036 retval = request_irq(irq, handler, irqflags, devname, dev_id);
1038 unbind_from_irq(irq);
1044 EXPORT_SYMBOL_GPL(bind_evtchn_to_irqhandler);
1046 int bind_interdomain_evtchn_to_irqhandler(unsigned int remote_domain,
1047 unsigned int remote_port,
1048 irq_handler_t handler,
1049 unsigned long irqflags,
1050 const char *devname,
1055 irq = bind_interdomain_evtchn_to_irq(remote_domain, remote_port);
1059 retval = request_irq(irq, handler, irqflags, devname, dev_id);
1061 unbind_from_irq(irq);
1067 EXPORT_SYMBOL_GPL(bind_interdomain_evtchn_to_irqhandler);
1069 int bind_virq_to_irqhandler(unsigned int virq, unsigned int cpu,
1070 irq_handler_t handler,
1071 unsigned long irqflags, const char *devname, void *dev_id)
1075 irq = bind_virq_to_irq(virq, cpu);
1078 retval = request_irq(irq, handler, irqflags, devname, dev_id);
1080 unbind_from_irq(irq);
1086 EXPORT_SYMBOL_GPL(bind_virq_to_irqhandler);
1088 int bind_ipi_to_irqhandler(enum ipi_vector ipi,
1090 irq_handler_t handler,
1091 unsigned long irqflags,
1092 const char *devname,
1097 irq = bind_ipi_to_irq(ipi, cpu);
1101 irqflags |= IRQF_NO_SUSPEND | IRQF_FORCE_RESUME | IRQF_EARLY_RESUME;
1102 retval = request_irq(irq, handler, irqflags, devname, dev_id);
1104 unbind_from_irq(irq);
1111 void unbind_from_irqhandler(unsigned int irq, void *dev_id)
1113 struct irq_info *info = irq_get_handler_data(irq);
1117 free_irq(irq, dev_id);
1118 unbind_from_irq(irq);
1120 EXPORT_SYMBOL_GPL(unbind_from_irqhandler);
1123 * xen_set_irq_priority() - set an event channel priority.
1124 * @irq:irq bound to an event channel.
1125 * @priority: priority between XEN_IRQ_PRIORITY_MAX and XEN_IRQ_PRIORITY_MIN.
1127 int xen_set_irq_priority(unsigned irq, unsigned priority)
1129 struct evtchn_set_priority set_priority;
1131 set_priority.port = evtchn_from_irq(irq);
1132 set_priority.priority = priority;
1134 return HYPERVISOR_event_channel_op(EVTCHNOP_set_priority,
1137 EXPORT_SYMBOL_GPL(xen_set_irq_priority);
1139 int evtchn_make_refcounted(unsigned int evtchn)
1141 int irq = get_evtchn_to_irq(evtchn);
1142 struct irq_info *info;
1147 info = irq_get_handler_data(irq);
1152 WARN_ON(info->refcnt != -1);
1158 EXPORT_SYMBOL_GPL(evtchn_make_refcounted);
1160 int evtchn_get(unsigned int evtchn)
1163 struct irq_info *info;
1166 if (evtchn >= xen_evtchn_max_channels())
1169 mutex_lock(&irq_mapping_update_lock);
1171 irq = get_evtchn_to_irq(evtchn);
1175 info = irq_get_handler_data(irq);
1181 if (info->refcnt <= 0)
1187 mutex_unlock(&irq_mapping_update_lock);
1191 EXPORT_SYMBOL_GPL(evtchn_get);
1193 void evtchn_put(unsigned int evtchn)
1195 int irq = get_evtchn_to_irq(evtchn);
1196 if (WARN_ON(irq == -1))
1198 unbind_from_irq(irq);
1200 EXPORT_SYMBOL_GPL(evtchn_put);
1202 void xen_send_IPI_one(unsigned int cpu, enum ipi_vector vector)
1207 if (unlikely(vector == XEN_NMI_VECTOR)) {
1208 int rc = HYPERVISOR_vcpu_op(VCPUOP_send_nmi, cpu, NULL);
1210 printk(KERN_WARNING "Sending nmi to CPU%d failed (rc:%d)\n", cpu, rc);
1214 irq = per_cpu(ipi_to_irq, cpu)[vector];
1216 notify_remote_via_irq(irq);
1219 static DEFINE_PER_CPU(unsigned, xed_nesting_count);
1221 static void __xen_evtchn_do_upcall(void)
1223 struct vcpu_info *vcpu_info = __this_cpu_read(xen_vcpu);
1224 int cpu = get_cpu();
1228 vcpu_info->evtchn_upcall_pending = 0;
1230 if (__this_cpu_inc_return(xed_nesting_count) - 1)
1233 xen_evtchn_handle_events(cpu);
1235 BUG_ON(!irqs_disabled());
1237 count = __this_cpu_read(xed_nesting_count);
1238 __this_cpu_write(xed_nesting_count, 0);
1239 } while (count != 1 || vcpu_info->evtchn_upcall_pending);
1246 void xen_evtchn_do_upcall(struct pt_regs *regs)
1248 struct pt_regs *old_regs = set_irq_regs(regs);
1255 __xen_evtchn_do_upcall();
1258 set_irq_regs(old_regs);
1261 void xen_hvm_evtchn_do_upcall(void)
1263 __xen_evtchn_do_upcall();
1265 EXPORT_SYMBOL_GPL(xen_hvm_evtchn_do_upcall);
1267 /* Rebind a new event channel to an existing irq. */
1268 void rebind_evtchn_irq(int evtchn, int irq)
1270 struct irq_info *info = info_for_irq(irq);
1275 /* Make sure the irq is masked, since the new event channel
1276 will also be masked. */
1279 mutex_lock(&irq_mapping_update_lock);
1281 /* After resume the irq<->evtchn mappings are all cleared out */
1282 BUG_ON(get_evtchn_to_irq(evtchn) != -1);
1283 /* Expect irq to have been bound before,
1284 so there should be a proper type */
1285 BUG_ON(info->type == IRQT_UNBOUND);
1287 (void)xen_irq_info_evtchn_setup(irq, evtchn);
1289 mutex_unlock(&irq_mapping_update_lock);
1291 /* new event channels are always bound to cpu 0 */
1292 irq_set_affinity(irq, cpumask_of(0));
1294 /* Unmask the event channel. */
1298 /* Rebind an evtchn so that it gets delivered to a specific cpu */
1299 static int rebind_irq_to_cpu(unsigned irq, unsigned tcpu)
1301 struct evtchn_bind_vcpu bind_vcpu;
1302 int evtchn = evtchn_from_irq(irq);
1305 if (!VALID_EVTCHN(evtchn))
1309 * Events delivered via platform PCI interrupts are always
1310 * routed to vcpu 0 and hence cannot be rebound.
1312 if (xen_hvm_domain() && !xen_have_vector_callback)
1315 /* Send future instances of this interrupt to other vcpu. */
1316 bind_vcpu.port = evtchn;
1317 bind_vcpu.vcpu = tcpu;
1320 * Mask the event while changing the VCPU binding to prevent
1321 * it being delivered on an unexpected VCPU.
1323 masked = test_and_set_mask(evtchn);
1326 * If this fails, it usually just indicates that we're dealing with a
1327 * virq or IPI channel, which don't actually need to be rebound. Ignore
1328 * it, but don't do the xenlinux-level rebind in that case.
1330 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_vcpu, &bind_vcpu) >= 0)
1331 bind_evtchn_to_cpu(evtchn, tcpu);
1334 unmask_evtchn(evtchn);
1339 static int set_affinity_irq(struct irq_data *data, const struct cpumask *dest,
1342 unsigned tcpu = cpumask_first(dest);
1344 return rebind_irq_to_cpu(data->irq, tcpu);
1347 static int retrigger_evtchn(int evtchn)
1351 if (!VALID_EVTCHN(evtchn))
1354 masked = test_and_set_mask(evtchn);
1357 unmask_evtchn(evtchn);
1362 int resend_irq_on_evtchn(unsigned int irq)
1364 return retrigger_evtchn(evtchn_from_irq(irq));
1367 static void enable_dynirq(struct irq_data *data)
1369 int evtchn = evtchn_from_irq(data->irq);
1371 if (VALID_EVTCHN(evtchn))
1372 unmask_evtchn(evtchn);
1375 static void disable_dynirq(struct irq_data *data)
1377 int evtchn = evtchn_from_irq(data->irq);
1379 if (VALID_EVTCHN(evtchn))
1380 mask_evtchn(evtchn);
1383 static void ack_dynirq(struct irq_data *data)
1385 int evtchn = evtchn_from_irq(data->irq);
1389 if (VALID_EVTCHN(evtchn))
1390 clear_evtchn(evtchn);
1393 static void mask_ack_dynirq(struct irq_data *data)
1395 disable_dynirq(data);
1399 static int retrigger_dynirq(struct irq_data *data)
1401 return retrigger_evtchn(evtchn_from_irq(data->irq));
1404 static void restore_pirqs(void)
1406 int pirq, rc, irq, gsi;
1407 struct physdev_map_pirq map_irq;
1408 struct irq_info *info;
1410 list_for_each_entry(info, &xen_irq_list_head, list) {
1411 if (info->type != IRQT_PIRQ)
1414 pirq = info->u.pirq.pirq;
1415 gsi = info->u.pirq.gsi;
1418 /* save/restore of PT devices doesn't work, so at this point the
1419 * only devices present are GSI based emulated devices */
1423 map_irq.domid = DOMID_SELF;
1424 map_irq.type = MAP_PIRQ_TYPE_GSI;
1425 map_irq.index = gsi;
1426 map_irq.pirq = pirq;
1428 rc = HYPERVISOR_physdev_op(PHYSDEVOP_map_pirq, &map_irq);
1430 pr_warn("xen map irq failed gsi=%d irq=%d pirq=%d rc=%d\n",
1431 gsi, irq, pirq, rc);
1436 printk(KERN_DEBUG "xen: --> irq=%d, pirq=%d\n", irq, map_irq.pirq);
1438 __startup_pirq(irq);
1442 static void restore_cpu_virqs(unsigned int cpu)
1444 struct evtchn_bind_virq bind_virq;
1445 int virq, irq, evtchn;
1447 for (virq = 0; virq < NR_VIRQS; virq++) {
1448 if ((irq = per_cpu(virq_to_irq, cpu)[virq]) == -1)
1451 BUG_ON(virq_from_irq(irq) != virq);
1453 /* Get a new binding from Xen. */
1454 bind_virq.virq = virq;
1455 bind_virq.vcpu = cpu;
1456 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_virq,
1459 evtchn = bind_virq.port;
1461 /* Record the new mapping. */
1462 (void)xen_irq_info_virq_setup(cpu, irq, evtchn, virq);
1463 bind_evtchn_to_cpu(evtchn, cpu);
1467 static void restore_cpu_ipis(unsigned int cpu)
1469 struct evtchn_bind_ipi bind_ipi;
1470 int ipi, irq, evtchn;
1472 for (ipi = 0; ipi < XEN_NR_IPIS; ipi++) {
1473 if ((irq = per_cpu(ipi_to_irq, cpu)[ipi]) == -1)
1476 BUG_ON(ipi_from_irq(irq) != ipi);
1478 /* Get a new binding from Xen. */
1479 bind_ipi.vcpu = cpu;
1480 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_ipi,
1483 evtchn = bind_ipi.port;
1485 /* Record the new mapping. */
1486 (void)xen_irq_info_ipi_setup(cpu, irq, evtchn, ipi);
1487 bind_evtchn_to_cpu(evtchn, cpu);
1491 /* Clear an irq's pending state, in preparation for polling on it */
1492 void xen_clear_irq_pending(int irq)
1494 int evtchn = evtchn_from_irq(irq);
1496 if (VALID_EVTCHN(evtchn))
1497 clear_evtchn(evtchn);
1499 EXPORT_SYMBOL(xen_clear_irq_pending);
1500 void xen_set_irq_pending(int irq)
1502 int evtchn = evtchn_from_irq(irq);
1504 if (VALID_EVTCHN(evtchn))
1508 bool xen_test_irq_pending(int irq)
1510 int evtchn = evtchn_from_irq(irq);
1513 if (VALID_EVTCHN(evtchn))
1514 ret = test_evtchn(evtchn);
1519 /* Poll waiting for an irq to become pending with timeout. In the usual case,
1520 * the irq will be disabled so it won't deliver an interrupt. */
1521 void xen_poll_irq_timeout(int irq, u64 timeout)
1523 evtchn_port_t evtchn = evtchn_from_irq(irq);
1525 if (VALID_EVTCHN(evtchn)) {
1526 struct sched_poll poll;
1529 poll.timeout = timeout;
1530 set_xen_guest_handle(poll.ports, &evtchn);
1532 if (HYPERVISOR_sched_op(SCHEDOP_poll, &poll) != 0)
1536 EXPORT_SYMBOL(xen_poll_irq_timeout);
1537 /* Poll waiting for an irq to become pending. In the usual case, the
1538 * irq will be disabled so it won't deliver an interrupt. */
1539 void xen_poll_irq(int irq)
1541 xen_poll_irq_timeout(irq, 0 /* no timeout */);
1544 /* Check whether the IRQ line is shared with other guests. */
1545 int xen_test_irq_shared(int irq)
1547 struct irq_info *info = info_for_irq(irq);
1548 struct physdev_irq_status_query irq_status;
1553 irq_status.irq = info->u.pirq.pirq;
1555 if (HYPERVISOR_physdev_op(PHYSDEVOP_irq_status_query, &irq_status))
1557 return !(irq_status.flags & XENIRQSTAT_shared);
1559 EXPORT_SYMBOL_GPL(xen_test_irq_shared);
1561 void xen_irq_resume(void)
1564 struct irq_info *info;
1566 /* New event-channel space is not 'live' yet. */
1567 xen_evtchn_mask_all();
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;
1646 if (xen_have_vector_callback) {
1647 callback_via = HVM_CALLBACK_VECTOR(HYPERVISOR_CALLBACK_VECTOR);
1648 rc = xen_set_callback_via(callback_via);
1650 pr_err("Request for Xen HVM callback vector failed\n");
1651 xen_have_vector_callback = 0;
1654 pr_info("Xen HVM callback vector for event delivery is enabled\n");
1655 /* in the restore case the vector has already been allocated */
1656 if (!test_bit(HYPERVISOR_CALLBACK_VECTOR, used_vectors))
1657 alloc_intr_gate(HYPERVISOR_CALLBACK_VECTOR,
1658 xen_hvm_callback_vector);
1662 void xen_callback_vector(void) {}
1665 #undef MODULE_PARAM_PREFIX
1666 #define MODULE_PARAM_PREFIX "xen."
1668 static bool fifo_events = true;
1669 module_param(fifo_events, bool, 0);
1671 void __init xen_init_IRQ(void)
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 xen_evtchn_mask_all();
1687 pirq_needs_eoi = pirq_needs_eoi_flag;
1690 if (xen_pv_domain()) {
1691 irq_ctx_init(smp_processor_id());
1692 if (xen_initial_domain())
1693 pci_xen_initial_domain();
1695 if (xen_feature(XENFEAT_hvm_callback_vector))
1696 xen_callback_vector();
1698 if (xen_hvm_domain()) {
1700 /* pci_xen_hvm_init must be called after native_init_IRQ so that
1701 * __acpi_register_gsi can point at the right function */
1705 struct physdev_pirq_eoi_gmfn eoi_gmfn;
1707 pirq_eoi_map = (void *)__get_free_page(GFP_KERNEL|__GFP_ZERO);
1708 eoi_gmfn.gmfn = virt_to_mfn(pirq_eoi_map);
1709 rc = HYPERVISOR_physdev_op(PHYSDEVOP_pirq_eoi_gmfn_v2, &eoi_gmfn);
1710 /* TODO: No PVH support for PIRQ EOI */
1712 free_page((unsigned long) pirq_eoi_map);
1713 pirq_eoi_map = NULL;
1715 pirq_needs_eoi = pirq_check_eoi_map;