1 // SPDX-License-Identifier: GPL-2.0-only
5 * Xen models interrupts with abstract event channels. Because each
6 * domain gets 1024 event channels, but NR_IRQ is not that large, we
7 * must dynamically map irqs<->event channels. The event channels
8 * interface with the rest of the kernel by defining a xen interrupt
9 * chip. When an event is received, it is mapped to an irq and sent
10 * through the normal interrupt processing path.
12 * There are four kinds of events which can be mapped to an event
15 * 1. Inter-domain notifications. This includes all the virtual
16 * device events, since they're driven by front-ends in another domain
18 * 2. VIRQs, typically used for timers. These are per-cpu events.
20 * 4. PIRQs - Hardware interrupts.
22 * Jeremy Fitzhardinge <jeremy@xensource.com>, XenSource Inc, 2007
25 #define pr_fmt(fmt) "xen:" KBUILD_MODNAME ": " fmt
27 #include <linux/linkage.h>
28 #include <linux/interrupt.h>
29 #include <linux/irq.h>
30 #include <linux/moduleparam.h>
31 #include <linux/string.h>
32 #include <linux/memblock.h>
33 #include <linux/slab.h>
34 #include <linux/irqnr.h>
35 #include <linux/pci.h>
36 #include <linux/spinlock.h>
37 #include <linux/cpuhotplug.h>
38 #include <linux/atomic.h>
39 #include <linux/ktime.h>
43 #include <asm/ptrace.h>
44 #include <asm/idtentry.h>
46 #include <asm/io_apic.h>
47 #include <asm/i8259.h>
48 #include <asm/xen/pci.h>
50 #include <asm/sync_bitops.h>
51 #include <asm/xen/hypercall.h>
52 #include <asm/xen/hypervisor.h>
57 #include <xen/xen-ops.h>
58 #include <xen/events.h>
59 #include <xen/interface/xen.h>
60 #include <xen/interface/event_channel.h>
61 #include <xen/interface/hvm/hvm_op.h>
62 #include <xen/interface/hvm/params.h>
63 #include <xen/interface/physdev.h>
64 #include <xen/interface/sched.h>
65 #include <xen/interface/vcpu.h>
66 #include <asm/hw_irq.h>
68 #include "events_internal.h"
70 #undef MODULE_PARAM_PREFIX
71 #define MODULE_PARAM_PREFIX "xen."
73 /* Interrupt types. */
83 * Packed IRQ information:
84 * type - enum xen_irq_type
85 * event channel - irq->event channel mapping
86 * cpu - cpu this event channel is bound to
87 * index - type-specific information:
88 * PIRQ - vector, with MSB being "needs EIO", or physical IRQ of the HVM
89 * guest, or GSI (real passthrough IRQ) of the device.
95 struct list_head list;
96 struct list_head eoi_list;
100 enum xen_irq_type type; /* type */
102 evtchn_port_t evtchn; /* event channel */
103 unsigned short cpu; /* cpu bound */
104 unsigned short eoi_cpu; /* EOI must happen on this cpu-1 */
105 unsigned int irq_epoch; /* If eoi_cpu valid: irq_epoch of event */
106 u64 eoi_time; /* Time in jiffies when to EOI. */
114 unsigned char vector;
121 #define PIRQ_NEEDS_EOI (1 << 0)
122 #define PIRQ_SHAREABLE (1 << 1)
123 #define PIRQ_MSI_GROUP (1 << 2)
125 static uint __read_mostly event_loop_timeout = 2;
126 module_param(event_loop_timeout, uint, 0644);
128 static uint __read_mostly event_eoi_delay = 10;
129 module_param(event_eoi_delay, uint, 0644);
131 const struct evtchn_ops *evtchn_ops;
134 * This lock protects updates to the following mapping and reference-count
135 * arrays. The lock does not need to be acquired to read the mapping tables.
137 static DEFINE_MUTEX(irq_mapping_update_lock);
140 * Lock protecting event handling loop against removing event channels.
141 * Adding of event channels is no issue as the associated IRQ becomes active
142 * only after everything is setup (before request_[threaded_]irq() the handler
143 * can't be entered for an event, as the event channel will be unmasked only
146 static DEFINE_RWLOCK(evtchn_rwlock);
151 * irq_mapping_update_lock
154 * percpu eoi_list_lock
157 static LIST_HEAD(xen_irq_list_head);
159 /* IRQ <-> VIRQ mapping. */
160 static DEFINE_PER_CPU(int [NR_VIRQS], virq_to_irq) = {[0 ... NR_VIRQS-1] = -1};
162 /* IRQ <-> IPI mapping */
163 static DEFINE_PER_CPU(int [XEN_NR_IPIS], ipi_to_irq) = {[0 ... XEN_NR_IPIS-1] = -1};
165 /* Event channel distribution data */
166 static atomic_t channels_on_cpu[NR_CPUS];
168 static int **evtchn_to_irq;
170 static unsigned long *pirq_eoi_map;
172 static bool (*pirq_needs_eoi)(unsigned irq);
174 #define EVTCHN_ROW(e) (e / (PAGE_SIZE/sizeof(**evtchn_to_irq)))
175 #define EVTCHN_COL(e) (e % (PAGE_SIZE/sizeof(**evtchn_to_irq)))
176 #define EVTCHN_PER_ROW (PAGE_SIZE / sizeof(**evtchn_to_irq))
178 /* Xen will never allocate port zero for any purpose. */
179 #define VALID_EVTCHN(chn) ((chn) != 0)
181 static struct irq_info *legacy_info_ptrs[NR_IRQS_LEGACY];
183 static struct irq_chip xen_dynamic_chip;
184 static struct irq_chip xen_lateeoi_chip;
185 static struct irq_chip xen_percpu_chip;
186 static struct irq_chip xen_pirq_chip;
187 static void enable_dynirq(struct irq_data *data);
188 static void disable_dynirq(struct irq_data *data);
190 static DEFINE_PER_CPU(unsigned int, irq_epoch);
192 static void clear_evtchn_to_irq_row(unsigned row)
196 for (col = 0; col < EVTCHN_PER_ROW; col++)
197 WRITE_ONCE(evtchn_to_irq[row][col], -1);
200 static void clear_evtchn_to_irq_all(void)
204 for (row = 0; row < EVTCHN_ROW(xen_evtchn_max_channels()); row++) {
205 if (evtchn_to_irq[row] == NULL)
207 clear_evtchn_to_irq_row(row);
211 static int set_evtchn_to_irq(evtchn_port_t evtchn, unsigned int irq)
216 if (evtchn >= xen_evtchn_max_channels())
219 row = EVTCHN_ROW(evtchn);
220 col = EVTCHN_COL(evtchn);
222 if (evtchn_to_irq[row] == NULL) {
223 /* Unallocated irq entries return -1 anyway */
227 evtchn_to_irq[row] = (int *)get_zeroed_page(GFP_KERNEL);
228 if (evtchn_to_irq[row] == NULL)
231 clear_evtchn_to_irq_row(row);
234 WRITE_ONCE(evtchn_to_irq[row][col], irq);
238 int get_evtchn_to_irq(evtchn_port_t evtchn)
240 if (evtchn >= xen_evtchn_max_channels())
242 if (evtchn_to_irq[EVTCHN_ROW(evtchn)] == NULL)
244 return READ_ONCE(evtchn_to_irq[EVTCHN_ROW(evtchn)][EVTCHN_COL(evtchn)]);
247 /* Get info for IRQ */
248 static struct irq_info *info_for_irq(unsigned irq)
250 if (irq < nr_legacy_irqs())
251 return legacy_info_ptrs[irq];
253 return irq_get_chip_data(irq);
256 static void set_info_for_irq(unsigned int irq, struct irq_info *info)
258 if (irq < nr_legacy_irqs())
259 legacy_info_ptrs[irq] = info;
261 irq_set_chip_data(irq, info);
264 /* Per CPU channel accounting */
265 static void channels_on_cpu_dec(struct irq_info *info)
267 if (!info->is_accounted)
270 info->is_accounted = 0;
272 if (WARN_ON_ONCE(info->cpu >= nr_cpu_ids))
275 WARN_ON_ONCE(!atomic_add_unless(&channels_on_cpu[info->cpu], -1 , 0));
278 static void channels_on_cpu_inc(struct irq_info *info)
280 if (WARN_ON_ONCE(info->cpu >= nr_cpu_ids))
283 if (WARN_ON_ONCE(!atomic_add_unless(&channels_on_cpu[info->cpu], 1,
287 info->is_accounted = 1;
290 /* Constructors for packed IRQ information. */
291 static int xen_irq_info_common_setup(struct irq_info *info,
293 enum xen_irq_type type,
294 evtchn_port_t evtchn,
299 BUG_ON(info->type != IRQT_UNBOUND && info->type != type);
303 info->evtchn = evtchn;
306 ret = set_evtchn_to_irq(evtchn, irq);
310 irq_clear_status_flags(irq, IRQ_NOREQUEST|IRQ_NOAUTOEN);
312 return xen_evtchn_port_setup(evtchn);
315 static int xen_irq_info_evtchn_setup(unsigned irq,
316 evtchn_port_t evtchn)
318 struct irq_info *info = info_for_irq(irq);
320 return xen_irq_info_common_setup(info, irq, IRQT_EVTCHN, evtchn, 0);
323 static int xen_irq_info_ipi_setup(unsigned cpu,
325 evtchn_port_t evtchn,
328 struct irq_info *info = info_for_irq(irq);
332 per_cpu(ipi_to_irq, cpu)[ipi] = irq;
334 return xen_irq_info_common_setup(info, irq, IRQT_IPI, evtchn, 0);
337 static int xen_irq_info_virq_setup(unsigned cpu,
339 evtchn_port_t evtchn,
342 struct irq_info *info = info_for_irq(irq);
346 per_cpu(virq_to_irq, cpu)[virq] = irq;
348 return xen_irq_info_common_setup(info, irq, IRQT_VIRQ, evtchn, 0);
351 static int xen_irq_info_pirq_setup(unsigned irq,
352 evtchn_port_t evtchn,
358 struct irq_info *info = info_for_irq(irq);
360 info->u.pirq.pirq = pirq;
361 info->u.pirq.gsi = gsi;
362 info->u.pirq.domid = domid;
363 info->u.pirq.flags = flags;
365 return xen_irq_info_common_setup(info, irq, IRQT_PIRQ, evtchn, 0);
368 static void xen_irq_info_cleanup(struct irq_info *info)
370 set_evtchn_to_irq(info->evtchn, -1);
372 channels_on_cpu_dec(info);
376 * Accessors for packed IRQ information.
378 evtchn_port_t evtchn_from_irq(unsigned irq)
380 const struct irq_info *info = NULL;
382 if (likely(irq < nr_irqs))
383 info = info_for_irq(irq);
390 unsigned int irq_from_evtchn(evtchn_port_t evtchn)
392 return get_evtchn_to_irq(evtchn);
394 EXPORT_SYMBOL_GPL(irq_from_evtchn);
396 int irq_from_virq(unsigned int cpu, unsigned int virq)
398 return per_cpu(virq_to_irq, cpu)[virq];
401 static enum ipi_vector ipi_from_irq(unsigned irq)
403 struct irq_info *info = info_for_irq(irq);
405 BUG_ON(info == NULL);
406 BUG_ON(info->type != IRQT_IPI);
411 static unsigned virq_from_irq(unsigned irq)
413 struct irq_info *info = info_for_irq(irq);
415 BUG_ON(info == NULL);
416 BUG_ON(info->type != IRQT_VIRQ);
421 static unsigned pirq_from_irq(unsigned irq)
423 struct irq_info *info = info_for_irq(irq);
425 BUG_ON(info == NULL);
426 BUG_ON(info->type != IRQT_PIRQ);
428 return info->u.pirq.pirq;
431 static enum xen_irq_type type_from_irq(unsigned irq)
433 return info_for_irq(irq)->type;
436 static unsigned cpu_from_irq(unsigned irq)
438 return info_for_irq(irq)->cpu;
441 unsigned int cpu_from_evtchn(evtchn_port_t evtchn)
443 int irq = get_evtchn_to_irq(evtchn);
447 ret = cpu_from_irq(irq);
453 static bool pirq_check_eoi_map(unsigned irq)
455 return test_bit(pirq_from_irq(irq), pirq_eoi_map);
459 static bool pirq_needs_eoi_flag(unsigned irq)
461 struct irq_info *info = info_for_irq(irq);
462 BUG_ON(info->type != IRQT_PIRQ);
464 return info->u.pirq.flags & PIRQ_NEEDS_EOI;
467 static void bind_evtchn_to_cpu(evtchn_port_t evtchn, unsigned int cpu,
470 int irq = get_evtchn_to_irq(evtchn);
471 struct irq_info *info = info_for_irq(irq);
475 if (IS_ENABLED(CONFIG_SMP) && force_affinity) {
476 cpumask_copy(irq_get_affinity_mask(irq), cpumask_of(cpu));
477 cpumask_copy(irq_get_effective_affinity_mask(irq),
481 xen_evtchn_port_bind_to_cpu(evtchn, cpu, info->cpu);
483 channels_on_cpu_dec(info);
485 channels_on_cpu_inc(info);
489 * notify_remote_via_irq - send event to remote end of event channel via irq
490 * @irq: irq of event channel to send event to
492 * Unlike notify_remote_via_evtchn(), this is safe to use across
493 * save/restore. Notifications on a broken connection are silently
496 void notify_remote_via_irq(int irq)
498 evtchn_port_t evtchn = evtchn_from_irq(irq);
500 if (VALID_EVTCHN(evtchn))
501 notify_remote_via_evtchn(evtchn);
503 EXPORT_SYMBOL_GPL(notify_remote_via_irq);
505 struct lateeoi_work {
506 struct delayed_work delayed;
507 spinlock_t eoi_list_lock;
508 struct list_head eoi_list;
511 static DEFINE_PER_CPU(struct lateeoi_work, lateeoi);
513 static void lateeoi_list_del(struct irq_info *info)
515 struct lateeoi_work *eoi = &per_cpu(lateeoi, info->eoi_cpu);
518 spin_lock_irqsave(&eoi->eoi_list_lock, flags);
519 list_del_init(&info->eoi_list);
520 spin_unlock_irqrestore(&eoi->eoi_list_lock, flags);
523 static void lateeoi_list_add(struct irq_info *info)
525 struct lateeoi_work *eoi = &per_cpu(lateeoi, info->eoi_cpu);
526 struct irq_info *elem;
527 u64 now = get_jiffies_64();
531 if (now < info->eoi_time)
532 delay = info->eoi_time - now;
536 spin_lock_irqsave(&eoi->eoi_list_lock, flags);
538 if (list_empty(&eoi->eoi_list)) {
539 list_add(&info->eoi_list, &eoi->eoi_list);
540 mod_delayed_work_on(info->eoi_cpu, system_wq,
541 &eoi->delayed, delay);
543 list_for_each_entry_reverse(elem, &eoi->eoi_list, eoi_list) {
544 if (elem->eoi_time <= info->eoi_time)
547 list_add(&info->eoi_list, &elem->eoi_list);
550 spin_unlock_irqrestore(&eoi->eoi_list_lock, flags);
553 static void xen_irq_lateeoi_locked(struct irq_info *info, bool spurious)
555 evtchn_port_t evtchn;
557 unsigned int delay = 0;
559 evtchn = info->evtchn;
560 if (!VALID_EVTCHN(evtchn) || !list_empty(&info->eoi_list))
564 if ((1 << info->spurious_cnt) < (HZ << 2)) {
565 if (info->spurious_cnt != 0xFF)
566 info->spurious_cnt++;
568 if (info->spurious_cnt > 1) {
569 delay = 1 << (info->spurious_cnt - 2);
573 info->eoi_cpu = smp_processor_id();
574 info->eoi_time = get_jiffies_64() + delay;
577 info->spurious_cnt = 0;
581 if (info->eoi_time &&
582 (info->irq_epoch == per_cpu(irq_epoch, cpu) || delay)) {
583 lateeoi_list_add(info);
588 unmask_evtchn(evtchn);
591 static void xen_irq_lateeoi_worker(struct work_struct *work)
593 struct lateeoi_work *eoi;
594 struct irq_info *info;
595 u64 now = get_jiffies_64();
598 eoi = container_of(to_delayed_work(work), struct lateeoi_work, delayed);
600 read_lock_irqsave(&evtchn_rwlock, flags);
603 spin_lock(&eoi->eoi_list_lock);
605 info = list_first_entry_or_null(&eoi->eoi_list, struct irq_info,
608 if (info == NULL || now < info->eoi_time) {
609 spin_unlock(&eoi->eoi_list_lock);
613 list_del_init(&info->eoi_list);
615 spin_unlock(&eoi->eoi_list_lock);
619 xen_irq_lateeoi_locked(info, false);
623 mod_delayed_work_on(info->eoi_cpu, system_wq,
624 &eoi->delayed, info->eoi_time - now);
626 read_unlock_irqrestore(&evtchn_rwlock, flags);
629 static void xen_cpu_init_eoi(unsigned int cpu)
631 struct lateeoi_work *eoi = &per_cpu(lateeoi, cpu);
633 INIT_DELAYED_WORK(&eoi->delayed, xen_irq_lateeoi_worker);
634 spin_lock_init(&eoi->eoi_list_lock);
635 INIT_LIST_HEAD(&eoi->eoi_list);
638 void xen_irq_lateeoi(unsigned int irq, unsigned int eoi_flags)
640 struct irq_info *info;
643 read_lock_irqsave(&evtchn_rwlock, flags);
645 info = info_for_irq(irq);
648 xen_irq_lateeoi_locked(info, eoi_flags & XEN_EOI_FLAG_SPURIOUS);
650 read_unlock_irqrestore(&evtchn_rwlock, flags);
652 EXPORT_SYMBOL_GPL(xen_irq_lateeoi);
654 static void xen_irq_init(unsigned irq)
656 struct irq_info *info;
658 info = kzalloc(sizeof(*info), GFP_KERNEL);
660 panic("Unable to allocate metadata for IRQ%d\n", irq);
662 info->type = IRQT_UNBOUND;
665 set_info_for_irq(irq, info);
667 * Interrupt affinity setting can be immediate. No point
668 * in delaying it until an interrupt is handled.
670 irq_set_status_flags(irq, IRQ_MOVE_PCNTXT);
672 INIT_LIST_HEAD(&info->eoi_list);
673 list_add_tail(&info->list, &xen_irq_list_head);
676 static int __must_check xen_allocate_irqs_dynamic(int nvec)
678 int i, irq = irq_alloc_descs(-1, 0, nvec, -1);
681 for (i = 0; i < nvec; i++)
682 xen_irq_init(irq + i);
688 static inline int __must_check xen_allocate_irq_dynamic(void)
691 return xen_allocate_irqs_dynamic(1);
694 static int __must_check xen_allocate_irq_gsi(unsigned gsi)
699 * A PV guest has no concept of a GSI (since it has no ACPI
700 * nor access to/knowledge of the physical APICs). Therefore
701 * all IRQs are dynamically allocated from the entire IRQ
704 if (xen_pv_domain() && !xen_initial_domain())
705 return xen_allocate_irq_dynamic();
707 /* Legacy IRQ descriptors are already allocated by the arch. */
708 if (gsi < nr_legacy_irqs())
711 irq = irq_alloc_desc_at(gsi, -1);
718 static void xen_free_irq(unsigned irq)
720 struct irq_info *info = info_for_irq(irq);
726 write_lock_irqsave(&evtchn_rwlock, flags);
728 if (!list_empty(&info->eoi_list))
729 lateeoi_list_del(info);
731 list_del(&info->list);
733 set_info_for_irq(irq, NULL);
735 WARN_ON(info->refcnt > 0);
737 write_unlock_irqrestore(&evtchn_rwlock, flags);
741 /* Legacy IRQ descriptors are managed by the arch. */
742 if (irq < nr_legacy_irqs())
748 static void xen_evtchn_close(evtchn_port_t port)
750 struct evtchn_close close;
753 if (HYPERVISOR_event_channel_op(EVTCHNOP_close, &close) != 0)
757 static void pirq_query_unmask(int irq)
759 struct physdev_irq_status_query irq_status;
760 struct irq_info *info = info_for_irq(irq);
762 BUG_ON(info->type != IRQT_PIRQ);
764 irq_status.irq = pirq_from_irq(irq);
765 if (HYPERVISOR_physdev_op(PHYSDEVOP_irq_status_query, &irq_status))
766 irq_status.flags = 0;
768 info->u.pirq.flags &= ~PIRQ_NEEDS_EOI;
769 if (irq_status.flags & XENIRQSTAT_needs_eoi)
770 info->u.pirq.flags |= PIRQ_NEEDS_EOI;
773 static void eoi_pirq(struct irq_data *data)
775 evtchn_port_t evtchn = evtchn_from_irq(data->irq);
776 struct physdev_eoi eoi = { .irq = pirq_from_irq(data->irq) };
779 if (!VALID_EVTCHN(evtchn))
782 clear_evtchn(evtchn);
784 if (pirq_needs_eoi(data->irq)) {
785 rc = HYPERVISOR_physdev_op(PHYSDEVOP_eoi, &eoi);
790 static void mask_ack_pirq(struct irq_data *data)
792 disable_dynirq(data);
796 static unsigned int __startup_pirq(unsigned int irq)
798 struct evtchn_bind_pirq bind_pirq;
799 struct irq_info *info = info_for_irq(irq);
800 evtchn_port_t evtchn = evtchn_from_irq(irq);
803 BUG_ON(info->type != IRQT_PIRQ);
805 if (VALID_EVTCHN(evtchn))
808 bind_pirq.pirq = pirq_from_irq(irq);
809 /* NB. We are happy to share unless we are probing. */
810 bind_pirq.flags = info->u.pirq.flags & PIRQ_SHAREABLE ?
811 BIND_PIRQ__WILL_SHARE : 0;
812 rc = HYPERVISOR_event_channel_op(EVTCHNOP_bind_pirq, &bind_pirq);
814 pr_warn("Failed to obtain physical IRQ %d\n", irq);
817 evtchn = bind_pirq.port;
819 pirq_query_unmask(irq);
821 rc = set_evtchn_to_irq(evtchn, irq);
825 info->evtchn = evtchn;
826 bind_evtchn_to_cpu(evtchn, 0, false);
828 rc = xen_evtchn_port_setup(evtchn);
833 unmask_evtchn(evtchn);
834 eoi_pirq(irq_get_irq_data(irq));
839 pr_err("irq%d: Failed to set port to irq mapping (%d)\n", irq, rc);
840 xen_evtchn_close(evtchn);
844 static unsigned int startup_pirq(struct irq_data *data)
846 return __startup_pirq(data->irq);
849 static void shutdown_pirq(struct irq_data *data)
851 unsigned int irq = data->irq;
852 struct irq_info *info = info_for_irq(irq);
853 evtchn_port_t evtchn = evtchn_from_irq(irq);
855 BUG_ON(info->type != IRQT_PIRQ);
857 if (!VALID_EVTCHN(evtchn))
861 xen_evtchn_close(evtchn);
862 xen_irq_info_cleanup(info);
865 static void enable_pirq(struct irq_data *data)
870 static void disable_pirq(struct irq_data *data)
872 disable_dynirq(data);
875 int xen_irq_from_gsi(unsigned gsi)
877 struct irq_info *info;
879 list_for_each_entry(info, &xen_irq_list_head, list) {
880 if (info->type != IRQT_PIRQ)
883 if (info->u.pirq.gsi == gsi)
889 EXPORT_SYMBOL_GPL(xen_irq_from_gsi);
891 static void __unbind_from_irq(unsigned int irq)
893 evtchn_port_t evtchn = evtchn_from_irq(irq);
894 struct irq_info *info = info_for_irq(irq);
896 if (info->refcnt > 0) {
898 if (info->refcnt != 0)
902 if (VALID_EVTCHN(evtchn)) {
903 unsigned int cpu = cpu_from_irq(irq);
905 xen_evtchn_close(evtchn);
907 switch (type_from_irq(irq)) {
909 per_cpu(virq_to_irq, cpu)[virq_from_irq(irq)] = -1;
912 per_cpu(ipi_to_irq, cpu)[ipi_from_irq(irq)] = -1;
918 xen_irq_info_cleanup(info);
925 * Do not make any assumptions regarding the relationship between the
926 * IRQ number returned here and the Xen pirq argument.
928 * Note: We don't assign an event channel until the irq actually started
929 * up. Return an existing irq if we've already got one for the gsi.
931 * Shareable implies level triggered, not shareable implies edge
934 int xen_bind_pirq_gsi_to_irq(unsigned gsi,
935 unsigned pirq, int shareable, char *name)
938 struct physdev_irq irq_op;
941 mutex_lock(&irq_mapping_update_lock);
943 irq = xen_irq_from_gsi(gsi);
945 pr_info("%s: returning irq %d for gsi %u\n",
950 irq = xen_allocate_irq_gsi(gsi);
957 /* Only the privileged domain can do this. For non-priv, the pcifront
958 * driver provides a PCI bus that does the call to do exactly
959 * this in the priv domain. */
960 if (xen_initial_domain() &&
961 HYPERVISOR_physdev_op(PHYSDEVOP_alloc_irq_vector, &irq_op)) {
967 ret = xen_irq_info_pirq_setup(irq, 0, pirq, gsi, DOMID_SELF,
968 shareable ? PIRQ_SHAREABLE : 0);
970 __unbind_from_irq(irq);
975 pirq_query_unmask(irq);
976 /* We try to use the handler with the appropriate semantic for the
977 * type of interrupt: if the interrupt is an edge triggered
978 * interrupt we use handle_edge_irq.
980 * On the other hand if the interrupt is level triggered we use
981 * handle_fasteoi_irq like the native code does for this kind of
984 * Depending on the Xen version, pirq_needs_eoi might return true
985 * not only for level triggered interrupts but for edge triggered
986 * interrupts too. In any case Xen always honors the eoi mechanism,
987 * not injecting any more pirqs of the same kind if the first one
988 * hasn't received an eoi yet. Therefore using the fasteoi handler
989 * is the right choice either way.
992 irq_set_chip_and_handler_name(irq, &xen_pirq_chip,
993 handle_fasteoi_irq, name);
995 irq_set_chip_and_handler_name(irq, &xen_pirq_chip,
996 handle_edge_irq, name);
999 mutex_unlock(&irq_mapping_update_lock);
1004 #ifdef CONFIG_PCI_MSI
1005 int xen_allocate_pirq_msi(struct pci_dev *dev, struct msi_desc *msidesc)
1008 struct physdev_get_free_pirq op_get_free_pirq;
1010 op_get_free_pirq.type = MAP_PIRQ_TYPE_MSI;
1011 rc = HYPERVISOR_physdev_op(PHYSDEVOP_get_free_pirq, &op_get_free_pirq);
1013 WARN_ONCE(rc == -ENOSYS,
1014 "hypervisor does not support the PHYSDEVOP_get_free_pirq interface\n");
1016 return rc ? -1 : op_get_free_pirq.pirq;
1019 int xen_bind_pirq_msi_to_irq(struct pci_dev *dev, struct msi_desc *msidesc,
1020 int pirq, int nvec, const char *name, domid_t domid)
1024 mutex_lock(&irq_mapping_update_lock);
1026 irq = xen_allocate_irqs_dynamic(nvec);
1030 for (i = 0; i < nvec; i++) {
1031 irq_set_chip_and_handler_name(irq + i, &xen_pirq_chip, handle_edge_irq, name);
1033 ret = xen_irq_info_pirq_setup(irq + i, 0, pirq + i, 0, domid,
1034 i == 0 ? 0 : PIRQ_MSI_GROUP);
1039 ret = irq_set_msi_desc(irq, msidesc);
1043 mutex_unlock(&irq_mapping_update_lock);
1047 __unbind_from_irq(irq + nvec);
1048 mutex_unlock(&irq_mapping_update_lock);
1053 int xen_destroy_irq(int irq)
1055 struct physdev_unmap_pirq unmap_irq;
1056 struct irq_info *info = info_for_irq(irq);
1059 mutex_lock(&irq_mapping_update_lock);
1062 * If trying to remove a vector in a MSI group different
1063 * than the first one skip the PIRQ unmap unless this vector
1064 * is the first one in the group.
1066 if (xen_initial_domain() && !(info->u.pirq.flags & PIRQ_MSI_GROUP)) {
1067 unmap_irq.pirq = info->u.pirq.pirq;
1068 unmap_irq.domid = info->u.pirq.domid;
1069 rc = HYPERVISOR_physdev_op(PHYSDEVOP_unmap_pirq, &unmap_irq);
1070 /* If another domain quits without making the pci_disable_msix
1071 * call, the Xen hypervisor takes care of freeing the PIRQs
1072 * (free_domain_pirqs).
1074 if ((rc == -ESRCH && info->u.pirq.domid != DOMID_SELF))
1075 pr_info("domain %d does not have %d anymore\n",
1076 info->u.pirq.domid, info->u.pirq.pirq);
1078 pr_warn("unmap irq failed %d\n", rc);
1086 mutex_unlock(&irq_mapping_update_lock);
1090 int xen_irq_from_pirq(unsigned pirq)
1094 struct irq_info *info;
1096 mutex_lock(&irq_mapping_update_lock);
1098 list_for_each_entry(info, &xen_irq_list_head, list) {
1099 if (info->type != IRQT_PIRQ)
1102 if (info->u.pirq.pirq == pirq)
1107 mutex_unlock(&irq_mapping_update_lock);
1113 int xen_pirq_from_irq(unsigned irq)
1115 return pirq_from_irq(irq);
1117 EXPORT_SYMBOL_GPL(xen_pirq_from_irq);
1119 static int bind_evtchn_to_irq_chip(evtchn_port_t evtchn, struct irq_chip *chip)
1124 if (evtchn >= xen_evtchn_max_channels())
1127 mutex_lock(&irq_mapping_update_lock);
1129 irq = get_evtchn_to_irq(evtchn);
1132 irq = xen_allocate_irq_dynamic();
1136 irq_set_chip_and_handler_name(irq, chip,
1137 handle_edge_irq, "event");
1139 ret = xen_irq_info_evtchn_setup(irq, evtchn);
1141 __unbind_from_irq(irq);
1146 * New interdomain events are initially bound to vCPU0 This
1147 * is required to setup the event channel in the first
1148 * place and also important for UP guests because the
1149 * affinity setting is not invoked on them so nothing would
1152 bind_evtchn_to_cpu(evtchn, 0, false);
1154 struct irq_info *info = info_for_irq(irq);
1155 WARN_ON(info == NULL || info->type != IRQT_EVTCHN);
1159 mutex_unlock(&irq_mapping_update_lock);
1164 int bind_evtchn_to_irq(evtchn_port_t evtchn)
1166 return bind_evtchn_to_irq_chip(evtchn, &xen_dynamic_chip);
1168 EXPORT_SYMBOL_GPL(bind_evtchn_to_irq);
1170 static int bind_ipi_to_irq(unsigned int ipi, unsigned int cpu)
1172 struct evtchn_bind_ipi bind_ipi;
1173 evtchn_port_t evtchn;
1176 mutex_lock(&irq_mapping_update_lock);
1178 irq = per_cpu(ipi_to_irq, cpu)[ipi];
1181 irq = xen_allocate_irq_dynamic();
1185 irq_set_chip_and_handler_name(irq, &xen_percpu_chip,
1186 handle_percpu_irq, "ipi");
1188 bind_ipi.vcpu = xen_vcpu_nr(cpu);
1189 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_ipi,
1192 evtchn = bind_ipi.port;
1194 ret = xen_irq_info_ipi_setup(cpu, irq, evtchn, ipi);
1196 __unbind_from_irq(irq);
1201 * Force the affinity mask to the target CPU so proc shows
1202 * the correct target.
1204 bind_evtchn_to_cpu(evtchn, cpu, true);
1206 struct irq_info *info = info_for_irq(irq);
1207 WARN_ON(info == NULL || info->type != IRQT_IPI);
1211 mutex_unlock(&irq_mapping_update_lock);
1215 static int bind_interdomain_evtchn_to_irq_chip(unsigned int remote_domain,
1216 evtchn_port_t remote_port,
1217 struct irq_chip *chip)
1219 struct evtchn_bind_interdomain bind_interdomain;
1222 bind_interdomain.remote_dom = remote_domain;
1223 bind_interdomain.remote_port = remote_port;
1225 err = HYPERVISOR_event_channel_op(EVTCHNOP_bind_interdomain,
1228 return err ? : bind_evtchn_to_irq_chip(bind_interdomain.local_port,
1232 int bind_interdomain_evtchn_to_irq_lateeoi(unsigned int remote_domain,
1233 evtchn_port_t remote_port)
1235 return bind_interdomain_evtchn_to_irq_chip(remote_domain, remote_port,
1238 EXPORT_SYMBOL_GPL(bind_interdomain_evtchn_to_irq_lateeoi);
1240 static int find_virq(unsigned int virq, unsigned int cpu, evtchn_port_t *evtchn)
1242 struct evtchn_status status;
1246 memset(&status, 0, sizeof(status));
1247 for (port = 0; port < xen_evtchn_max_channels(); port++) {
1248 status.dom = DOMID_SELF;
1250 rc = HYPERVISOR_event_channel_op(EVTCHNOP_status, &status);
1253 if (status.status != EVTCHNSTAT_virq)
1255 if (status.u.virq == virq && status.vcpu == xen_vcpu_nr(cpu)) {
1264 * xen_evtchn_nr_channels - number of usable event channel ports
1266 * This may be less than the maximum supported by the current
1267 * hypervisor ABI. Use xen_evtchn_max_channels() for the maximum
1270 unsigned xen_evtchn_nr_channels(void)
1272 return evtchn_ops->nr_channels();
1274 EXPORT_SYMBOL_GPL(xen_evtchn_nr_channels);
1276 int bind_virq_to_irq(unsigned int virq, unsigned int cpu, bool percpu)
1278 struct evtchn_bind_virq bind_virq;
1279 evtchn_port_t evtchn = 0;
1282 mutex_lock(&irq_mapping_update_lock);
1284 irq = per_cpu(virq_to_irq, cpu)[virq];
1287 irq = xen_allocate_irq_dynamic();
1292 irq_set_chip_and_handler_name(irq, &xen_percpu_chip,
1293 handle_percpu_irq, "virq");
1295 irq_set_chip_and_handler_name(irq, &xen_dynamic_chip,
1296 handle_edge_irq, "virq");
1298 bind_virq.virq = virq;
1299 bind_virq.vcpu = xen_vcpu_nr(cpu);
1300 ret = HYPERVISOR_event_channel_op(EVTCHNOP_bind_virq,
1303 evtchn = bind_virq.port;
1306 ret = find_virq(virq, cpu, &evtchn);
1310 ret = xen_irq_info_virq_setup(cpu, irq, evtchn, virq);
1312 __unbind_from_irq(irq);
1318 * Force the affinity mask for percpu interrupts so proc
1319 * shows the correct target.
1321 bind_evtchn_to_cpu(evtchn, cpu, percpu);
1323 struct irq_info *info = info_for_irq(irq);
1324 WARN_ON(info == NULL || info->type != IRQT_VIRQ);
1328 mutex_unlock(&irq_mapping_update_lock);
1333 static void unbind_from_irq(unsigned int irq)
1335 mutex_lock(&irq_mapping_update_lock);
1336 __unbind_from_irq(irq);
1337 mutex_unlock(&irq_mapping_update_lock);
1340 static int bind_evtchn_to_irqhandler_chip(evtchn_port_t evtchn,
1341 irq_handler_t handler,
1342 unsigned long irqflags,
1343 const char *devname, void *dev_id,
1344 struct irq_chip *chip)
1348 irq = bind_evtchn_to_irq_chip(evtchn, chip);
1351 retval = request_irq(irq, handler, irqflags, devname, dev_id);
1353 unbind_from_irq(irq);
1360 int bind_evtchn_to_irqhandler(evtchn_port_t evtchn,
1361 irq_handler_t handler,
1362 unsigned long irqflags,
1363 const char *devname, void *dev_id)
1365 return bind_evtchn_to_irqhandler_chip(evtchn, handler, irqflags,
1369 EXPORT_SYMBOL_GPL(bind_evtchn_to_irqhandler);
1371 int bind_evtchn_to_irqhandler_lateeoi(evtchn_port_t evtchn,
1372 irq_handler_t handler,
1373 unsigned long irqflags,
1374 const char *devname, void *dev_id)
1376 return bind_evtchn_to_irqhandler_chip(evtchn, handler, irqflags,
1380 EXPORT_SYMBOL_GPL(bind_evtchn_to_irqhandler_lateeoi);
1382 static int bind_interdomain_evtchn_to_irqhandler_chip(
1383 unsigned int remote_domain, evtchn_port_t remote_port,
1384 irq_handler_t handler, unsigned long irqflags,
1385 const char *devname, void *dev_id, struct irq_chip *chip)
1389 irq = bind_interdomain_evtchn_to_irq_chip(remote_domain, remote_port,
1394 retval = request_irq(irq, handler, irqflags, devname, dev_id);
1396 unbind_from_irq(irq);
1403 int bind_interdomain_evtchn_to_irqhandler_lateeoi(unsigned int remote_domain,
1404 evtchn_port_t remote_port,
1405 irq_handler_t handler,
1406 unsigned long irqflags,
1407 const char *devname,
1410 return bind_interdomain_evtchn_to_irqhandler_chip(remote_domain,
1411 remote_port, handler, irqflags, devname,
1412 dev_id, &xen_lateeoi_chip);
1414 EXPORT_SYMBOL_GPL(bind_interdomain_evtchn_to_irqhandler_lateeoi);
1416 int bind_virq_to_irqhandler(unsigned int virq, unsigned int cpu,
1417 irq_handler_t handler,
1418 unsigned long irqflags, const char *devname, void *dev_id)
1422 irq = bind_virq_to_irq(virq, cpu, irqflags & IRQF_PERCPU);
1425 retval = request_irq(irq, handler, irqflags, devname, dev_id);
1427 unbind_from_irq(irq);
1433 EXPORT_SYMBOL_GPL(bind_virq_to_irqhandler);
1435 int bind_ipi_to_irqhandler(enum ipi_vector ipi,
1437 irq_handler_t handler,
1438 unsigned long irqflags,
1439 const char *devname,
1444 irq = bind_ipi_to_irq(ipi, cpu);
1448 irqflags |= IRQF_NO_SUSPEND | IRQF_FORCE_RESUME | IRQF_EARLY_RESUME;
1449 retval = request_irq(irq, handler, irqflags, devname, dev_id);
1451 unbind_from_irq(irq);
1458 void unbind_from_irqhandler(unsigned int irq, void *dev_id)
1460 struct irq_info *info = info_for_irq(irq);
1464 free_irq(irq, dev_id);
1465 unbind_from_irq(irq);
1467 EXPORT_SYMBOL_GPL(unbind_from_irqhandler);
1470 * xen_set_irq_priority() - set an event channel priority.
1471 * @irq:irq bound to an event channel.
1472 * @priority: priority between XEN_IRQ_PRIORITY_MAX and XEN_IRQ_PRIORITY_MIN.
1474 int xen_set_irq_priority(unsigned irq, unsigned priority)
1476 struct evtchn_set_priority set_priority;
1478 set_priority.port = evtchn_from_irq(irq);
1479 set_priority.priority = priority;
1481 return HYPERVISOR_event_channel_op(EVTCHNOP_set_priority,
1484 EXPORT_SYMBOL_GPL(xen_set_irq_priority);
1486 int evtchn_make_refcounted(evtchn_port_t evtchn)
1488 int irq = get_evtchn_to_irq(evtchn);
1489 struct irq_info *info;
1494 info = info_for_irq(irq);
1499 WARN_ON(info->refcnt != -1);
1505 EXPORT_SYMBOL_GPL(evtchn_make_refcounted);
1507 int evtchn_get(evtchn_port_t evtchn)
1510 struct irq_info *info;
1513 if (evtchn >= xen_evtchn_max_channels())
1516 mutex_lock(&irq_mapping_update_lock);
1518 irq = get_evtchn_to_irq(evtchn);
1522 info = info_for_irq(irq);
1528 if (info->refcnt <= 0 || info->refcnt == SHRT_MAX)
1534 mutex_unlock(&irq_mapping_update_lock);
1538 EXPORT_SYMBOL_GPL(evtchn_get);
1540 void evtchn_put(evtchn_port_t evtchn)
1542 int irq = get_evtchn_to_irq(evtchn);
1543 if (WARN_ON(irq == -1))
1545 unbind_from_irq(irq);
1547 EXPORT_SYMBOL_GPL(evtchn_put);
1549 void xen_send_IPI_one(unsigned int cpu, enum ipi_vector vector)
1554 if (unlikely(vector == XEN_NMI_VECTOR)) {
1555 int rc = HYPERVISOR_vcpu_op(VCPUOP_send_nmi, xen_vcpu_nr(cpu),
1558 printk(KERN_WARNING "Sending nmi to CPU%d failed (rc:%d)\n", cpu, rc);
1562 irq = per_cpu(ipi_to_irq, cpu)[vector];
1564 notify_remote_via_irq(irq);
1567 struct evtchn_loop_ctrl {
1573 void handle_irq_for_port(evtchn_port_t port, struct evtchn_loop_ctrl *ctrl)
1576 struct irq_info *info;
1578 irq = get_evtchn_to_irq(port);
1583 * Check for timeout every 256 events.
1584 * We are setting the timeout value only after the first 256
1585 * events in order to not hurt the common case of few loop
1586 * iterations. The 256 is basically an arbitrary value.
1588 * In case we are hitting the timeout we need to defer all further
1589 * EOIs in order to ensure to leave the event handling loop rather
1590 * sooner than later.
1592 if (!ctrl->defer_eoi && !(++ctrl->count & 0xff)) {
1593 ktime_t kt = ktime_get();
1595 if (!ctrl->timeout) {
1596 kt = ktime_add_ms(kt,
1597 jiffies_to_msecs(event_loop_timeout));
1599 } else if (kt > ctrl->timeout) {
1600 ctrl->defer_eoi = true;
1604 info = info_for_irq(irq);
1606 if (ctrl->defer_eoi) {
1607 info->eoi_cpu = smp_processor_id();
1608 info->irq_epoch = __this_cpu_read(irq_epoch);
1609 info->eoi_time = get_jiffies_64() + event_eoi_delay;
1612 generic_handle_irq(irq);
1615 static void __xen_evtchn_do_upcall(void)
1617 struct vcpu_info *vcpu_info = __this_cpu_read(xen_vcpu);
1618 int cpu = smp_processor_id();
1619 struct evtchn_loop_ctrl ctrl = { 0 };
1621 read_lock(&evtchn_rwlock);
1624 vcpu_info->evtchn_upcall_pending = 0;
1626 xen_evtchn_handle_events(cpu, &ctrl);
1628 BUG_ON(!irqs_disabled());
1630 virt_rmb(); /* Hypervisor can set upcall pending. */
1632 } while (vcpu_info->evtchn_upcall_pending);
1634 read_unlock(&evtchn_rwlock);
1637 * Increment irq_epoch only now to defer EOIs only for
1638 * xen_irq_lateeoi() invocations occurring from inside the loop
1641 __this_cpu_inc(irq_epoch);
1644 void xen_evtchn_do_upcall(struct pt_regs *regs)
1646 struct pt_regs *old_regs = set_irq_regs(regs);
1650 __xen_evtchn_do_upcall();
1653 set_irq_regs(old_regs);
1656 void xen_hvm_evtchn_do_upcall(void)
1658 __xen_evtchn_do_upcall();
1660 EXPORT_SYMBOL_GPL(xen_hvm_evtchn_do_upcall);
1662 /* Rebind a new event channel to an existing irq. */
1663 void rebind_evtchn_irq(evtchn_port_t evtchn, int irq)
1665 struct irq_info *info = info_for_irq(irq);
1670 /* Make sure the irq is masked, since the new event channel
1671 will also be masked. */
1674 mutex_lock(&irq_mapping_update_lock);
1676 /* After resume the irq<->evtchn mappings are all cleared out */
1677 BUG_ON(get_evtchn_to_irq(evtchn) != -1);
1678 /* Expect irq to have been bound before,
1679 so there should be a proper type */
1680 BUG_ON(info->type == IRQT_UNBOUND);
1682 (void)xen_irq_info_evtchn_setup(irq, evtchn);
1684 mutex_unlock(&irq_mapping_update_lock);
1686 bind_evtchn_to_cpu(evtchn, info->cpu, false);
1688 /* Unmask the event channel. */
1692 /* Rebind an evtchn so that it gets delivered to a specific cpu */
1693 static int xen_rebind_evtchn_to_cpu(evtchn_port_t evtchn, unsigned int tcpu)
1695 struct evtchn_bind_vcpu bind_vcpu;
1698 if (!VALID_EVTCHN(evtchn))
1701 if (!xen_support_evtchn_rebind())
1704 /* Send future instances of this interrupt to other vcpu. */
1705 bind_vcpu.port = evtchn;
1706 bind_vcpu.vcpu = xen_vcpu_nr(tcpu);
1709 * Mask the event while changing the VCPU binding to prevent
1710 * it being delivered on an unexpected VCPU.
1712 masked = test_and_set_mask(evtchn);
1715 * If this fails, it usually just indicates that we're dealing with a
1716 * virq or IPI channel, which don't actually need to be rebound. Ignore
1717 * it, but don't do the xenlinux-level rebind in that case.
1719 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_vcpu, &bind_vcpu) >= 0)
1720 bind_evtchn_to_cpu(evtchn, tcpu, false);
1723 unmask_evtchn(evtchn);
1729 * Find the CPU within @dest mask which has the least number of channels
1730 * assigned. This is not precise as the per cpu counts can be modified
1733 static unsigned int select_target_cpu(const struct cpumask *dest)
1735 unsigned int cpu, best_cpu = UINT_MAX, minch = UINT_MAX;
1737 for_each_cpu_and(cpu, dest, cpu_online_mask) {
1738 unsigned int curch = atomic_read(&channels_on_cpu[cpu]);
1740 if (curch < minch) {
1747 * Catch the unlikely case that dest contains no online CPUs. Can't
1750 if (best_cpu == UINT_MAX)
1751 return select_target_cpu(cpu_online_mask);
1756 static int set_affinity_irq(struct irq_data *data, const struct cpumask *dest,
1759 unsigned int tcpu = select_target_cpu(dest);
1762 ret = xen_rebind_evtchn_to_cpu(evtchn_from_irq(data->irq), tcpu);
1764 irq_data_update_effective_affinity(data, cpumask_of(tcpu));
1769 static void enable_dynirq(struct irq_data *data)
1771 evtchn_port_t evtchn = evtchn_from_irq(data->irq);
1773 if (VALID_EVTCHN(evtchn))
1774 unmask_evtchn(evtchn);
1777 static void disable_dynirq(struct irq_data *data)
1779 evtchn_port_t evtchn = evtchn_from_irq(data->irq);
1781 if (VALID_EVTCHN(evtchn))
1782 mask_evtchn(evtchn);
1785 static void ack_dynirq(struct irq_data *data)
1787 evtchn_port_t evtchn = evtchn_from_irq(data->irq);
1789 if (!VALID_EVTCHN(evtchn))
1792 clear_evtchn(evtchn);
1795 static void mask_ack_dynirq(struct irq_data *data)
1797 disable_dynirq(data);
1801 static int retrigger_dynirq(struct irq_data *data)
1803 evtchn_port_t evtchn = evtchn_from_irq(data->irq);
1806 if (!VALID_EVTCHN(evtchn))
1809 masked = test_and_set_mask(evtchn);
1812 unmask_evtchn(evtchn);
1817 static void restore_pirqs(void)
1819 int pirq, rc, irq, gsi;
1820 struct physdev_map_pirq map_irq;
1821 struct irq_info *info;
1823 list_for_each_entry(info, &xen_irq_list_head, list) {
1824 if (info->type != IRQT_PIRQ)
1827 pirq = info->u.pirq.pirq;
1828 gsi = info->u.pirq.gsi;
1831 /* save/restore of PT devices doesn't work, so at this point the
1832 * only devices present are GSI based emulated devices */
1836 map_irq.domid = DOMID_SELF;
1837 map_irq.type = MAP_PIRQ_TYPE_GSI;
1838 map_irq.index = gsi;
1839 map_irq.pirq = pirq;
1841 rc = HYPERVISOR_physdev_op(PHYSDEVOP_map_pirq, &map_irq);
1843 pr_warn("xen map irq failed gsi=%d irq=%d pirq=%d rc=%d\n",
1844 gsi, irq, pirq, rc);
1849 printk(KERN_DEBUG "xen: --> irq=%d, pirq=%d\n", irq, map_irq.pirq);
1851 __startup_pirq(irq);
1855 static void restore_cpu_virqs(unsigned int cpu)
1857 struct evtchn_bind_virq bind_virq;
1858 evtchn_port_t evtchn;
1861 for (virq = 0; virq < NR_VIRQS; virq++) {
1862 if ((irq = per_cpu(virq_to_irq, cpu)[virq]) == -1)
1865 BUG_ON(virq_from_irq(irq) != virq);
1867 /* Get a new binding from Xen. */
1868 bind_virq.virq = virq;
1869 bind_virq.vcpu = xen_vcpu_nr(cpu);
1870 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_virq,
1873 evtchn = bind_virq.port;
1875 /* Record the new mapping. */
1876 (void)xen_irq_info_virq_setup(cpu, irq, evtchn, virq);
1877 /* The affinity mask is still valid */
1878 bind_evtchn_to_cpu(evtchn, cpu, false);
1882 static void restore_cpu_ipis(unsigned int cpu)
1884 struct evtchn_bind_ipi bind_ipi;
1885 evtchn_port_t evtchn;
1888 for (ipi = 0; ipi < XEN_NR_IPIS; ipi++) {
1889 if ((irq = per_cpu(ipi_to_irq, cpu)[ipi]) == -1)
1892 BUG_ON(ipi_from_irq(irq) != ipi);
1894 /* Get a new binding from Xen. */
1895 bind_ipi.vcpu = xen_vcpu_nr(cpu);
1896 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_ipi,
1899 evtchn = bind_ipi.port;
1901 /* Record the new mapping. */
1902 (void)xen_irq_info_ipi_setup(cpu, irq, evtchn, ipi);
1903 /* The affinity mask is still valid */
1904 bind_evtchn_to_cpu(evtchn, cpu, false);
1908 /* Clear an irq's pending state, in preparation for polling on it */
1909 void xen_clear_irq_pending(int irq)
1911 evtchn_port_t evtchn = evtchn_from_irq(irq);
1913 if (VALID_EVTCHN(evtchn))
1914 clear_evtchn(evtchn);
1916 EXPORT_SYMBOL(xen_clear_irq_pending);
1917 void xen_set_irq_pending(int irq)
1919 evtchn_port_t evtchn = evtchn_from_irq(irq);
1921 if (VALID_EVTCHN(evtchn))
1925 bool xen_test_irq_pending(int irq)
1927 evtchn_port_t evtchn = evtchn_from_irq(irq);
1930 if (VALID_EVTCHN(evtchn))
1931 ret = test_evtchn(evtchn);
1936 /* Poll waiting for an irq to become pending with timeout. In the usual case,
1937 * the irq will be disabled so it won't deliver an interrupt. */
1938 void xen_poll_irq_timeout(int irq, u64 timeout)
1940 evtchn_port_t evtchn = evtchn_from_irq(irq);
1942 if (VALID_EVTCHN(evtchn)) {
1943 struct sched_poll poll;
1946 poll.timeout = timeout;
1947 set_xen_guest_handle(poll.ports, &evtchn);
1949 if (HYPERVISOR_sched_op(SCHEDOP_poll, &poll) != 0)
1953 EXPORT_SYMBOL(xen_poll_irq_timeout);
1954 /* Poll waiting for an irq to become pending. In the usual case, the
1955 * irq will be disabled so it won't deliver an interrupt. */
1956 void xen_poll_irq(int irq)
1958 xen_poll_irq_timeout(irq, 0 /* no timeout */);
1961 /* Check whether the IRQ line is shared with other guests. */
1962 int xen_test_irq_shared(int irq)
1964 struct irq_info *info = info_for_irq(irq);
1965 struct physdev_irq_status_query irq_status;
1970 irq_status.irq = info->u.pirq.pirq;
1972 if (HYPERVISOR_physdev_op(PHYSDEVOP_irq_status_query, &irq_status))
1974 return !(irq_status.flags & XENIRQSTAT_shared);
1976 EXPORT_SYMBOL_GPL(xen_test_irq_shared);
1978 void xen_irq_resume(void)
1981 struct irq_info *info;
1983 /* New event-channel space is not 'live' yet. */
1984 xen_evtchn_resume();
1986 /* No IRQ <-> event-channel mappings. */
1987 list_for_each_entry(info, &xen_irq_list_head, list) {
1988 /* Zap event-channel binding */
1990 /* Adjust accounting */
1991 channels_on_cpu_dec(info);
1994 clear_evtchn_to_irq_all();
1996 for_each_possible_cpu(cpu) {
1997 restore_cpu_virqs(cpu);
1998 restore_cpu_ipis(cpu);
2004 static struct irq_chip xen_dynamic_chip __read_mostly = {
2007 .irq_disable = disable_dynirq,
2008 .irq_mask = disable_dynirq,
2009 .irq_unmask = enable_dynirq,
2011 .irq_ack = ack_dynirq,
2012 .irq_mask_ack = mask_ack_dynirq,
2014 .irq_set_affinity = set_affinity_irq,
2015 .irq_retrigger = retrigger_dynirq,
2018 static struct irq_chip xen_lateeoi_chip __read_mostly = {
2019 /* The chip name needs to contain "xen-dyn" for irqbalance to work. */
2020 .name = "xen-dyn-lateeoi",
2022 .irq_disable = disable_dynirq,
2023 .irq_mask = disable_dynirq,
2024 .irq_unmask = enable_dynirq,
2026 .irq_ack = mask_ack_dynirq,
2027 .irq_mask_ack = mask_ack_dynirq,
2029 .irq_set_affinity = set_affinity_irq,
2030 .irq_retrigger = retrigger_dynirq,
2033 static struct irq_chip xen_pirq_chip __read_mostly = {
2036 .irq_startup = startup_pirq,
2037 .irq_shutdown = shutdown_pirq,
2038 .irq_enable = enable_pirq,
2039 .irq_disable = disable_pirq,
2041 .irq_mask = disable_dynirq,
2042 .irq_unmask = enable_dynirq,
2044 .irq_ack = eoi_pirq,
2045 .irq_eoi = eoi_pirq,
2046 .irq_mask_ack = mask_ack_pirq,
2048 .irq_set_affinity = set_affinity_irq,
2050 .irq_retrigger = retrigger_dynirq,
2053 static struct irq_chip xen_percpu_chip __read_mostly = {
2054 .name = "xen-percpu",
2056 .irq_disable = disable_dynirq,
2057 .irq_mask = disable_dynirq,
2058 .irq_unmask = enable_dynirq,
2060 .irq_ack = ack_dynirq,
2063 #ifdef CONFIG_XEN_PVHVM
2064 /* Vector callbacks are better than PCI interrupts to receive event
2065 * channel notifications because we can receive vector callbacks on any
2066 * vcpu and we don't need PCI support or APIC interactions. */
2067 void xen_setup_callback_vector(void)
2069 uint64_t callback_via;
2071 if (xen_have_vector_callback) {
2072 callback_via = HVM_CALLBACK_VECTOR(HYPERVISOR_CALLBACK_VECTOR);
2073 if (xen_set_callback_via(callback_via)) {
2074 pr_err("Request for Xen HVM callback vector failed\n");
2075 xen_have_vector_callback = 0;
2080 static __init void xen_alloc_callback_vector(void)
2082 if (!xen_have_vector_callback)
2085 pr_info("Xen HVM callback vector for event delivery is enabled\n");
2086 alloc_intr_gate(HYPERVISOR_CALLBACK_VECTOR, asm_sysvec_xen_hvm_callback);
2089 void xen_setup_callback_vector(void) {}
2090 static inline void xen_alloc_callback_vector(void) {}
2093 bool xen_fifo_events = true;
2094 module_param_named(fifo_events, xen_fifo_events, bool, 0);
2096 static int xen_evtchn_cpu_prepare(unsigned int cpu)
2100 xen_cpu_init_eoi(cpu);
2102 if (evtchn_ops->percpu_init)
2103 ret = evtchn_ops->percpu_init(cpu);
2108 static int xen_evtchn_cpu_dead(unsigned int cpu)
2112 if (evtchn_ops->percpu_deinit)
2113 ret = evtchn_ops->percpu_deinit(cpu);
2118 void __init xen_init_IRQ(void)
2121 evtchn_port_t evtchn;
2123 if (xen_fifo_events)
2124 ret = xen_evtchn_fifo_init();
2126 xen_evtchn_2l_init();
2127 xen_fifo_events = false;
2130 xen_cpu_init_eoi(smp_processor_id());
2132 cpuhp_setup_state_nocalls(CPUHP_XEN_EVTCHN_PREPARE,
2133 "xen/evtchn:prepare",
2134 xen_evtchn_cpu_prepare, xen_evtchn_cpu_dead);
2136 evtchn_to_irq = kcalloc(EVTCHN_ROW(xen_evtchn_max_channels()),
2137 sizeof(*evtchn_to_irq), GFP_KERNEL);
2138 BUG_ON(!evtchn_to_irq);
2140 /* No event channels are 'live' right now. */
2141 for (evtchn = 0; evtchn < xen_evtchn_nr_channels(); evtchn++)
2142 mask_evtchn(evtchn);
2144 pirq_needs_eoi = pirq_needs_eoi_flag;
2147 if (xen_pv_domain()) {
2148 if (xen_initial_domain())
2149 pci_xen_initial_domain();
2151 if (xen_feature(XENFEAT_hvm_callback_vector)) {
2152 xen_setup_callback_vector();
2153 xen_alloc_callback_vector();
2156 if (xen_hvm_domain()) {
2158 /* pci_xen_hvm_init must be called after native_init_IRQ so that
2159 * __acpi_register_gsi can point at the right function */
2163 struct physdev_pirq_eoi_gmfn eoi_gmfn;
2165 pirq_eoi_map = (void *)__get_free_page(GFP_KERNEL|__GFP_ZERO);
2166 eoi_gmfn.gmfn = virt_to_gfn(pirq_eoi_map);
2167 rc = HYPERVISOR_physdev_op(PHYSDEVOP_pirq_eoi_gmfn_v2, &eoi_gmfn);
2169 free_page((unsigned long) pirq_eoi_map);
2170 pirq_eoi_map = NULL;
2172 pirq_needs_eoi = pirq_check_eoi_map;