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 recieved, 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. Hardware interrupts. Not supported at present.
21 * Jeremy Fitzhardinge <jeremy@xensource.com>, XenSource Inc, 2007
24 #include <linux/linkage.h>
25 #include <linux/interrupt.h>
26 #include <linux/irq.h>
27 #include <linux/module.h>
28 #include <linux/string.h>
30 #include <asm/ptrace.h>
32 #include <asm/sync_bitops.h>
33 #include <asm/xen/hypercall.h>
34 #include <asm/xen/hypervisor.h>
36 #include <xen/xen-ops.h>
37 #include <xen/events.h>
38 #include <xen/interface/xen.h>
39 #include <xen/interface/event_channel.h>
42 * This lock protects updates to the following mapping and reference-count
43 * arrays. The lock does not need to be acquired to read the mapping tables.
45 static DEFINE_SPINLOCK(irq_mapping_update_lock);
47 /* IRQ <-> VIRQ mapping. */
48 static DEFINE_PER_CPU(int, virq_to_irq[NR_VIRQS]) = {[0 ... NR_VIRQS-1] = -1};
50 /* IRQ <-> IPI mapping */
51 static DEFINE_PER_CPU(int, ipi_to_irq[XEN_NR_IPIS]) = {[0 ... XEN_NR_IPIS-1] = -1};
53 /* Packed IRQ information: binding type, sub-type index, and event channel. */
56 unsigned short evtchn;
61 static struct packed_irq irq_info[NR_IRQS];
72 /* Convenient shorthand for packed representation of an unbound IRQ. */
73 #define IRQ_UNBOUND mk_irq_info(IRQT_UNBOUND, 0, 0)
75 static int evtchn_to_irq[NR_EVENT_CHANNELS] = {
76 [0 ... NR_EVENT_CHANNELS-1] = -1
78 static unsigned long cpu_evtchn_mask[NR_CPUS][NR_EVENT_CHANNELS/BITS_PER_LONG];
79 static u8 cpu_evtchn[NR_EVENT_CHANNELS];
81 /* Reference counts for bindings to IRQs. */
82 static int irq_bindcount[NR_IRQS];
84 /* Xen will never allocate port zero for any purpose. */
85 #define VALID_EVTCHN(chn) ((chn) != 0)
87 static struct irq_chip xen_dynamic_chip;
89 /* Constructor for packed IRQ information. */
90 static inline struct packed_irq mk_irq_info(u32 type, u32 index, u32 evtchn)
92 return (struct packed_irq) { evtchn, index, type };
96 * Accessors for packed IRQ information.
98 static inline unsigned int evtchn_from_irq(int irq)
100 return irq_info[irq].evtchn;
103 static inline unsigned int index_from_irq(int irq)
105 return irq_info[irq].index;
108 static inline unsigned int type_from_irq(int irq)
110 return irq_info[irq].type;
113 static inline unsigned long active_evtchns(unsigned int cpu,
114 struct shared_info *sh,
117 return (sh->evtchn_pending[idx] &
118 cpu_evtchn_mask[cpu][idx] &
119 ~sh->evtchn_mask[idx]);
122 static void bind_evtchn_to_cpu(unsigned int chn, unsigned int cpu)
124 int irq = evtchn_to_irq[chn];
128 irq_desc[irq].affinity = cpumask_of_cpu(cpu);
131 __clear_bit(chn, cpu_evtchn_mask[cpu_evtchn[chn]]);
132 __set_bit(chn, cpu_evtchn_mask[cpu]);
134 cpu_evtchn[chn] = cpu;
137 static void init_evtchn_cpu_bindings(void)
141 /* By default all event channels notify CPU#0. */
142 for (i = 0; i < NR_IRQS; i++)
143 irq_desc[i].affinity = cpumask_of_cpu(0);
146 memset(cpu_evtchn, 0, sizeof(cpu_evtchn));
147 memset(cpu_evtchn_mask[0], ~0, sizeof(cpu_evtchn_mask[0]));
150 static inline unsigned int cpu_from_evtchn(unsigned int evtchn)
152 return cpu_evtchn[evtchn];
155 static inline void clear_evtchn(int port)
157 struct shared_info *s = HYPERVISOR_shared_info;
158 sync_clear_bit(port, &s->evtchn_pending[0]);
161 static inline void set_evtchn(int port)
163 struct shared_info *s = HYPERVISOR_shared_info;
164 sync_set_bit(port, &s->evtchn_pending[0]);
169 * notify_remote_via_irq - send event to remote end of event channel via irq
170 * @irq: irq of event channel to send event to
172 * Unlike notify_remote_via_evtchn(), this is safe to use across
173 * save/restore. Notifications on a broken connection are silently
176 void notify_remote_via_irq(int irq)
178 int evtchn = evtchn_from_irq(irq);
180 if (VALID_EVTCHN(evtchn))
181 notify_remote_via_evtchn(evtchn);
183 EXPORT_SYMBOL_GPL(notify_remote_via_irq);
185 static void mask_evtchn(int port)
187 struct shared_info *s = HYPERVISOR_shared_info;
188 sync_set_bit(port, &s->evtchn_mask[0]);
191 static void unmask_evtchn(int port)
193 struct shared_info *s = HYPERVISOR_shared_info;
194 unsigned int cpu = get_cpu();
196 BUG_ON(!irqs_disabled());
198 /* Slow path (hypercall) if this is a non-local port. */
199 if (unlikely(cpu != cpu_from_evtchn(port))) {
200 struct evtchn_unmask unmask = { .port = port };
201 (void)HYPERVISOR_event_channel_op(EVTCHNOP_unmask, &unmask);
203 struct vcpu_info *vcpu_info = __get_cpu_var(xen_vcpu);
205 sync_clear_bit(port, &s->evtchn_mask[0]);
208 * The following is basically the equivalent of
209 * 'hw_resend_irq'. Just like a real IO-APIC we 'lose
210 * the interrupt edge' if the channel is masked.
212 if (sync_test_bit(port, &s->evtchn_pending[0]) &&
213 !sync_test_and_set_bit(port / BITS_PER_LONG,
214 &vcpu_info->evtchn_pending_sel))
215 vcpu_info->evtchn_upcall_pending = 1;
221 static int find_unbound_irq(void)
225 /* Only allocate from dynirq range */
226 for (irq = 0; irq < NR_IRQS; irq++)
227 if (irq_bindcount[irq] == 0)
231 panic("No available IRQ to bind to: increase NR_IRQS!\n");
236 int bind_evtchn_to_irq(unsigned int evtchn)
240 spin_lock(&irq_mapping_update_lock);
242 irq = evtchn_to_irq[evtchn];
245 irq = find_unbound_irq();
247 dynamic_irq_init(irq);
248 set_irq_chip_and_handler_name(irq, &xen_dynamic_chip,
249 handle_level_irq, "event");
251 evtchn_to_irq[evtchn] = irq;
252 irq_info[irq] = mk_irq_info(IRQT_EVTCHN, 0, evtchn);
255 irq_bindcount[irq]++;
257 spin_unlock(&irq_mapping_update_lock);
261 EXPORT_SYMBOL_GPL(bind_evtchn_to_irq);
263 static int bind_ipi_to_irq(unsigned int ipi, unsigned int cpu)
265 struct evtchn_bind_ipi bind_ipi;
268 spin_lock(&irq_mapping_update_lock);
270 irq = per_cpu(ipi_to_irq, cpu)[ipi];
272 irq = find_unbound_irq();
276 dynamic_irq_init(irq);
277 set_irq_chip_and_handler_name(irq, &xen_dynamic_chip,
278 handle_level_irq, "ipi");
281 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_ipi,
284 evtchn = bind_ipi.port;
286 evtchn_to_irq[evtchn] = irq;
287 irq_info[irq] = mk_irq_info(IRQT_IPI, ipi, evtchn);
289 per_cpu(ipi_to_irq, cpu)[ipi] = irq;
291 bind_evtchn_to_cpu(evtchn, cpu);
294 irq_bindcount[irq]++;
297 spin_unlock(&irq_mapping_update_lock);
302 static int bind_virq_to_irq(unsigned int virq, unsigned int cpu)
304 struct evtchn_bind_virq bind_virq;
307 spin_lock(&irq_mapping_update_lock);
309 irq = per_cpu(virq_to_irq, cpu)[virq];
312 bind_virq.virq = virq;
313 bind_virq.vcpu = cpu;
314 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_virq,
317 evtchn = bind_virq.port;
319 irq = find_unbound_irq();
321 dynamic_irq_init(irq);
322 set_irq_chip_and_handler_name(irq, &xen_dynamic_chip,
323 handle_level_irq, "virq");
325 evtchn_to_irq[evtchn] = irq;
326 irq_info[irq] = mk_irq_info(IRQT_VIRQ, virq, evtchn);
328 per_cpu(virq_to_irq, cpu)[virq] = irq;
330 bind_evtchn_to_cpu(evtchn, cpu);
333 irq_bindcount[irq]++;
335 spin_unlock(&irq_mapping_update_lock);
340 static void unbind_from_irq(unsigned int irq)
342 struct evtchn_close close;
343 int evtchn = evtchn_from_irq(irq);
345 spin_lock(&irq_mapping_update_lock);
347 if ((--irq_bindcount[irq] == 0) && VALID_EVTCHN(evtchn)) {
349 if (HYPERVISOR_event_channel_op(EVTCHNOP_close, &close) != 0)
352 switch (type_from_irq(irq)) {
354 per_cpu(virq_to_irq, cpu_from_evtchn(evtchn))
355 [index_from_irq(irq)] = -1;
361 /* Closed ports are implicitly re-bound to VCPU0. */
362 bind_evtchn_to_cpu(evtchn, 0);
364 evtchn_to_irq[evtchn] = -1;
365 irq_info[irq] = IRQ_UNBOUND;
367 dynamic_irq_cleanup(irq);
370 spin_unlock(&irq_mapping_update_lock);
373 int bind_evtchn_to_irqhandler(unsigned int evtchn,
374 irq_handler_t handler,
375 unsigned long irqflags,
376 const char *devname, void *dev_id)
381 irq = bind_evtchn_to_irq(evtchn);
382 retval = request_irq(irq, handler, irqflags, devname, dev_id);
384 unbind_from_irq(irq);
390 EXPORT_SYMBOL_GPL(bind_evtchn_to_irqhandler);
392 int bind_virq_to_irqhandler(unsigned int virq, unsigned int cpu,
393 irq_handler_t handler,
394 unsigned long irqflags, const char *devname, void *dev_id)
399 irq = bind_virq_to_irq(virq, cpu);
400 retval = request_irq(irq, handler, irqflags, devname, dev_id);
402 unbind_from_irq(irq);
408 EXPORT_SYMBOL_GPL(bind_virq_to_irqhandler);
410 int bind_ipi_to_irqhandler(enum ipi_vector ipi,
412 irq_handler_t handler,
413 unsigned long irqflags,
419 irq = bind_ipi_to_irq(ipi, cpu);
423 retval = request_irq(irq, handler, irqflags, devname, dev_id);
425 unbind_from_irq(irq);
432 void unbind_from_irqhandler(unsigned int irq, void *dev_id)
434 free_irq(irq, dev_id);
435 unbind_from_irq(irq);
437 EXPORT_SYMBOL_GPL(unbind_from_irqhandler);
439 void xen_send_IPI_one(unsigned int cpu, enum ipi_vector vector)
441 int irq = per_cpu(ipi_to_irq, cpu)[vector];
443 notify_remote_via_irq(irq);
446 irqreturn_t xen_debug_interrupt(int irq, void *dev_id)
448 struct shared_info *sh = HYPERVISOR_shared_info;
449 int cpu = smp_processor_id();
452 static DEFINE_SPINLOCK(debug_lock);
454 spin_lock_irqsave(&debug_lock, flags);
456 printk("vcpu %d\n ", cpu);
458 for_each_online_cpu(i) {
459 struct vcpu_info *v = per_cpu(xen_vcpu, i);
460 printk("%d: masked=%d pending=%d event_sel %08lx\n ", i,
461 (get_irq_regs() && i == cpu) ? xen_irqs_disabled(get_irq_regs()) : v->evtchn_upcall_mask,
462 v->evtchn_upcall_pending,
463 v->evtchn_pending_sel);
465 printk("pending:\n ");
466 for(i = ARRAY_SIZE(sh->evtchn_pending)-1; i >= 0; i--)
467 printk("%08lx%s", sh->evtchn_pending[i],
468 i % 8 == 0 ? "\n " : " ");
469 printk("\nmasks:\n ");
470 for(i = ARRAY_SIZE(sh->evtchn_mask)-1; i >= 0; i--)
471 printk("%08lx%s", sh->evtchn_mask[i],
472 i % 8 == 0 ? "\n " : " ");
474 printk("\nunmasked:\n ");
475 for(i = ARRAY_SIZE(sh->evtchn_mask)-1; i >= 0; i--)
476 printk("%08lx%s", sh->evtchn_pending[i] & ~sh->evtchn_mask[i],
477 i % 8 == 0 ? "\n " : " ");
479 printk("\npending list:\n");
480 for(i = 0; i < NR_EVENT_CHANNELS; i++) {
481 if (sync_test_bit(i, sh->evtchn_pending)) {
482 printk(" %d: event %d -> irq %d\n",
488 spin_unlock_irqrestore(&debug_lock, flags);
495 * Search the CPUs pending events bitmasks. For each one found, map
496 * the event number to an irq, and feed it into do_IRQ() for
499 * Xen uses a two-level bitmap to speed searching. The first level is
500 * a bitset of words which contain pending event bits. The second
501 * level is a bitset of pending events themselves.
503 void xen_evtchn_do_upcall(struct pt_regs *regs)
506 struct shared_info *s = HYPERVISOR_shared_info;
507 struct vcpu_info *vcpu_info = __get_cpu_var(xen_vcpu);
508 static DEFINE_PER_CPU(unsigned, nesting_count);
512 unsigned long pending_words;
514 vcpu_info->evtchn_upcall_pending = 0;
516 if (__get_cpu_var(nesting_count)++)
519 #ifndef CONFIG_X86 /* No need for a barrier -- XCHG is a barrier on x86. */
520 /* Clear master flag /before/ clearing selector flag. */
523 pending_words = xchg(&vcpu_info->evtchn_pending_sel, 0);
524 while (pending_words != 0) {
525 unsigned long pending_bits;
526 int word_idx = __ffs(pending_words);
527 pending_words &= ~(1UL << word_idx);
529 while ((pending_bits = active_evtchns(cpu, s, word_idx)) != 0) {
530 int bit_idx = __ffs(pending_bits);
531 int port = (word_idx * BITS_PER_LONG) + bit_idx;
532 int irq = evtchn_to_irq[port];
535 xen_do_IRQ(irq, regs);
539 BUG_ON(!irqs_disabled());
541 count = __get_cpu_var(nesting_count);
542 __get_cpu_var(nesting_count) = 0;
549 /* Rebind a new event channel to an existing irq. */
550 void rebind_evtchn_irq(int evtchn, int irq)
552 /* Make sure the irq is masked, since the new event channel
553 will also be masked. */
556 spin_lock(&irq_mapping_update_lock);
558 /* After resume the irq<->evtchn mappings are all cleared out */
559 BUG_ON(evtchn_to_irq[evtchn] != -1);
560 /* Expect irq to have been bound before,
561 so the bindcount should be non-0 */
562 BUG_ON(irq_bindcount[irq] == 0);
564 evtchn_to_irq[evtchn] = irq;
565 irq_info[irq] = mk_irq_info(IRQT_EVTCHN, 0, evtchn);
567 spin_unlock(&irq_mapping_update_lock);
569 /* new event channels are always bound to cpu 0 */
570 irq_set_affinity(irq, cpumask_of_cpu(0));
572 /* Unmask the event channel. */
576 /* Rebind an evtchn so that it gets delivered to a specific cpu */
577 static void rebind_irq_to_cpu(unsigned irq, unsigned tcpu)
579 struct evtchn_bind_vcpu bind_vcpu;
580 int evtchn = evtchn_from_irq(irq);
582 if (!VALID_EVTCHN(evtchn))
585 /* Send future instances of this interrupt to other vcpu. */
586 bind_vcpu.port = evtchn;
587 bind_vcpu.vcpu = tcpu;
590 * If this fails, it usually just indicates that we're dealing with a
591 * virq or IPI channel, which don't actually need to be rebound. Ignore
592 * it, but don't do the xenlinux-level rebind in that case.
594 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_vcpu, &bind_vcpu) >= 0)
595 bind_evtchn_to_cpu(evtchn, tcpu);
599 static void set_affinity_irq(unsigned irq, cpumask_t dest)
601 unsigned tcpu = first_cpu(dest);
602 rebind_irq_to_cpu(irq, tcpu);
605 int resend_irq_on_evtchn(unsigned int irq)
607 int masked, evtchn = evtchn_from_irq(irq);
608 struct shared_info *s = HYPERVISOR_shared_info;
610 if (!VALID_EVTCHN(evtchn))
613 masked = sync_test_and_set_bit(evtchn, s->evtchn_mask);
614 sync_set_bit(evtchn, s->evtchn_pending);
616 unmask_evtchn(evtchn);
621 static void enable_dynirq(unsigned int irq)
623 int evtchn = evtchn_from_irq(irq);
625 if (VALID_EVTCHN(evtchn))
626 unmask_evtchn(evtchn);
629 static void disable_dynirq(unsigned int irq)
631 int evtchn = evtchn_from_irq(irq);
633 if (VALID_EVTCHN(evtchn))
637 static void ack_dynirq(unsigned int irq)
639 int evtchn = evtchn_from_irq(irq);
641 move_native_irq(irq);
643 if (VALID_EVTCHN(evtchn))
644 clear_evtchn(evtchn);
647 static int retrigger_dynirq(unsigned int irq)
649 int evtchn = evtchn_from_irq(irq);
650 struct shared_info *sh = HYPERVISOR_shared_info;
653 if (VALID_EVTCHN(evtchn)) {
656 masked = sync_test_and_set_bit(evtchn, sh->evtchn_mask);
657 sync_set_bit(evtchn, sh->evtchn_pending);
659 unmask_evtchn(evtchn);
666 static void restore_cpu_virqs(unsigned int cpu)
668 struct evtchn_bind_virq bind_virq;
669 int virq, irq, evtchn;
671 for (virq = 0; virq < NR_VIRQS; virq++) {
672 if ((irq = per_cpu(virq_to_irq, cpu)[virq]) == -1)
675 BUG_ON(irq_info[irq].type != IRQT_VIRQ);
676 BUG_ON(irq_info[irq].index != virq);
678 /* Get a new binding from Xen. */
679 bind_virq.virq = virq;
680 bind_virq.vcpu = cpu;
681 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_virq,
684 evtchn = bind_virq.port;
686 /* Record the new mapping. */
687 evtchn_to_irq[evtchn] = irq;
688 irq_info[irq] = mk_irq_info(IRQT_VIRQ, virq, evtchn);
689 bind_evtchn_to_cpu(evtchn, cpu);
692 unmask_evtchn(evtchn);
696 static void restore_cpu_ipis(unsigned int cpu)
698 struct evtchn_bind_ipi bind_ipi;
699 int ipi, irq, evtchn;
701 for (ipi = 0; ipi < XEN_NR_IPIS; ipi++) {
702 if ((irq = per_cpu(ipi_to_irq, cpu)[ipi]) == -1)
705 BUG_ON(irq_info[irq].type != IRQT_IPI);
706 BUG_ON(irq_info[irq].index != ipi);
708 /* Get a new binding from Xen. */
710 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_ipi,
713 evtchn = bind_ipi.port;
715 /* Record the new mapping. */
716 evtchn_to_irq[evtchn] = irq;
717 irq_info[irq] = mk_irq_info(IRQT_IPI, ipi, evtchn);
718 bind_evtchn_to_cpu(evtchn, cpu);
721 unmask_evtchn(evtchn);
726 /* Clear an irq's pending state, in preparation for polling on it */
727 void xen_clear_irq_pending(int irq)
729 int evtchn = evtchn_from_irq(irq);
731 if (VALID_EVTCHN(evtchn))
732 clear_evtchn(evtchn);
735 /* Poll waiting for an irq to become pending. In the usual case, the
736 irq will be disabled so it won't deliver an interrupt. */
737 void xen_poll_irq(int irq)
739 evtchn_port_t evtchn = evtchn_from_irq(irq);
741 if (VALID_EVTCHN(evtchn)) {
742 struct sched_poll poll;
746 poll.ports = &evtchn;
748 if (HYPERVISOR_sched_op(SCHEDOP_poll, &poll) != 0)
753 void xen_irq_resume(void)
755 unsigned int cpu, irq, evtchn;
757 init_evtchn_cpu_bindings();
759 /* New event-channel space is not 'live' yet. */
760 for (evtchn = 0; evtchn < NR_EVENT_CHANNELS; evtchn++)
763 /* No IRQ <-> event-channel mappings. */
764 for (irq = 0; irq < NR_IRQS; irq++)
765 irq_info[irq].evtchn = 0; /* zap event-channel binding */
767 for (evtchn = 0; evtchn < NR_EVENT_CHANNELS; evtchn++)
768 evtchn_to_irq[evtchn] = -1;
770 for_each_possible_cpu(cpu) {
771 restore_cpu_virqs(cpu);
772 restore_cpu_ipis(cpu);
776 static struct irq_chip xen_dynamic_chip __read_mostly = {
778 .mask = disable_dynirq,
779 .unmask = enable_dynirq,
781 .set_affinity = set_affinity_irq,
782 .retrigger = retrigger_dynirq,
785 void __init xen_init_IRQ(void)
789 init_evtchn_cpu_bindings();
791 /* No event channels are 'live' right now. */
792 for (i = 0; i < NR_EVENT_CHANNELS; i++)
795 /* Dynamic IRQ space is currently unbound. Zero the refcnts. */
796 for (i = 0; i < NR_IRQS; i++)
797 irq_bindcount[i] = 0;
799 irq_ctx_init(smp_processor_id());