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)
88 * Force a proper event-channel callback from Xen after clearing the
89 * callback mask. We do this in a very simple manner, by making a call
90 * down into Xen. The pending flag will be checked by Xen on return.
92 void force_evtchn_callback(void)
94 (void)HYPERVISOR_xen_version(0, NULL);
96 EXPORT_SYMBOL_GPL(force_evtchn_callback);
98 static struct irq_chip xen_dynamic_chip;
100 /* Constructor for packed IRQ information. */
101 static inline struct packed_irq mk_irq_info(u32 type, u32 index, u32 evtchn)
103 return (struct packed_irq) { evtchn, index, type };
107 * Accessors for packed IRQ information.
109 static inline unsigned int evtchn_from_irq(int irq)
111 return irq_info[irq].evtchn;
114 static inline unsigned int index_from_irq(int irq)
116 return irq_info[irq].index;
119 static inline unsigned int type_from_irq(int irq)
121 return irq_info[irq].type;
124 static inline unsigned long active_evtchns(unsigned int cpu,
125 struct shared_info *sh,
128 return (sh->evtchn_pending[idx] &
129 cpu_evtchn_mask[cpu][idx] &
130 ~sh->evtchn_mask[idx]);
133 static void bind_evtchn_to_cpu(unsigned int chn, unsigned int cpu)
135 int irq = evtchn_to_irq[chn];
139 irq_desc[irq].affinity = cpumask_of_cpu(cpu);
142 __clear_bit(chn, cpu_evtchn_mask[cpu_evtchn[chn]]);
143 __set_bit(chn, cpu_evtchn_mask[cpu]);
145 cpu_evtchn[chn] = cpu;
148 static void init_evtchn_cpu_bindings(void)
152 /* By default all event channels notify CPU#0. */
153 for (i = 0; i < NR_IRQS; i++)
154 irq_desc[i].affinity = cpumask_of_cpu(0);
157 memset(cpu_evtchn, 0, sizeof(cpu_evtchn));
158 memset(cpu_evtchn_mask[0], ~0, sizeof(cpu_evtchn_mask[0]));
161 static inline unsigned int cpu_from_evtchn(unsigned int evtchn)
163 return cpu_evtchn[evtchn];
166 static inline void clear_evtchn(int port)
168 struct shared_info *s = HYPERVISOR_shared_info;
169 sync_clear_bit(port, &s->evtchn_pending[0]);
172 static inline void set_evtchn(int port)
174 struct shared_info *s = HYPERVISOR_shared_info;
175 sync_set_bit(port, &s->evtchn_pending[0]);
180 * notify_remote_via_irq - send event to remote end of event channel via irq
181 * @irq: irq of event channel to send event to
183 * Unlike notify_remote_via_evtchn(), this is safe to use across
184 * save/restore. Notifications on a broken connection are silently
187 void notify_remote_via_irq(int irq)
189 int evtchn = evtchn_from_irq(irq);
191 if (VALID_EVTCHN(evtchn))
192 notify_remote_via_evtchn(evtchn);
194 EXPORT_SYMBOL_GPL(notify_remote_via_irq);
196 static void mask_evtchn(int port)
198 struct shared_info *s = HYPERVISOR_shared_info;
199 sync_set_bit(port, &s->evtchn_mask[0]);
202 static void unmask_evtchn(int port)
204 struct shared_info *s = HYPERVISOR_shared_info;
205 unsigned int cpu = get_cpu();
207 BUG_ON(!irqs_disabled());
209 /* Slow path (hypercall) if this is a non-local port. */
210 if (unlikely(cpu != cpu_from_evtchn(port))) {
211 struct evtchn_unmask unmask = { .port = port };
212 (void)HYPERVISOR_event_channel_op(EVTCHNOP_unmask, &unmask);
214 struct vcpu_info *vcpu_info = __get_cpu_var(xen_vcpu);
216 sync_clear_bit(port, &s->evtchn_mask[0]);
219 * The following is basically the equivalent of
220 * 'hw_resend_irq'. Just like a real IO-APIC we 'lose
221 * the interrupt edge' if the channel is masked.
223 if (sync_test_bit(port, &s->evtchn_pending[0]) &&
224 !sync_test_and_set_bit(port / BITS_PER_LONG,
225 &vcpu_info->evtchn_pending_sel))
226 vcpu_info->evtchn_upcall_pending = 1;
232 static int find_unbound_irq(void)
236 /* Only allocate from dynirq range */
237 for (irq = 0; irq < NR_IRQS; irq++)
238 if (irq_bindcount[irq] == 0)
242 panic("No available IRQ to bind to: increase NR_IRQS!\n");
247 int bind_evtchn_to_irq(unsigned int evtchn)
251 spin_lock(&irq_mapping_update_lock);
253 irq = evtchn_to_irq[evtchn];
256 irq = find_unbound_irq();
258 dynamic_irq_init(irq);
259 set_irq_chip_and_handler_name(irq, &xen_dynamic_chip,
260 handle_level_irq, "event");
262 evtchn_to_irq[evtchn] = irq;
263 irq_info[irq] = mk_irq_info(IRQT_EVTCHN, 0, evtchn);
266 irq_bindcount[irq]++;
268 spin_unlock(&irq_mapping_update_lock);
272 EXPORT_SYMBOL_GPL(bind_evtchn_to_irq);
274 static int bind_ipi_to_irq(unsigned int ipi, unsigned int cpu)
276 struct evtchn_bind_ipi bind_ipi;
279 spin_lock(&irq_mapping_update_lock);
281 irq = per_cpu(ipi_to_irq, cpu)[ipi];
283 irq = find_unbound_irq();
287 dynamic_irq_init(irq);
288 set_irq_chip_and_handler_name(irq, &xen_dynamic_chip,
289 handle_level_irq, "ipi");
292 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_ipi,
295 evtchn = bind_ipi.port;
297 evtchn_to_irq[evtchn] = irq;
298 irq_info[irq] = mk_irq_info(IRQT_IPI, ipi, evtchn);
300 per_cpu(ipi_to_irq, cpu)[ipi] = irq;
302 bind_evtchn_to_cpu(evtchn, cpu);
305 irq_bindcount[irq]++;
308 spin_unlock(&irq_mapping_update_lock);
313 static int bind_virq_to_irq(unsigned int virq, unsigned int cpu)
315 struct evtchn_bind_virq bind_virq;
318 spin_lock(&irq_mapping_update_lock);
320 irq = per_cpu(virq_to_irq, cpu)[virq];
323 bind_virq.virq = virq;
324 bind_virq.vcpu = cpu;
325 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_virq,
328 evtchn = bind_virq.port;
330 irq = find_unbound_irq();
332 dynamic_irq_init(irq);
333 set_irq_chip_and_handler_name(irq, &xen_dynamic_chip,
334 handle_level_irq, "virq");
336 evtchn_to_irq[evtchn] = irq;
337 irq_info[irq] = mk_irq_info(IRQT_VIRQ, virq, evtchn);
339 per_cpu(virq_to_irq, cpu)[virq] = irq;
341 bind_evtchn_to_cpu(evtchn, cpu);
344 irq_bindcount[irq]++;
346 spin_unlock(&irq_mapping_update_lock);
351 static void unbind_from_irq(unsigned int irq)
353 struct evtchn_close close;
354 int evtchn = evtchn_from_irq(irq);
356 spin_lock(&irq_mapping_update_lock);
358 if ((--irq_bindcount[irq] == 0) && VALID_EVTCHN(evtchn)) {
360 if (HYPERVISOR_event_channel_op(EVTCHNOP_close, &close) != 0)
363 switch (type_from_irq(irq)) {
365 per_cpu(virq_to_irq, cpu_from_evtchn(evtchn))
366 [index_from_irq(irq)] = -1;
372 /* Closed ports are implicitly re-bound to VCPU0. */
373 bind_evtchn_to_cpu(evtchn, 0);
375 evtchn_to_irq[evtchn] = -1;
376 irq_info[irq] = IRQ_UNBOUND;
378 dynamic_irq_cleanup(irq);
381 spin_unlock(&irq_mapping_update_lock);
384 int bind_evtchn_to_irqhandler(unsigned int evtchn,
385 irq_handler_t handler,
386 unsigned long irqflags,
387 const char *devname, void *dev_id)
392 irq = bind_evtchn_to_irq(evtchn);
393 retval = request_irq(irq, handler, irqflags, devname, dev_id);
395 unbind_from_irq(irq);
401 EXPORT_SYMBOL_GPL(bind_evtchn_to_irqhandler);
403 int bind_virq_to_irqhandler(unsigned int virq, unsigned int cpu,
404 irq_handler_t handler,
405 unsigned long irqflags, const char *devname, void *dev_id)
410 irq = bind_virq_to_irq(virq, cpu);
411 retval = request_irq(irq, handler, irqflags, devname, dev_id);
413 unbind_from_irq(irq);
419 EXPORT_SYMBOL_GPL(bind_virq_to_irqhandler);
421 int bind_ipi_to_irqhandler(enum ipi_vector ipi,
423 irq_handler_t handler,
424 unsigned long irqflags,
430 irq = bind_ipi_to_irq(ipi, cpu);
434 retval = request_irq(irq, handler, irqflags, devname, dev_id);
436 unbind_from_irq(irq);
443 void unbind_from_irqhandler(unsigned int irq, void *dev_id)
445 free_irq(irq, dev_id);
446 unbind_from_irq(irq);
448 EXPORT_SYMBOL_GPL(unbind_from_irqhandler);
450 void xen_send_IPI_one(unsigned int cpu, enum ipi_vector vector)
452 int irq = per_cpu(ipi_to_irq, cpu)[vector];
454 notify_remote_via_irq(irq);
457 irqreturn_t xen_debug_interrupt(int irq, void *dev_id)
459 struct shared_info *sh = HYPERVISOR_shared_info;
460 int cpu = smp_processor_id();
463 static DEFINE_SPINLOCK(debug_lock);
465 spin_lock_irqsave(&debug_lock, flags);
467 printk("vcpu %d\n ", cpu);
469 for_each_online_cpu(i) {
470 struct vcpu_info *v = per_cpu(xen_vcpu, i);
471 printk("%d: masked=%d pending=%d event_sel %08lx\n ", i,
472 (get_irq_regs() && i == cpu) ? xen_irqs_disabled(get_irq_regs()) : v->evtchn_upcall_mask,
473 v->evtchn_upcall_pending,
474 v->evtchn_pending_sel);
476 printk("pending:\n ");
477 for(i = ARRAY_SIZE(sh->evtchn_pending)-1; i >= 0; i--)
478 printk("%08lx%s", sh->evtchn_pending[i],
479 i % 8 == 0 ? "\n " : " ");
480 printk("\nmasks:\n ");
481 for(i = ARRAY_SIZE(sh->evtchn_mask)-1; i >= 0; i--)
482 printk("%08lx%s", sh->evtchn_mask[i],
483 i % 8 == 0 ? "\n " : " ");
485 printk("\nunmasked:\n ");
486 for(i = ARRAY_SIZE(sh->evtchn_mask)-1; i >= 0; i--)
487 printk("%08lx%s", sh->evtchn_pending[i] & ~sh->evtchn_mask[i],
488 i % 8 == 0 ? "\n " : " ");
490 printk("\npending list:\n");
491 for(i = 0; i < NR_EVENT_CHANNELS; i++) {
492 if (sync_test_bit(i, sh->evtchn_pending)) {
493 printk(" %d: event %d -> irq %d\n",
499 spin_unlock_irqrestore(&debug_lock, flags);
506 * Search the CPUs pending events bitmasks. For each one found, map
507 * the event number to an irq, and feed it into do_IRQ() for
510 * Xen uses a two-level bitmap to speed searching. The first level is
511 * a bitset of words which contain pending event bits. The second
512 * level is a bitset of pending events themselves.
514 void xen_evtchn_do_upcall(struct pt_regs *regs)
517 struct shared_info *s = HYPERVISOR_shared_info;
518 struct vcpu_info *vcpu_info = __get_cpu_var(xen_vcpu);
519 static DEFINE_PER_CPU(unsigned, nesting_count);
523 unsigned long pending_words;
525 vcpu_info->evtchn_upcall_pending = 0;
527 if (__get_cpu_var(nesting_count)++)
530 #ifndef CONFIG_X86 /* No need for a barrier -- XCHG is a barrier on x86. */
531 /* Clear master flag /before/ clearing selector flag. */
534 pending_words = xchg(&vcpu_info->evtchn_pending_sel, 0);
535 while (pending_words != 0) {
536 unsigned long pending_bits;
537 int word_idx = __ffs(pending_words);
538 pending_words &= ~(1UL << word_idx);
540 while ((pending_bits = active_evtchns(cpu, s, word_idx)) != 0) {
541 int bit_idx = __ffs(pending_bits);
542 int port = (word_idx * BITS_PER_LONG) + bit_idx;
543 int irq = evtchn_to_irq[port];
546 xen_do_IRQ(irq, regs);
550 BUG_ON(!irqs_disabled());
552 count = __get_cpu_var(nesting_count);
553 __get_cpu_var(nesting_count) = 0;
560 /* Rebind a new event channel to an existing irq. */
561 void rebind_evtchn_irq(int evtchn, int irq)
563 /* Make sure the irq is masked, since the new event channel
564 will also be masked. */
567 spin_lock(&irq_mapping_update_lock);
569 /* After resume the irq<->evtchn mappings are all cleared out */
570 BUG_ON(evtchn_to_irq[evtchn] != -1);
571 /* Expect irq to have been bound before,
572 so the bindcount should be non-0 */
573 BUG_ON(irq_bindcount[irq] == 0);
575 evtchn_to_irq[evtchn] = irq;
576 irq_info[irq] = mk_irq_info(IRQT_EVTCHN, 0, evtchn);
578 spin_unlock(&irq_mapping_update_lock);
580 /* new event channels are always bound to cpu 0 */
581 irq_set_affinity(irq, cpumask_of_cpu(0));
583 /* Unmask the event channel. */
587 /* Rebind an evtchn so that it gets delivered to a specific cpu */
588 static void rebind_irq_to_cpu(unsigned irq, unsigned tcpu)
590 struct evtchn_bind_vcpu bind_vcpu;
591 int evtchn = evtchn_from_irq(irq);
593 if (!VALID_EVTCHN(evtchn))
596 /* Send future instances of this interrupt to other vcpu. */
597 bind_vcpu.port = evtchn;
598 bind_vcpu.vcpu = tcpu;
601 * If this fails, it usually just indicates that we're dealing with a
602 * virq or IPI channel, which don't actually need to be rebound. Ignore
603 * it, but don't do the xenlinux-level rebind in that case.
605 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_vcpu, &bind_vcpu) >= 0)
606 bind_evtchn_to_cpu(evtchn, tcpu);
610 static void set_affinity_irq(unsigned irq, cpumask_t dest)
612 unsigned tcpu = first_cpu(dest);
613 rebind_irq_to_cpu(irq, tcpu);
616 int resend_irq_on_evtchn(unsigned int irq)
618 int masked, evtchn = evtchn_from_irq(irq);
619 struct shared_info *s = HYPERVISOR_shared_info;
621 if (!VALID_EVTCHN(evtchn))
624 masked = sync_test_and_set_bit(evtchn, s->evtchn_mask);
625 sync_set_bit(evtchn, s->evtchn_pending);
627 unmask_evtchn(evtchn);
632 static void enable_dynirq(unsigned int irq)
634 int evtchn = evtchn_from_irq(irq);
636 if (VALID_EVTCHN(evtchn))
637 unmask_evtchn(evtchn);
640 static void disable_dynirq(unsigned int irq)
642 int evtchn = evtchn_from_irq(irq);
644 if (VALID_EVTCHN(evtchn))
648 static void ack_dynirq(unsigned int irq)
650 int evtchn = evtchn_from_irq(irq);
652 move_native_irq(irq);
654 if (VALID_EVTCHN(evtchn))
655 clear_evtchn(evtchn);
658 static int retrigger_dynirq(unsigned int irq)
660 int evtchn = evtchn_from_irq(irq);
661 struct shared_info *sh = HYPERVISOR_shared_info;
664 if (VALID_EVTCHN(evtchn)) {
667 masked = sync_test_and_set_bit(evtchn, sh->evtchn_mask);
668 sync_set_bit(evtchn, sh->evtchn_pending);
670 unmask_evtchn(evtchn);
677 static void restore_cpu_virqs(unsigned int cpu)
679 struct evtchn_bind_virq bind_virq;
680 int virq, irq, evtchn;
682 for (virq = 0; virq < NR_VIRQS; virq++) {
683 if ((irq = per_cpu(virq_to_irq, cpu)[virq]) == -1)
686 BUG_ON(irq_info[irq].type != IRQT_VIRQ);
687 BUG_ON(irq_info[irq].index != virq);
689 /* Get a new binding from Xen. */
690 bind_virq.virq = virq;
691 bind_virq.vcpu = cpu;
692 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_virq,
695 evtchn = bind_virq.port;
697 /* Record the new mapping. */
698 evtchn_to_irq[evtchn] = irq;
699 irq_info[irq] = mk_irq_info(IRQT_VIRQ, virq, evtchn);
700 bind_evtchn_to_cpu(evtchn, cpu);
703 unmask_evtchn(evtchn);
707 static void restore_cpu_ipis(unsigned int cpu)
709 struct evtchn_bind_ipi bind_ipi;
710 int ipi, irq, evtchn;
712 for (ipi = 0; ipi < XEN_NR_IPIS; ipi++) {
713 if ((irq = per_cpu(ipi_to_irq, cpu)[ipi]) == -1)
716 BUG_ON(irq_info[irq].type != IRQT_IPI);
717 BUG_ON(irq_info[irq].index != ipi);
719 /* Get a new binding from Xen. */
721 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_ipi,
724 evtchn = bind_ipi.port;
726 /* Record the new mapping. */
727 evtchn_to_irq[evtchn] = irq;
728 irq_info[irq] = mk_irq_info(IRQT_IPI, ipi, evtchn);
729 bind_evtchn_to_cpu(evtchn, cpu);
732 unmask_evtchn(evtchn);
737 /* Clear an irq's pending state, in preparation for polling on it */
738 void xen_clear_irq_pending(int irq)
740 int evtchn = evtchn_from_irq(irq);
742 if (VALID_EVTCHN(evtchn))
743 clear_evtchn(evtchn);
746 /* Poll waiting for an irq to become pending. In the usual case, the
747 irq will be disabled so it won't deliver an interrupt. */
748 void xen_poll_irq(int irq)
750 evtchn_port_t evtchn = evtchn_from_irq(irq);
752 if (VALID_EVTCHN(evtchn)) {
753 struct sched_poll poll;
757 poll.ports = &evtchn;
759 if (HYPERVISOR_sched_op(SCHEDOP_poll, &poll) != 0)
764 void xen_irq_resume(void)
766 unsigned int cpu, irq, evtchn;
768 init_evtchn_cpu_bindings();
770 /* New event-channel space is not 'live' yet. */
771 for (evtchn = 0; evtchn < NR_EVENT_CHANNELS; evtchn++)
774 /* No IRQ <-> event-channel mappings. */
775 for (irq = 0; irq < NR_IRQS; irq++)
776 irq_info[irq].evtchn = 0; /* zap event-channel binding */
778 for (evtchn = 0; evtchn < NR_EVENT_CHANNELS; evtchn++)
779 evtchn_to_irq[evtchn] = -1;
781 for_each_possible_cpu(cpu) {
782 restore_cpu_virqs(cpu);
783 restore_cpu_ipis(cpu);
787 static struct irq_chip xen_dynamic_chip __read_mostly = {
789 .mask = disable_dynirq,
790 .unmask = enable_dynirq,
792 .set_affinity = set_affinity_irq,
793 .retrigger = retrigger_dynirq,
796 void __init xen_init_IRQ(void)
800 init_evtchn_cpu_bindings();
802 /* No event channels are 'live' right now. */
803 for (i = 0; i < NR_EVENT_CHANNELS; i++)
806 /* Dynamic IRQ space is currently unbound. Zero the refcnts. */
807 for (i = 0; i < NR_IRQS; i++)
808 irq_bindcount[i] = 0;
810 irq_ctx_init(smp_processor_id());