ce602dd09bc18c60381c68f0f1310bb168817bed
[safe/jmp/linux-2.6] / drivers / xen / events.c
1 /*
2  * Xen event channels
3  *
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.
10  *
11  * There are four kinds of events which can be mapped to an event
12  * channel:
13  *
14  * 1. Inter-domain notifications.  This includes all the virtual
15  *    device events, since they're driven by front-ends in another domain
16  *    (typically dom0).
17  * 2. VIRQs, typically used for timers.  These are per-cpu events.
18  * 3. IPIs.
19  * 4. Hardware interrupts. Not supported at present.
20  *
21  * Jeremy Fitzhardinge <jeremy@xensource.com>, XenSource Inc, 2007
22  */
23
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>
29 #include <linux/bootmem.h>
30
31 #include <asm/ptrace.h>
32 #include <asm/irq.h>
33 #include <asm/idle.h>
34 #include <asm/sync_bitops.h>
35 #include <asm/xen/hypercall.h>
36 #include <asm/xen/hypervisor.h>
37
38 #include <xen/xen-ops.h>
39 #include <xen/events.h>
40 #include <xen/interface/xen.h>
41 #include <xen/interface/event_channel.h>
42
43 /*
44  * This lock protects updates to the following mapping and reference-count
45  * arrays. The lock does not need to be acquired to read the mapping tables.
46  */
47 static DEFINE_SPINLOCK(irq_mapping_update_lock);
48
49 /* IRQ <-> VIRQ mapping. */
50 static DEFINE_PER_CPU(int [NR_VIRQS], virq_to_irq) = {[0 ... NR_VIRQS-1] = -1};
51
52 /* IRQ <-> IPI mapping */
53 static DEFINE_PER_CPU(int [XEN_NR_IPIS], ipi_to_irq) = {[0 ... XEN_NR_IPIS-1] = -1};
54
55 /* Interrupt types. */
56 enum xen_irq_type {
57         IRQT_UNBOUND = 0,
58         IRQT_PIRQ,
59         IRQT_VIRQ,
60         IRQT_IPI,
61         IRQT_EVTCHN
62 };
63
64 /*
65  * Packed IRQ information:
66  * type - enum xen_irq_type
67  * event channel - irq->event channel mapping
68  * cpu - cpu this event channel is bound to
69  * index - type-specific information:
70  *    PIRQ - vector, with MSB being "needs EIO"
71  *    VIRQ - virq number
72  *    IPI - IPI vector
73  *    EVTCHN -
74  */
75 struct irq_info
76 {
77         enum xen_irq_type type; /* type */
78         unsigned short evtchn;  /* event channel */
79         unsigned short cpu;     /* cpu bound */
80
81         union {
82                 unsigned short virq;
83                 enum ipi_vector ipi;
84                 struct {
85                         unsigned short gsi;
86                         unsigned short vector;
87                 } pirq;
88         } u;
89 };
90
91 static struct irq_info irq_info[NR_IRQS];
92
93 static int evtchn_to_irq[NR_EVENT_CHANNELS] = {
94         [0 ... NR_EVENT_CHANNELS-1] = -1
95 };
96 struct cpu_evtchn_s {
97         unsigned long bits[NR_EVENT_CHANNELS/BITS_PER_LONG];
98 };
99 static struct cpu_evtchn_s *cpu_evtchn_mask_p;
100 static inline unsigned long *cpu_evtchn_mask(int cpu)
101 {
102         return cpu_evtchn_mask_p[cpu].bits;
103 }
104
105 /* Xen will never allocate port zero for any purpose. */
106 #define VALID_EVTCHN(chn)       ((chn) != 0)
107
108 static struct irq_chip xen_dynamic_chip;
109
110 /* Constructor for packed IRQ information. */
111 static struct irq_info mk_unbound_info(void)
112 {
113         return (struct irq_info) { .type = IRQT_UNBOUND };
114 }
115
116 static struct irq_info mk_evtchn_info(unsigned short evtchn)
117 {
118         return (struct irq_info) { .type = IRQT_EVTCHN, .evtchn = evtchn,
119                         .cpu = 0 };
120 }
121
122 static struct irq_info mk_ipi_info(unsigned short evtchn, enum ipi_vector ipi)
123 {
124         return (struct irq_info) { .type = IRQT_IPI, .evtchn = evtchn,
125                         .cpu = 0, .u.ipi = ipi };
126 }
127
128 static struct irq_info mk_virq_info(unsigned short evtchn, unsigned short virq)
129 {
130         return (struct irq_info) { .type = IRQT_VIRQ, .evtchn = evtchn,
131                         .cpu = 0, .u.virq = virq };
132 }
133
134 static struct irq_info mk_pirq_info(unsigned short evtchn,
135                                     unsigned short gsi, unsigned short vector)
136 {
137         return (struct irq_info) { .type = IRQT_PIRQ, .evtchn = evtchn,
138                         .cpu = 0, .u.pirq = { .gsi = gsi, .vector = vector } };
139 }
140
141 /*
142  * Accessors for packed IRQ information.
143  */
144 static struct irq_info *info_for_irq(unsigned irq)
145 {
146         return &irq_info[irq];
147 }
148
149 static unsigned int evtchn_from_irq(unsigned irq)
150 {
151         return info_for_irq(irq)->evtchn;
152 }
153
154 unsigned irq_from_evtchn(unsigned int evtchn)
155 {
156         return evtchn_to_irq[evtchn];
157 }
158 EXPORT_SYMBOL_GPL(irq_from_evtchn);
159
160 static enum ipi_vector ipi_from_irq(unsigned irq)
161 {
162         struct irq_info *info = info_for_irq(irq);
163
164         BUG_ON(info == NULL);
165         BUG_ON(info->type != IRQT_IPI);
166
167         return info->u.ipi;
168 }
169
170 static unsigned virq_from_irq(unsigned irq)
171 {
172         struct irq_info *info = info_for_irq(irq);
173
174         BUG_ON(info == NULL);
175         BUG_ON(info->type != IRQT_VIRQ);
176
177         return info->u.virq;
178 }
179
180 static unsigned gsi_from_irq(unsigned irq)
181 {
182         struct irq_info *info = info_for_irq(irq);
183
184         BUG_ON(info == NULL);
185         BUG_ON(info->type != IRQT_PIRQ);
186
187         return info->u.pirq.gsi;
188 }
189
190 static unsigned vector_from_irq(unsigned irq)
191 {
192         struct irq_info *info = info_for_irq(irq);
193
194         BUG_ON(info == NULL);
195         BUG_ON(info->type != IRQT_PIRQ);
196
197         return info->u.pirq.vector;
198 }
199
200 static enum xen_irq_type type_from_irq(unsigned irq)
201 {
202         return info_for_irq(irq)->type;
203 }
204
205 static unsigned cpu_from_irq(unsigned irq)
206 {
207         return info_for_irq(irq)->cpu;
208 }
209
210 static unsigned int cpu_from_evtchn(unsigned int evtchn)
211 {
212         int irq = evtchn_to_irq[evtchn];
213         unsigned ret = 0;
214
215         if (irq != -1)
216                 ret = cpu_from_irq(irq);
217
218         return ret;
219 }
220
221 static inline unsigned long active_evtchns(unsigned int cpu,
222                                            struct shared_info *sh,
223                                            unsigned int idx)
224 {
225         return (sh->evtchn_pending[idx] &
226                 cpu_evtchn_mask(cpu)[idx] &
227                 ~sh->evtchn_mask[idx]);
228 }
229
230 static void bind_evtchn_to_cpu(unsigned int chn, unsigned int cpu)
231 {
232         int irq = evtchn_to_irq[chn];
233
234         BUG_ON(irq == -1);
235 #ifdef CONFIG_SMP
236         cpumask_copy(irq_to_desc(irq)->affinity, cpumask_of(cpu));
237 #endif
238
239         __clear_bit(chn, cpu_evtchn_mask(cpu_from_irq(irq)));
240         __set_bit(chn, cpu_evtchn_mask(cpu));
241
242         irq_info[irq].cpu = cpu;
243 }
244
245 static void init_evtchn_cpu_bindings(void)
246 {
247 #ifdef CONFIG_SMP
248         struct irq_desc *desc;
249         int i;
250
251         /* By default all event channels notify CPU#0. */
252         for_each_irq_desc(i, desc) {
253                 cpumask_copy(desc->affinity, cpumask_of(0));
254         }
255 #endif
256
257         memset(cpu_evtchn_mask(0), ~0, sizeof(cpu_evtchn_mask(0)));
258 }
259
260 static inline void clear_evtchn(int port)
261 {
262         struct shared_info *s = HYPERVISOR_shared_info;
263         sync_clear_bit(port, &s->evtchn_pending[0]);
264 }
265
266 static inline void set_evtchn(int port)
267 {
268         struct shared_info *s = HYPERVISOR_shared_info;
269         sync_set_bit(port, &s->evtchn_pending[0]);
270 }
271
272 static inline int test_evtchn(int port)
273 {
274         struct shared_info *s = HYPERVISOR_shared_info;
275         return sync_test_bit(port, &s->evtchn_pending[0]);
276 }
277
278
279 /**
280  * notify_remote_via_irq - send event to remote end of event channel via irq
281  * @irq: irq of event channel to send event to
282  *
283  * Unlike notify_remote_via_evtchn(), this is safe to use across
284  * save/restore. Notifications on a broken connection are silently
285  * dropped.
286  */
287 void notify_remote_via_irq(int irq)
288 {
289         int evtchn = evtchn_from_irq(irq);
290
291         if (VALID_EVTCHN(evtchn))
292                 notify_remote_via_evtchn(evtchn);
293 }
294 EXPORT_SYMBOL_GPL(notify_remote_via_irq);
295
296 static void mask_evtchn(int port)
297 {
298         struct shared_info *s = HYPERVISOR_shared_info;
299         sync_set_bit(port, &s->evtchn_mask[0]);
300 }
301
302 static void unmask_evtchn(int port)
303 {
304         struct shared_info *s = HYPERVISOR_shared_info;
305         unsigned int cpu = get_cpu();
306
307         BUG_ON(!irqs_disabled());
308
309         /* Slow path (hypercall) if this is a non-local port. */
310         if (unlikely(cpu != cpu_from_evtchn(port))) {
311                 struct evtchn_unmask unmask = { .port = port };
312                 (void)HYPERVISOR_event_channel_op(EVTCHNOP_unmask, &unmask);
313         } else {
314                 struct vcpu_info *vcpu_info = __get_cpu_var(xen_vcpu);
315
316                 sync_clear_bit(port, &s->evtchn_mask[0]);
317
318                 /*
319                  * The following is basically the equivalent of
320                  * 'hw_resend_irq'. Just like a real IO-APIC we 'lose
321                  * the interrupt edge' if the channel is masked.
322                  */
323                 if (sync_test_bit(port, &s->evtchn_pending[0]) &&
324                     !sync_test_and_set_bit(port / BITS_PER_LONG,
325                                            &vcpu_info->evtchn_pending_sel))
326                         vcpu_info->evtchn_upcall_pending = 1;
327         }
328
329         put_cpu();
330 }
331
332 static int find_unbound_irq(void)
333 {
334         int irq;
335         struct irq_desc *desc;
336
337         for (irq = 0; irq < nr_irqs; irq++)
338                 if (irq_info[irq].type == IRQT_UNBOUND)
339                         break;
340
341         if (irq == nr_irqs)
342                 panic("No available IRQ to bind to: increase nr_irqs!\n");
343
344         desc = irq_to_desc_alloc_node(irq, 0);
345         if (WARN_ON(desc == NULL))
346                 return -1;
347
348         dynamic_irq_init(irq);
349
350         return irq;
351 }
352
353 int bind_evtchn_to_irq(unsigned int evtchn)
354 {
355         int irq;
356
357         spin_lock(&irq_mapping_update_lock);
358
359         irq = evtchn_to_irq[evtchn];
360
361         if (irq == -1) {
362                 irq = find_unbound_irq();
363
364                 set_irq_chip_and_handler_name(irq, &xen_dynamic_chip,
365                                               handle_level_irq, "event");
366
367                 evtchn_to_irq[evtchn] = irq;
368                 irq_info[irq] = mk_evtchn_info(evtchn);
369         }
370
371         spin_unlock(&irq_mapping_update_lock);
372
373         return irq;
374 }
375 EXPORT_SYMBOL_GPL(bind_evtchn_to_irq);
376
377 static int bind_ipi_to_irq(unsigned int ipi, unsigned int cpu)
378 {
379         struct evtchn_bind_ipi bind_ipi;
380         int evtchn, irq;
381
382         spin_lock(&irq_mapping_update_lock);
383
384         irq = per_cpu(ipi_to_irq, cpu)[ipi];
385
386         if (irq == -1) {
387                 irq = find_unbound_irq();
388                 if (irq < 0)
389                         goto out;
390
391                 set_irq_chip_and_handler_name(irq, &xen_dynamic_chip,
392                                               handle_level_irq, "ipi");
393
394                 bind_ipi.vcpu = cpu;
395                 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_ipi,
396                                                 &bind_ipi) != 0)
397                         BUG();
398                 evtchn = bind_ipi.port;
399
400                 evtchn_to_irq[evtchn] = irq;
401                 irq_info[irq] = mk_ipi_info(evtchn, ipi);
402                 per_cpu(ipi_to_irq, cpu)[ipi] = irq;
403
404                 bind_evtchn_to_cpu(evtchn, cpu);
405         }
406
407  out:
408         spin_unlock(&irq_mapping_update_lock);
409         return irq;
410 }
411
412
413 static int bind_virq_to_irq(unsigned int virq, unsigned int cpu)
414 {
415         struct evtchn_bind_virq bind_virq;
416         int evtchn, irq;
417
418         spin_lock(&irq_mapping_update_lock);
419
420         irq = per_cpu(virq_to_irq, cpu)[virq];
421
422         if (irq == -1) {
423                 bind_virq.virq = virq;
424                 bind_virq.vcpu = cpu;
425                 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_virq,
426                                                 &bind_virq) != 0)
427                         BUG();
428                 evtchn = bind_virq.port;
429
430                 irq = find_unbound_irq();
431
432                 set_irq_chip_and_handler_name(irq, &xen_dynamic_chip,
433                                               handle_level_irq, "virq");
434
435                 evtchn_to_irq[evtchn] = irq;
436                 irq_info[irq] = mk_virq_info(evtchn, virq);
437
438                 per_cpu(virq_to_irq, cpu)[virq] = irq;
439
440                 bind_evtchn_to_cpu(evtchn, cpu);
441         }
442
443         spin_unlock(&irq_mapping_update_lock);
444
445         return irq;
446 }
447
448 static void unbind_from_irq(unsigned int irq)
449 {
450         struct evtchn_close close;
451         int evtchn = evtchn_from_irq(irq);
452
453         spin_lock(&irq_mapping_update_lock);
454
455         if (VALID_EVTCHN(evtchn)) {
456                 close.port = evtchn;
457                 if (HYPERVISOR_event_channel_op(EVTCHNOP_close, &close) != 0)
458                         BUG();
459
460                 switch (type_from_irq(irq)) {
461                 case IRQT_VIRQ:
462                         per_cpu(virq_to_irq, cpu_from_evtchn(evtchn))
463                                 [virq_from_irq(irq)] = -1;
464                         break;
465                 case IRQT_IPI:
466                         per_cpu(ipi_to_irq, cpu_from_evtchn(evtchn))
467                                 [ipi_from_irq(irq)] = -1;
468                         break;
469                 default:
470                         break;
471                 }
472
473                 /* Closed ports are implicitly re-bound to VCPU0. */
474                 bind_evtchn_to_cpu(evtchn, 0);
475
476                 evtchn_to_irq[evtchn] = -1;
477         }
478
479         if (irq_info[irq].type != IRQT_UNBOUND) {
480                 irq_info[irq] = mk_unbound_info();
481
482                 dynamic_irq_cleanup(irq);
483         }
484
485         spin_unlock(&irq_mapping_update_lock);
486 }
487
488 int bind_evtchn_to_irqhandler(unsigned int evtchn,
489                               irq_handler_t handler,
490                               unsigned long irqflags,
491                               const char *devname, void *dev_id)
492 {
493         unsigned int irq;
494         int retval;
495
496         irq = bind_evtchn_to_irq(evtchn);
497         retval = request_irq(irq, handler, irqflags, devname, dev_id);
498         if (retval != 0) {
499                 unbind_from_irq(irq);
500                 return retval;
501         }
502
503         return irq;
504 }
505 EXPORT_SYMBOL_GPL(bind_evtchn_to_irqhandler);
506
507 int bind_virq_to_irqhandler(unsigned int virq, unsigned int cpu,
508                             irq_handler_t handler,
509                             unsigned long irqflags, const char *devname, void *dev_id)
510 {
511         unsigned int irq;
512         int retval;
513
514         irq = bind_virq_to_irq(virq, cpu);
515         retval = request_irq(irq, handler, irqflags, devname, dev_id);
516         if (retval != 0) {
517                 unbind_from_irq(irq);
518                 return retval;
519         }
520
521         return irq;
522 }
523 EXPORT_SYMBOL_GPL(bind_virq_to_irqhandler);
524
525 int bind_ipi_to_irqhandler(enum ipi_vector ipi,
526                            unsigned int cpu,
527                            irq_handler_t handler,
528                            unsigned long irqflags,
529                            const char *devname,
530                            void *dev_id)
531 {
532         int irq, retval;
533
534         irq = bind_ipi_to_irq(ipi, cpu);
535         if (irq < 0)
536                 return irq;
537
538         retval = request_irq(irq, handler, irqflags, devname, dev_id);
539         if (retval != 0) {
540                 unbind_from_irq(irq);
541                 return retval;
542         }
543
544         return irq;
545 }
546
547 void unbind_from_irqhandler(unsigned int irq, void *dev_id)
548 {
549         free_irq(irq, dev_id);
550         unbind_from_irq(irq);
551 }
552 EXPORT_SYMBOL_GPL(unbind_from_irqhandler);
553
554 void xen_send_IPI_one(unsigned int cpu, enum ipi_vector vector)
555 {
556         int irq = per_cpu(ipi_to_irq, cpu)[vector];
557         BUG_ON(irq < 0);
558         notify_remote_via_irq(irq);
559 }
560
561 irqreturn_t xen_debug_interrupt(int irq, void *dev_id)
562 {
563         struct shared_info *sh = HYPERVISOR_shared_info;
564         int cpu = smp_processor_id();
565         int i;
566         unsigned long flags;
567         static DEFINE_SPINLOCK(debug_lock);
568
569         spin_lock_irqsave(&debug_lock, flags);
570
571         printk("vcpu %d\n  ", cpu);
572
573         for_each_online_cpu(i) {
574                 struct vcpu_info *v = per_cpu(xen_vcpu, i);
575                 printk("%d: masked=%d pending=%d event_sel %08lx\n  ", i,
576                         (get_irq_regs() && i == cpu) ? xen_irqs_disabled(get_irq_regs()) : v->evtchn_upcall_mask,
577                         v->evtchn_upcall_pending,
578                         v->evtchn_pending_sel);
579         }
580         printk("pending:\n   ");
581         for(i = ARRAY_SIZE(sh->evtchn_pending)-1; i >= 0; i--)
582                 printk("%08lx%s", sh->evtchn_pending[i],
583                         i % 8 == 0 ? "\n   " : " ");
584         printk("\nmasks:\n   ");
585         for(i = ARRAY_SIZE(sh->evtchn_mask)-1; i >= 0; i--)
586                 printk("%08lx%s", sh->evtchn_mask[i],
587                         i % 8 == 0 ? "\n   " : " ");
588
589         printk("\nunmasked:\n   ");
590         for(i = ARRAY_SIZE(sh->evtchn_mask)-1; i >= 0; i--)
591                 printk("%08lx%s", sh->evtchn_pending[i] & ~sh->evtchn_mask[i],
592                         i % 8 == 0 ? "\n   " : " ");
593
594         printk("\npending list:\n");
595         for(i = 0; i < NR_EVENT_CHANNELS; i++) {
596                 if (sync_test_bit(i, sh->evtchn_pending)) {
597                         printk("  %d: event %d -> irq %d\n",
598                                cpu_from_evtchn(i), i,
599                                evtchn_to_irq[i]);
600                 }
601         }
602
603         spin_unlock_irqrestore(&debug_lock, flags);
604
605         return IRQ_HANDLED;
606 }
607
608 static DEFINE_PER_CPU(unsigned, xed_nesting_count);
609
610 /*
611  * Search the CPUs pending events bitmasks.  For each one found, map
612  * the event number to an irq, and feed it into do_IRQ() for
613  * handling.
614  *
615  * Xen uses a two-level bitmap to speed searching.  The first level is
616  * a bitset of words which contain pending event bits.  The second
617  * level is a bitset of pending events themselves.
618  */
619 void xen_evtchn_do_upcall(struct pt_regs *regs)
620 {
621         int cpu = get_cpu();
622         struct pt_regs *old_regs = set_irq_regs(regs);
623         struct shared_info *s = HYPERVISOR_shared_info;
624         struct vcpu_info *vcpu_info = __get_cpu_var(xen_vcpu);
625         unsigned count;
626
627         exit_idle();
628         irq_enter();
629
630         do {
631                 unsigned long pending_words;
632
633                 vcpu_info->evtchn_upcall_pending = 0;
634
635                 if (__get_cpu_var(xed_nesting_count)++)
636                         goto out;
637
638 #ifndef CONFIG_X86 /* No need for a barrier -- XCHG is a barrier on x86. */
639                 /* Clear master flag /before/ clearing selector flag. */
640                 wmb();
641 #endif
642                 pending_words = xchg(&vcpu_info->evtchn_pending_sel, 0);
643                 while (pending_words != 0) {
644                         unsigned long pending_bits;
645                         int word_idx = __ffs(pending_words);
646                         pending_words &= ~(1UL << word_idx);
647
648                         while ((pending_bits = active_evtchns(cpu, s, word_idx)) != 0) {
649                                 int bit_idx = __ffs(pending_bits);
650                                 int port = (word_idx * BITS_PER_LONG) + bit_idx;
651                                 int irq = evtchn_to_irq[port];
652
653                                 if (irq != -1)
654                                         handle_irq(irq, regs);
655                         }
656                 }
657
658                 BUG_ON(!irqs_disabled());
659
660                 count = __get_cpu_var(xed_nesting_count);
661                 __get_cpu_var(xed_nesting_count) = 0;
662         } while(count != 1);
663
664 out:
665         irq_exit();
666         set_irq_regs(old_regs);
667
668         put_cpu();
669 }
670
671 /* Rebind a new event channel to an existing irq. */
672 void rebind_evtchn_irq(int evtchn, int irq)
673 {
674         struct irq_info *info = info_for_irq(irq);
675
676         /* Make sure the irq is masked, since the new event channel
677            will also be masked. */
678         disable_irq(irq);
679
680         spin_lock(&irq_mapping_update_lock);
681
682         /* After resume the irq<->evtchn mappings are all cleared out */
683         BUG_ON(evtchn_to_irq[evtchn] != -1);
684         /* Expect irq to have been bound before,
685            so there should be a proper type */
686         BUG_ON(info->type == IRQT_UNBOUND);
687
688         evtchn_to_irq[evtchn] = irq;
689         irq_info[irq] = mk_evtchn_info(evtchn);
690
691         spin_unlock(&irq_mapping_update_lock);
692
693         /* new event channels are always bound to cpu 0 */
694         irq_set_affinity(irq, cpumask_of(0));
695
696         /* Unmask the event channel. */
697         enable_irq(irq);
698 }
699
700 /* Rebind an evtchn so that it gets delivered to a specific cpu */
701 static int rebind_irq_to_cpu(unsigned irq, unsigned tcpu)
702 {
703         struct evtchn_bind_vcpu bind_vcpu;
704         int evtchn = evtchn_from_irq(irq);
705
706         if (!VALID_EVTCHN(evtchn))
707                 return -1;
708
709         /* Send future instances of this interrupt to other vcpu. */
710         bind_vcpu.port = evtchn;
711         bind_vcpu.vcpu = tcpu;
712
713         /*
714          * If this fails, it usually just indicates that we're dealing with a
715          * virq or IPI channel, which don't actually need to be rebound. Ignore
716          * it, but don't do the xenlinux-level rebind in that case.
717          */
718         if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_vcpu, &bind_vcpu) >= 0)
719                 bind_evtchn_to_cpu(evtchn, tcpu);
720
721         return 0;
722 }
723
724 static int set_affinity_irq(unsigned irq, const struct cpumask *dest)
725 {
726         unsigned tcpu = cpumask_first(dest);
727
728         return rebind_irq_to_cpu(irq, tcpu);
729 }
730
731 int resend_irq_on_evtchn(unsigned int irq)
732 {
733         int masked, evtchn = evtchn_from_irq(irq);
734         struct shared_info *s = HYPERVISOR_shared_info;
735
736         if (!VALID_EVTCHN(evtchn))
737                 return 1;
738
739         masked = sync_test_and_set_bit(evtchn, s->evtchn_mask);
740         sync_set_bit(evtchn, s->evtchn_pending);
741         if (!masked)
742                 unmask_evtchn(evtchn);
743
744         return 1;
745 }
746
747 static void enable_dynirq(unsigned int irq)
748 {
749         int evtchn = evtchn_from_irq(irq);
750
751         if (VALID_EVTCHN(evtchn))
752                 unmask_evtchn(evtchn);
753 }
754
755 static void disable_dynirq(unsigned int irq)
756 {
757         int evtchn = evtchn_from_irq(irq);
758
759         if (VALID_EVTCHN(evtchn))
760                 mask_evtchn(evtchn);
761 }
762
763 static void ack_dynirq(unsigned int irq)
764 {
765         int evtchn = evtchn_from_irq(irq);
766
767         move_native_irq(irq);
768
769         if (VALID_EVTCHN(evtchn))
770                 clear_evtchn(evtchn);
771 }
772
773 static int retrigger_dynirq(unsigned int irq)
774 {
775         int evtchn = evtchn_from_irq(irq);
776         struct shared_info *sh = HYPERVISOR_shared_info;
777         int ret = 0;
778
779         if (VALID_EVTCHN(evtchn)) {
780                 int masked;
781
782                 masked = sync_test_and_set_bit(evtchn, sh->evtchn_mask);
783                 sync_set_bit(evtchn, sh->evtchn_pending);
784                 if (!masked)
785                         unmask_evtchn(evtchn);
786                 ret = 1;
787         }
788
789         return ret;
790 }
791
792 static void restore_cpu_virqs(unsigned int cpu)
793 {
794         struct evtchn_bind_virq bind_virq;
795         int virq, irq, evtchn;
796
797         for (virq = 0; virq < NR_VIRQS; virq++) {
798                 if ((irq = per_cpu(virq_to_irq, cpu)[virq]) == -1)
799                         continue;
800
801                 BUG_ON(virq_from_irq(irq) != virq);
802
803                 /* Get a new binding from Xen. */
804                 bind_virq.virq = virq;
805                 bind_virq.vcpu = cpu;
806                 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_virq,
807                                                 &bind_virq) != 0)
808                         BUG();
809                 evtchn = bind_virq.port;
810
811                 /* Record the new mapping. */
812                 evtchn_to_irq[evtchn] = irq;
813                 irq_info[irq] = mk_virq_info(evtchn, virq);
814                 bind_evtchn_to_cpu(evtchn, cpu);
815
816                 /* Ready for use. */
817                 unmask_evtchn(evtchn);
818         }
819 }
820
821 static void restore_cpu_ipis(unsigned int cpu)
822 {
823         struct evtchn_bind_ipi bind_ipi;
824         int ipi, irq, evtchn;
825
826         for (ipi = 0; ipi < XEN_NR_IPIS; ipi++) {
827                 if ((irq = per_cpu(ipi_to_irq, cpu)[ipi]) == -1)
828                         continue;
829
830                 BUG_ON(ipi_from_irq(irq) != ipi);
831
832                 /* Get a new binding from Xen. */
833                 bind_ipi.vcpu = cpu;
834                 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_ipi,
835                                                 &bind_ipi) != 0)
836                         BUG();
837                 evtchn = bind_ipi.port;
838
839                 /* Record the new mapping. */
840                 evtchn_to_irq[evtchn] = irq;
841                 irq_info[irq] = mk_ipi_info(evtchn, ipi);
842                 bind_evtchn_to_cpu(evtchn, cpu);
843
844                 /* Ready for use. */
845                 unmask_evtchn(evtchn);
846
847         }
848 }
849
850 /* Clear an irq's pending state, in preparation for polling on it */
851 void xen_clear_irq_pending(int irq)
852 {
853         int evtchn = evtchn_from_irq(irq);
854
855         if (VALID_EVTCHN(evtchn))
856                 clear_evtchn(evtchn);
857 }
858
859 void xen_set_irq_pending(int irq)
860 {
861         int evtchn = evtchn_from_irq(irq);
862
863         if (VALID_EVTCHN(evtchn))
864                 set_evtchn(evtchn);
865 }
866
867 bool xen_test_irq_pending(int irq)
868 {
869         int evtchn = evtchn_from_irq(irq);
870         bool ret = false;
871
872         if (VALID_EVTCHN(evtchn))
873                 ret = test_evtchn(evtchn);
874
875         return ret;
876 }
877
878 /* Poll waiting for an irq to become pending.  In the usual case, the
879    irq will be disabled so it won't deliver an interrupt. */
880 void xen_poll_irq(int irq)
881 {
882         evtchn_port_t evtchn = evtchn_from_irq(irq);
883
884         if (VALID_EVTCHN(evtchn)) {
885                 struct sched_poll poll;
886
887                 poll.nr_ports = 1;
888                 poll.timeout = 0;
889                 set_xen_guest_handle(poll.ports, &evtchn);
890
891                 if (HYPERVISOR_sched_op(SCHEDOP_poll, &poll) != 0)
892                         BUG();
893         }
894 }
895
896 void xen_irq_resume(void)
897 {
898         unsigned int cpu, irq, evtchn;
899
900         init_evtchn_cpu_bindings();
901
902         /* New event-channel space is not 'live' yet. */
903         for (evtchn = 0; evtchn < NR_EVENT_CHANNELS; evtchn++)
904                 mask_evtchn(evtchn);
905
906         /* No IRQ <-> event-channel mappings. */
907         for (irq = 0; irq < nr_irqs; irq++)
908                 irq_info[irq].evtchn = 0; /* zap event-channel binding */
909
910         for (evtchn = 0; evtchn < NR_EVENT_CHANNELS; evtchn++)
911                 evtchn_to_irq[evtchn] = -1;
912
913         for_each_possible_cpu(cpu) {
914                 restore_cpu_virqs(cpu);
915                 restore_cpu_ipis(cpu);
916         }
917 }
918
919 static struct irq_chip xen_dynamic_chip __read_mostly = {
920         .name           = "xen-dyn",
921
922         .disable        = disable_dynirq,
923         .mask           = disable_dynirq,
924         .unmask         = enable_dynirq,
925
926         .ack            = ack_dynirq,
927         .set_affinity   = set_affinity_irq,
928         .retrigger      = retrigger_dynirq,
929 };
930
931 void __init xen_init_IRQ(void)
932 {
933         int i;
934
935         cpu_evtchn_mask_p = kcalloc(nr_cpu_ids, sizeof(struct cpu_evtchn_s),
936                                     GFP_KERNEL);
937         BUG_ON(cpu_evtchn_mask_p == NULL);
938
939         init_evtchn_cpu_bindings();
940
941         /* No event channels are 'live' right now. */
942         for (i = 0; i < NR_EVENT_CHANNELS; i++)
943                 mask_evtchn(i);
944
945         irq_ctx_init(smp_processor_id());
946 }