975e90fa6d5aba1650a7bd3a5cf9543e6e26ef79
[pandora-kernel.git] / 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. PIRQs - Hardware interrupts.
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 #include <linux/slab.h>
31 #include <linux/irqnr.h>
32 #include <linux/pci.h>
33
34 #include <asm/desc.h>
35 #include <asm/ptrace.h>
36 #include <asm/irq.h>
37 #include <asm/idle.h>
38 #include <asm/io_apic.h>
39 #include <asm/sync_bitops.h>
40 #include <asm/xen/pci.h>
41 #include <asm/xen/hypercall.h>
42 #include <asm/xen/hypervisor.h>
43
44 #include <xen/xen.h>
45 #include <xen/hvm.h>
46 #include <xen/xen-ops.h>
47 #include <xen/events.h>
48 #include <xen/interface/xen.h>
49 #include <xen/interface/event_channel.h>
50 #include <xen/interface/hvm/hvm_op.h>
51 #include <xen/interface/hvm/params.h>
52
53 /*
54  * This lock protects updates to the following mapping and reference-count
55  * arrays. The lock does not need to be acquired to read the mapping tables.
56  */
57 static DEFINE_SPINLOCK(irq_mapping_update_lock);
58
59 /* IRQ <-> VIRQ mapping. */
60 static DEFINE_PER_CPU(int [NR_VIRQS], virq_to_irq) = {[0 ... NR_VIRQS-1] = -1};
61
62 /* IRQ <-> IPI mapping */
63 static DEFINE_PER_CPU(int [XEN_NR_IPIS], ipi_to_irq) = {[0 ... XEN_NR_IPIS-1] = -1};
64
65 /* Interrupt types. */
66 enum xen_irq_type {
67         IRQT_UNBOUND = 0,
68         IRQT_PIRQ,
69         IRQT_VIRQ,
70         IRQT_IPI,
71         IRQT_EVTCHN
72 };
73
74 /*
75  * Packed IRQ information:
76  * type - enum xen_irq_type
77  * event channel - irq->event channel mapping
78  * cpu - cpu this event channel is bound to
79  * index - type-specific information:
80  *    PIRQ - vector, with MSB being "needs EIO", or physical IRQ of the HVM
81  *           guest, or GSI (real passthrough IRQ) of the device.
82  *    VIRQ - virq number
83  *    IPI - IPI vector
84  *    EVTCHN -
85  */
86 struct irq_info
87 {
88         enum xen_irq_type type; /* type */
89         unsigned short evtchn;  /* event channel */
90         unsigned short cpu;     /* cpu bound */
91
92         union {
93                 unsigned short virq;
94                 enum ipi_vector ipi;
95                 struct {
96                         unsigned short pirq;
97                         unsigned short gsi;
98                         unsigned char vector;
99                         unsigned char flags;
100                 } pirq;
101         } u;
102 };
103 #define PIRQ_NEEDS_EOI  (1 << 0)
104 #define PIRQ_SHAREABLE  (1 << 1)
105
106 static struct irq_info *irq_info;
107 static int *pirq_to_irq;
108
109 static int *evtchn_to_irq;
110 struct cpu_evtchn_s {
111         unsigned long bits[NR_EVENT_CHANNELS/BITS_PER_LONG];
112 };
113
114 static __initdata struct cpu_evtchn_s init_evtchn_mask = {
115         .bits[0 ... (NR_EVENT_CHANNELS/BITS_PER_LONG)-1] = ~0ul,
116 };
117 static struct cpu_evtchn_s *cpu_evtchn_mask_p = &init_evtchn_mask;
118
119 static inline unsigned long *cpu_evtchn_mask(int cpu)
120 {
121         return cpu_evtchn_mask_p[cpu].bits;
122 }
123
124 /* Xen will never allocate port zero for any purpose. */
125 #define VALID_EVTCHN(chn)       ((chn) != 0)
126
127 static struct irq_chip xen_dynamic_chip;
128 static struct irq_chip xen_percpu_chip;
129 static struct irq_chip xen_pirq_chip;
130
131 /* Constructor for packed IRQ information. */
132 static struct irq_info mk_unbound_info(void)
133 {
134         return (struct irq_info) { .type = IRQT_UNBOUND };
135 }
136
137 static struct irq_info mk_evtchn_info(unsigned short evtchn)
138 {
139         return (struct irq_info) { .type = IRQT_EVTCHN, .evtchn = evtchn,
140                         .cpu = 0 };
141 }
142
143 static struct irq_info mk_ipi_info(unsigned short evtchn, enum ipi_vector ipi)
144 {
145         return (struct irq_info) { .type = IRQT_IPI, .evtchn = evtchn,
146                         .cpu = 0, .u.ipi = ipi };
147 }
148
149 static struct irq_info mk_virq_info(unsigned short evtchn, unsigned short virq)
150 {
151         return (struct irq_info) { .type = IRQT_VIRQ, .evtchn = evtchn,
152                         .cpu = 0, .u.virq = virq };
153 }
154
155 static struct irq_info mk_pirq_info(unsigned short evtchn, unsigned short pirq,
156                                     unsigned short gsi, unsigned short vector)
157 {
158         return (struct irq_info) { .type = IRQT_PIRQ, .evtchn = evtchn,
159                         .cpu = 0,
160                         .u.pirq = { .pirq = pirq, .gsi = gsi, .vector = vector } };
161 }
162
163 /*
164  * Accessors for packed IRQ information.
165  */
166 static struct irq_info *info_for_irq(unsigned irq)
167 {
168         return &irq_info[irq];
169 }
170
171 static unsigned int evtchn_from_irq(unsigned irq)
172 {
173         if (unlikely(WARN(irq < 0 || irq >= nr_irqs, "Invalid irq %d!\n", irq)))
174                 return 0;
175
176         return info_for_irq(irq)->evtchn;
177 }
178
179 unsigned irq_from_evtchn(unsigned int evtchn)
180 {
181         return evtchn_to_irq[evtchn];
182 }
183 EXPORT_SYMBOL_GPL(irq_from_evtchn);
184
185 static enum ipi_vector ipi_from_irq(unsigned irq)
186 {
187         struct irq_info *info = info_for_irq(irq);
188
189         BUG_ON(info == NULL);
190         BUG_ON(info->type != IRQT_IPI);
191
192         return info->u.ipi;
193 }
194
195 static unsigned virq_from_irq(unsigned irq)
196 {
197         struct irq_info *info = info_for_irq(irq);
198
199         BUG_ON(info == NULL);
200         BUG_ON(info->type != IRQT_VIRQ);
201
202         return info->u.virq;
203 }
204
205 static unsigned pirq_from_irq(unsigned irq)
206 {
207         struct irq_info *info = info_for_irq(irq);
208
209         BUG_ON(info == NULL);
210         BUG_ON(info->type != IRQT_PIRQ);
211
212         return info->u.pirq.pirq;
213 }
214
215 static unsigned gsi_from_irq(unsigned irq)
216 {
217         struct irq_info *info = info_for_irq(irq);
218
219         BUG_ON(info == NULL);
220         BUG_ON(info->type != IRQT_PIRQ);
221
222         return info->u.pirq.gsi;
223 }
224
225 static unsigned vector_from_irq(unsigned irq)
226 {
227         struct irq_info *info = info_for_irq(irq);
228
229         BUG_ON(info == NULL);
230         BUG_ON(info->type != IRQT_PIRQ);
231
232         return info->u.pirq.vector;
233 }
234
235 static enum xen_irq_type type_from_irq(unsigned irq)
236 {
237         return info_for_irq(irq)->type;
238 }
239
240 static unsigned cpu_from_irq(unsigned irq)
241 {
242         return info_for_irq(irq)->cpu;
243 }
244
245 static unsigned int cpu_from_evtchn(unsigned int evtchn)
246 {
247         int irq = evtchn_to_irq[evtchn];
248         unsigned ret = 0;
249
250         if (irq != -1)
251                 ret = cpu_from_irq(irq);
252
253         return ret;
254 }
255
256 static bool pirq_needs_eoi(unsigned irq)
257 {
258         struct irq_info *info = info_for_irq(irq);
259
260         BUG_ON(info->type != IRQT_PIRQ);
261
262         return info->u.pirq.flags & PIRQ_NEEDS_EOI;
263 }
264
265 static inline unsigned long active_evtchns(unsigned int cpu,
266                                            struct shared_info *sh,
267                                            unsigned int idx)
268 {
269         return (sh->evtchn_pending[idx] &
270                 cpu_evtchn_mask(cpu)[idx] &
271                 ~sh->evtchn_mask[idx]);
272 }
273
274 static void bind_evtchn_to_cpu(unsigned int chn, unsigned int cpu)
275 {
276         int irq = evtchn_to_irq[chn];
277
278         BUG_ON(irq == -1);
279 #ifdef CONFIG_SMP
280         cpumask_copy(irq_to_desc(irq)->irq_data.affinity, cpumask_of(cpu));
281 #endif
282
283         clear_bit(chn, cpu_evtchn_mask(cpu_from_irq(irq)));
284         set_bit(chn, cpu_evtchn_mask(cpu));
285
286         irq_info[irq].cpu = cpu;
287 }
288
289 static void init_evtchn_cpu_bindings(void)
290 {
291         int i;
292 #ifdef CONFIG_SMP
293         struct irq_desc *desc;
294
295         /* By default all event channels notify CPU#0. */
296         for_each_irq_desc(i, desc) {
297                 cpumask_copy(desc->irq_data.affinity, cpumask_of(0));
298         }
299 #endif
300
301         for_each_possible_cpu(i)
302                 memset(cpu_evtchn_mask(i),
303                        (i == 0) ? ~0 : 0, sizeof(struct cpu_evtchn_s));
304
305 }
306
307 static inline void clear_evtchn(int port)
308 {
309         struct shared_info *s = HYPERVISOR_shared_info;
310         sync_clear_bit(port, &s->evtchn_pending[0]);
311 }
312
313 static inline void set_evtchn(int port)
314 {
315         struct shared_info *s = HYPERVISOR_shared_info;
316         sync_set_bit(port, &s->evtchn_pending[0]);
317 }
318
319 static inline int test_evtchn(int port)
320 {
321         struct shared_info *s = HYPERVISOR_shared_info;
322         return sync_test_bit(port, &s->evtchn_pending[0]);
323 }
324
325
326 /**
327  * notify_remote_via_irq - send event to remote end of event channel via irq
328  * @irq: irq of event channel to send event to
329  *
330  * Unlike notify_remote_via_evtchn(), this is safe to use across
331  * save/restore. Notifications on a broken connection are silently
332  * dropped.
333  */
334 void notify_remote_via_irq(int irq)
335 {
336         int evtchn = evtchn_from_irq(irq);
337
338         if (VALID_EVTCHN(evtchn))
339                 notify_remote_via_evtchn(evtchn);
340 }
341 EXPORT_SYMBOL_GPL(notify_remote_via_irq);
342
343 static void mask_evtchn(int port)
344 {
345         struct shared_info *s = HYPERVISOR_shared_info;
346         sync_set_bit(port, &s->evtchn_mask[0]);
347 }
348
349 static void unmask_evtchn(int port)
350 {
351         struct shared_info *s = HYPERVISOR_shared_info;
352         unsigned int cpu = get_cpu();
353
354         BUG_ON(!irqs_disabled());
355
356         /* Slow path (hypercall) if this is a non-local port. */
357         if (unlikely(cpu != cpu_from_evtchn(port))) {
358                 struct evtchn_unmask unmask = { .port = port };
359                 (void)HYPERVISOR_event_channel_op(EVTCHNOP_unmask, &unmask);
360         } else {
361                 struct vcpu_info *vcpu_info = __this_cpu_read(xen_vcpu);
362
363                 sync_clear_bit(port, &s->evtchn_mask[0]);
364
365                 /*
366                  * The following is basically the equivalent of
367                  * 'hw_resend_irq'. Just like a real IO-APIC we 'lose
368                  * the interrupt edge' if the channel is masked.
369                  */
370                 if (sync_test_bit(port, &s->evtchn_pending[0]) &&
371                     !sync_test_and_set_bit(port / BITS_PER_LONG,
372                                            &vcpu_info->evtchn_pending_sel))
373                         vcpu_info->evtchn_upcall_pending = 1;
374         }
375
376         put_cpu();
377 }
378
379 static int xen_allocate_irq_dynamic(void)
380 {
381         int first = 0;
382         int irq;
383
384 #ifdef CONFIG_X86_IO_APIC
385         /*
386          * For an HVM guest or domain 0 which see "real" (emulated or
387          * actual repectively) GSIs we allocate dynamic IRQs
388          * e.g. those corresponding to event channels or MSIs
389          * etc. from the range above those "real" GSIs to avoid
390          * collisions.
391          */
392         if (xen_initial_domain() || xen_hvm_domain())
393                 first = get_nr_irqs_gsi();
394 #endif
395
396 retry:
397         irq = irq_alloc_desc_from(first, -1);
398
399         if (irq == -ENOMEM && first > NR_IRQS_LEGACY) {
400                 printk(KERN_ERR "Out of dynamic IRQ space and eating into GSI space. You should increase nr_irqs\n");
401                 first = max(NR_IRQS_LEGACY, first - NR_IRQS_LEGACY);
402                 goto retry;
403         }
404
405         if (irq < 0)
406                 panic("No available IRQ to bind to: increase nr_irqs!\n");
407
408         return irq;
409 }
410
411 static int xen_allocate_irq_gsi(unsigned gsi)
412 {
413         int irq;
414
415         /*
416          * A PV guest has no concept of a GSI (since it has no ACPI
417          * nor access to/knowledge of the physical APICs). Therefore
418          * all IRQs are dynamically allocated from the entire IRQ
419          * space.
420          */
421         if (xen_pv_domain() && !xen_initial_domain())
422                 return xen_allocate_irq_dynamic();
423
424         /* Legacy IRQ descriptors are already allocated by the arch. */
425         if (gsi < NR_IRQS_LEGACY)
426                 return gsi;
427
428         irq = irq_alloc_desc_at(gsi, -1);
429         if (irq < 0)
430                 panic("Unable to allocate to IRQ%d (%d)\n", gsi, irq);
431
432         return irq;
433 }
434
435 static void xen_free_irq(unsigned irq)
436 {
437         /* Legacy IRQ descriptors are managed by the arch. */
438         if (irq < NR_IRQS_LEGACY)
439                 return;
440
441         irq_free_desc(irq);
442 }
443
444 static void pirq_unmask_notify(int irq)
445 {
446         struct physdev_eoi eoi = { .irq = pirq_from_irq(irq) };
447
448         if (unlikely(pirq_needs_eoi(irq))) {
449                 int rc = HYPERVISOR_physdev_op(PHYSDEVOP_eoi, &eoi);
450                 WARN_ON(rc);
451         }
452 }
453
454 static void pirq_query_unmask(int irq)
455 {
456         struct physdev_irq_status_query irq_status;
457         struct irq_info *info = info_for_irq(irq);
458
459         BUG_ON(info->type != IRQT_PIRQ);
460
461         irq_status.irq = pirq_from_irq(irq);
462         if (HYPERVISOR_physdev_op(PHYSDEVOP_irq_status_query, &irq_status))
463                 irq_status.flags = 0;
464
465         info->u.pirq.flags &= ~PIRQ_NEEDS_EOI;
466         if (irq_status.flags & XENIRQSTAT_needs_eoi)
467                 info->u.pirq.flags |= PIRQ_NEEDS_EOI;
468 }
469
470 static bool probing_irq(int irq)
471 {
472         struct irq_desc *desc = irq_to_desc(irq);
473
474         return desc && desc->action == NULL;
475 }
476
477 static unsigned int __startup_pirq(unsigned int irq)
478 {
479         struct evtchn_bind_pirq bind_pirq;
480         struct irq_info *info = info_for_irq(irq);
481         int evtchn = evtchn_from_irq(irq);
482         int rc;
483
484         BUG_ON(info->type != IRQT_PIRQ);
485
486         if (VALID_EVTCHN(evtchn))
487                 goto out;
488
489         bind_pirq.pirq = pirq_from_irq(irq);
490         /* NB. We are happy to share unless we are probing. */
491         bind_pirq.flags = info->u.pirq.flags & PIRQ_SHAREABLE ?
492                                         BIND_PIRQ__WILL_SHARE : 0;
493         rc = HYPERVISOR_event_channel_op(EVTCHNOP_bind_pirq, &bind_pirq);
494         if (rc != 0) {
495                 if (!probing_irq(irq))
496                         printk(KERN_INFO "Failed to obtain physical IRQ %d\n",
497                                irq);
498                 return 0;
499         }
500         evtchn = bind_pirq.port;
501
502         pirq_query_unmask(irq);
503
504         evtchn_to_irq[evtchn] = irq;
505         bind_evtchn_to_cpu(evtchn, 0);
506         info->evtchn = evtchn;
507
508 out:
509         unmask_evtchn(evtchn);
510         pirq_unmask_notify(irq);
511
512         return 0;
513 }
514
515 static unsigned int startup_pirq(struct irq_data *data)
516 {
517         return __startup_pirq(data->irq);
518 }
519
520 static void shutdown_pirq(struct irq_data *data)
521 {
522         struct evtchn_close close;
523         unsigned int irq = data->irq;
524         struct irq_info *info = info_for_irq(irq);
525         int evtchn = evtchn_from_irq(irq);
526
527         BUG_ON(info->type != IRQT_PIRQ);
528
529         if (!VALID_EVTCHN(evtchn))
530                 return;
531
532         mask_evtchn(evtchn);
533
534         close.port = evtchn;
535         if (HYPERVISOR_event_channel_op(EVTCHNOP_close, &close) != 0)
536                 BUG();
537
538         bind_evtchn_to_cpu(evtchn, 0);
539         evtchn_to_irq[evtchn] = -1;
540         info->evtchn = 0;
541 }
542
543 static void enable_pirq(struct irq_data *data)
544 {
545         startup_pirq(data);
546 }
547
548 static void disable_pirq(struct irq_data *data)
549 {
550 }
551
552 static void ack_pirq(struct irq_data *data)
553 {
554         int evtchn = evtchn_from_irq(data->irq);
555
556         move_native_irq(data->irq);
557
558         if (VALID_EVTCHN(evtchn)) {
559                 mask_evtchn(evtchn);
560                 clear_evtchn(evtchn);
561         }
562 }
563
564 static int find_irq_by_gsi(unsigned gsi)
565 {
566         int irq;
567
568         for (irq = 0; irq < nr_irqs; irq++) {
569                 struct irq_info *info = info_for_irq(irq);
570
571                 if (info == NULL || info->type != IRQT_PIRQ)
572                         continue;
573
574                 if (gsi_from_irq(irq) == gsi)
575                         return irq;
576         }
577
578         return -1;
579 }
580
581 int xen_allocate_pirq(unsigned gsi, int shareable, char *name)
582 {
583         return xen_map_pirq_gsi(gsi, gsi, shareable, name);
584 }
585
586 /* xen_map_pirq_gsi might allocate irqs from the top down, as a
587  * consequence don't assume that the irq number returned has a low value
588  * or can be used as a pirq number unless you know otherwise.
589  *
590  * One notable exception is when xen_map_pirq_gsi is called passing an
591  * hardware gsi as argument, in that case the irq number returned
592  * matches the gsi number passed as second argument.
593  *
594  * Note: We don't assign an event channel until the irq actually started
595  * up.  Return an existing irq if we've already got one for the gsi.
596  */
597 int xen_map_pirq_gsi(unsigned pirq, unsigned gsi, int shareable, char *name)
598 {
599         int irq = 0;
600         struct physdev_irq irq_op;
601
602         spin_lock(&irq_mapping_update_lock);
603
604         if ((pirq > nr_irqs) || (gsi > nr_irqs)) {
605                 printk(KERN_WARNING "xen_map_pirq_gsi: %s %s is incorrect!\n",
606                         pirq > nr_irqs ? "pirq" :"",
607                         gsi > nr_irqs ? "gsi" : "");
608                 goto out;
609         }
610
611         irq = find_irq_by_gsi(gsi);
612         if (irq != -1) {
613                 printk(KERN_INFO "xen_map_pirq_gsi: returning irq %d for gsi %u\n",
614                        irq, gsi);
615                 goto out;       /* XXX need refcount? */
616         }
617
618         irq = xen_allocate_irq_gsi(gsi);
619
620         set_irq_chip_and_handler_name(irq, &xen_pirq_chip,
621                                       handle_level_irq, name);
622
623         irq_op.irq = irq;
624         irq_op.vector = 0;
625
626         /* Only the privileged domain can do this. For non-priv, the pcifront
627          * driver provides a PCI bus that does the call to do exactly
628          * this in the priv domain. */
629         if (xen_initial_domain() &&
630             HYPERVISOR_physdev_op(PHYSDEVOP_alloc_irq_vector, &irq_op)) {
631                 xen_free_irq(irq);
632                 irq = -ENOSPC;
633                 goto out;
634         }
635
636         irq_info[irq] = mk_pirq_info(0, pirq, gsi, irq_op.vector);
637         irq_info[irq].u.pirq.flags |= shareable ? PIRQ_SHAREABLE : 0;
638         pirq_to_irq[pirq] = irq;
639
640 out:
641         spin_unlock(&irq_mapping_update_lock);
642
643         return irq;
644 }
645
646 #ifdef CONFIG_PCI_MSI
647 #include <linux/msi.h>
648 #include "../pci/msi.h"
649
650 static int find_unbound_pirq(int type)
651 {
652         int rc, i;
653         struct physdev_get_free_pirq op_get_free_pirq;
654         op_get_free_pirq.type = type;
655
656         rc = HYPERVISOR_physdev_op(PHYSDEVOP_get_free_pirq, &op_get_free_pirq);
657         if (!rc)
658                 return op_get_free_pirq.pirq;
659
660         for (i = 0; i < nr_irqs; i++) {
661                 if (pirq_to_irq[i] < 0)
662                         return i;
663         }
664         return -1;
665 }
666
667 void xen_allocate_pirq_msi(char *name, int *irq, int *pirq, int alloc)
668 {
669         spin_lock(&irq_mapping_update_lock);
670
671         if (alloc & XEN_ALLOC_IRQ) {
672                 *irq = xen_allocate_irq_dynamic();
673                 if (*irq == -1)
674                         goto out;
675         }
676
677         if (alloc & XEN_ALLOC_PIRQ) {
678                 *pirq = find_unbound_pirq(MAP_PIRQ_TYPE_MSI);
679                 if (*pirq == -1)
680                         goto out;
681         }
682
683         set_irq_chip_and_handler_name(*irq, &xen_pirq_chip,
684                                       handle_level_irq, name);
685
686         irq_info[*irq] = mk_pirq_info(0, *pirq, 0, 0);
687         pirq_to_irq[*pirq] = *irq;
688
689 out:
690         spin_unlock(&irq_mapping_update_lock);
691 }
692
693 int xen_create_msi_irq(struct pci_dev *dev, struct msi_desc *msidesc, int type)
694 {
695         int irq = -1;
696         struct physdev_map_pirq map_irq;
697         int rc;
698         int pos;
699         u32 table_offset, bir;
700
701         memset(&map_irq, 0, sizeof(map_irq));
702         map_irq.domid = DOMID_SELF;
703         map_irq.type = MAP_PIRQ_TYPE_MSI;
704         map_irq.index = -1;
705         map_irq.pirq = -1;
706         map_irq.bus = dev->bus->number;
707         map_irq.devfn = dev->devfn;
708
709         if (type == PCI_CAP_ID_MSIX) {
710                 pos = pci_find_capability(dev, PCI_CAP_ID_MSIX);
711
712                 pci_read_config_dword(dev, msix_table_offset_reg(pos),
713                                         &table_offset);
714                 bir = (u8)(table_offset & PCI_MSIX_FLAGS_BIRMASK);
715
716                 map_irq.table_base = pci_resource_start(dev, bir);
717                 map_irq.entry_nr = msidesc->msi_attrib.entry_nr;
718         }
719
720         spin_lock(&irq_mapping_update_lock);
721
722         irq = xen_allocate_irq_dynamic();
723
724         if (irq == -1)
725                 goto out;
726
727         rc = HYPERVISOR_physdev_op(PHYSDEVOP_map_pirq, &map_irq);
728         if (rc) {
729                 printk(KERN_WARNING "xen map irq failed %d\n", rc);
730
731                 xen_free_irq(irq);
732
733                 irq = -1;
734                 goto out;
735         }
736         irq_info[irq] = mk_pirq_info(0, map_irq.pirq, 0, map_irq.index);
737
738         set_irq_chip_and_handler_name(irq, &xen_pirq_chip,
739                         handle_level_irq,
740                         (type == PCI_CAP_ID_MSIX) ? "msi-x":"msi");
741
742 out:
743         spin_unlock(&irq_mapping_update_lock);
744         return irq;
745 }
746 #endif
747
748 int xen_destroy_irq(int irq)
749 {
750         struct irq_desc *desc;
751         struct physdev_unmap_pirq unmap_irq;
752         struct irq_info *info = info_for_irq(irq);
753         int rc = -ENOENT;
754
755         spin_lock(&irq_mapping_update_lock);
756
757         desc = irq_to_desc(irq);
758         if (!desc)
759                 goto out;
760
761         if (xen_initial_domain()) {
762                 unmap_irq.pirq = info->u.pirq.pirq;
763                 unmap_irq.domid = DOMID_SELF;
764                 rc = HYPERVISOR_physdev_op(PHYSDEVOP_unmap_pirq, &unmap_irq);
765                 if (rc) {
766                         printk(KERN_WARNING "unmap irq failed %d\n", rc);
767                         goto out;
768                 }
769                 pirq_to_irq[info->u.pirq.pirq] = -1;
770         }
771         irq_info[irq] = mk_unbound_info();
772
773         xen_free_irq(irq);
774
775 out:
776         spin_unlock(&irq_mapping_update_lock);
777         return rc;
778 }
779
780 int xen_vector_from_irq(unsigned irq)
781 {
782         return vector_from_irq(irq);
783 }
784
785 int xen_gsi_from_irq(unsigned irq)
786 {
787         return gsi_from_irq(irq);
788 }
789
790 int xen_irq_from_pirq(unsigned pirq)
791 {
792         return pirq_to_irq[pirq];
793 }
794
795 int bind_evtchn_to_irq(unsigned int evtchn)
796 {
797         int irq;
798
799         spin_lock(&irq_mapping_update_lock);
800
801         irq = evtchn_to_irq[evtchn];
802
803         if (irq == -1) {
804                 irq = xen_allocate_irq_dynamic();
805
806                 set_irq_chip_and_handler_name(irq, &xen_dynamic_chip,
807                                               handle_fasteoi_irq, "event");
808
809                 evtchn_to_irq[evtchn] = irq;
810                 irq_info[irq] = mk_evtchn_info(evtchn);
811         }
812
813         spin_unlock(&irq_mapping_update_lock);
814
815         return irq;
816 }
817 EXPORT_SYMBOL_GPL(bind_evtchn_to_irq);
818
819 static int bind_ipi_to_irq(unsigned int ipi, unsigned int cpu)
820 {
821         struct evtchn_bind_ipi bind_ipi;
822         int evtchn, irq;
823
824         spin_lock(&irq_mapping_update_lock);
825
826         irq = per_cpu(ipi_to_irq, cpu)[ipi];
827
828         if (irq == -1) {
829                 irq = xen_allocate_irq_dynamic();
830                 if (irq < 0)
831                         goto out;
832
833                 set_irq_chip_and_handler_name(irq, &xen_percpu_chip,
834                                               handle_percpu_irq, "ipi");
835
836                 bind_ipi.vcpu = cpu;
837                 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_ipi,
838                                                 &bind_ipi) != 0)
839                         BUG();
840                 evtchn = bind_ipi.port;
841
842                 evtchn_to_irq[evtchn] = irq;
843                 irq_info[irq] = mk_ipi_info(evtchn, ipi);
844                 per_cpu(ipi_to_irq, cpu)[ipi] = irq;
845
846                 bind_evtchn_to_cpu(evtchn, cpu);
847         }
848
849  out:
850         spin_unlock(&irq_mapping_update_lock);
851         return irq;
852 }
853
854
855 int bind_virq_to_irq(unsigned int virq, unsigned int cpu)
856 {
857         struct evtchn_bind_virq bind_virq;
858         int evtchn, irq;
859
860         spin_lock(&irq_mapping_update_lock);
861
862         irq = per_cpu(virq_to_irq, cpu)[virq];
863
864         if (irq == -1) {
865                 irq = xen_allocate_irq_dynamic();
866
867                 set_irq_chip_and_handler_name(irq, &xen_percpu_chip,
868                                               handle_percpu_irq, "virq");
869
870                 bind_virq.virq = virq;
871                 bind_virq.vcpu = cpu;
872                 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_virq,
873                                                 &bind_virq) != 0)
874                         BUG();
875                 evtchn = bind_virq.port;
876
877                 evtchn_to_irq[evtchn] = irq;
878                 irq_info[irq] = mk_virq_info(evtchn, virq);
879
880                 per_cpu(virq_to_irq, cpu)[virq] = irq;
881
882                 bind_evtchn_to_cpu(evtchn, cpu);
883         }
884
885         spin_unlock(&irq_mapping_update_lock);
886
887         return irq;
888 }
889
890 static void unbind_from_irq(unsigned int irq)
891 {
892         struct evtchn_close close;
893         int evtchn = evtchn_from_irq(irq);
894
895         spin_lock(&irq_mapping_update_lock);
896
897         if (VALID_EVTCHN(evtchn)) {
898                 close.port = evtchn;
899                 if (HYPERVISOR_event_channel_op(EVTCHNOP_close, &close) != 0)
900                         BUG();
901
902                 switch (type_from_irq(irq)) {
903                 case IRQT_VIRQ:
904                         per_cpu(virq_to_irq, cpu_from_evtchn(evtchn))
905                                 [virq_from_irq(irq)] = -1;
906                         break;
907                 case IRQT_IPI:
908                         per_cpu(ipi_to_irq, cpu_from_evtchn(evtchn))
909                                 [ipi_from_irq(irq)] = -1;
910                         break;
911                 default:
912                         break;
913                 }
914
915                 /* Closed ports are implicitly re-bound to VCPU0. */
916                 bind_evtchn_to_cpu(evtchn, 0);
917
918                 evtchn_to_irq[evtchn] = -1;
919         }
920
921         if (irq_info[irq].type != IRQT_UNBOUND) {
922                 irq_info[irq] = mk_unbound_info();
923
924                 xen_free_irq(irq);
925         }
926
927         spin_unlock(&irq_mapping_update_lock);
928 }
929
930 int bind_evtchn_to_irqhandler(unsigned int evtchn,
931                               irq_handler_t handler,
932                               unsigned long irqflags,
933                               const char *devname, void *dev_id)
934 {
935         unsigned int irq;
936         int retval;
937
938         irq = bind_evtchn_to_irq(evtchn);
939         retval = request_irq(irq, handler, irqflags, devname, dev_id);
940         if (retval != 0) {
941                 unbind_from_irq(irq);
942                 return retval;
943         }
944
945         return irq;
946 }
947 EXPORT_SYMBOL_GPL(bind_evtchn_to_irqhandler);
948
949 int bind_virq_to_irqhandler(unsigned int virq, unsigned int cpu,
950                             irq_handler_t handler,
951                             unsigned long irqflags, const char *devname, void *dev_id)
952 {
953         unsigned int irq;
954         int retval;
955
956         irq = bind_virq_to_irq(virq, cpu);
957         retval = request_irq(irq, handler, irqflags, devname, dev_id);
958         if (retval != 0) {
959                 unbind_from_irq(irq);
960                 return retval;
961         }
962
963         return irq;
964 }
965 EXPORT_SYMBOL_GPL(bind_virq_to_irqhandler);
966
967 int bind_ipi_to_irqhandler(enum ipi_vector ipi,
968                            unsigned int cpu,
969                            irq_handler_t handler,
970                            unsigned long irqflags,
971                            const char *devname,
972                            void *dev_id)
973 {
974         int irq, retval;
975
976         irq = bind_ipi_to_irq(ipi, cpu);
977         if (irq < 0)
978                 return irq;
979
980         irqflags |= IRQF_NO_SUSPEND | IRQF_FORCE_RESUME;
981         retval = request_irq(irq, handler, irqflags, devname, dev_id);
982         if (retval != 0) {
983                 unbind_from_irq(irq);
984                 return retval;
985         }
986
987         return irq;
988 }
989
990 void unbind_from_irqhandler(unsigned int irq, void *dev_id)
991 {
992         free_irq(irq, dev_id);
993         unbind_from_irq(irq);
994 }
995 EXPORT_SYMBOL_GPL(unbind_from_irqhandler);
996
997 void xen_send_IPI_one(unsigned int cpu, enum ipi_vector vector)
998 {
999         int irq = per_cpu(ipi_to_irq, cpu)[vector];
1000         BUG_ON(irq < 0);
1001         notify_remote_via_irq(irq);
1002 }
1003
1004 irqreturn_t xen_debug_interrupt(int irq, void *dev_id)
1005 {
1006         struct shared_info *sh = HYPERVISOR_shared_info;
1007         int cpu = smp_processor_id();
1008         unsigned long *cpu_evtchn = cpu_evtchn_mask(cpu);
1009         int i;
1010         unsigned long flags;
1011         static DEFINE_SPINLOCK(debug_lock);
1012         struct vcpu_info *v;
1013
1014         spin_lock_irqsave(&debug_lock, flags);
1015
1016         printk("\nvcpu %d\n  ", cpu);
1017
1018         for_each_online_cpu(i) {
1019                 int pending;
1020                 v = per_cpu(xen_vcpu, i);
1021                 pending = (get_irq_regs() && i == cpu)
1022                         ? xen_irqs_disabled(get_irq_regs())
1023                         : v->evtchn_upcall_mask;
1024                 printk("%d: masked=%d pending=%d event_sel %0*lx\n  ", i,
1025                        pending, v->evtchn_upcall_pending,
1026                        (int)(sizeof(v->evtchn_pending_sel)*2),
1027                        v->evtchn_pending_sel);
1028         }
1029         v = per_cpu(xen_vcpu, cpu);
1030
1031         printk("\npending:\n   ");
1032         for (i = ARRAY_SIZE(sh->evtchn_pending)-1; i >= 0; i--)
1033                 printk("%0*lx%s", (int)sizeof(sh->evtchn_pending[0])*2,
1034                        sh->evtchn_pending[i],
1035                        i % 8 == 0 ? "\n   " : " ");
1036         printk("\nglobal mask:\n   ");
1037         for (i = ARRAY_SIZE(sh->evtchn_mask)-1; i >= 0; i--)
1038                 printk("%0*lx%s",
1039                        (int)(sizeof(sh->evtchn_mask[0])*2),
1040                        sh->evtchn_mask[i],
1041                        i % 8 == 0 ? "\n   " : " ");
1042
1043         printk("\nglobally unmasked:\n   ");
1044         for (i = ARRAY_SIZE(sh->evtchn_mask)-1; i >= 0; i--)
1045                 printk("%0*lx%s", (int)(sizeof(sh->evtchn_mask[0])*2),
1046                        sh->evtchn_pending[i] & ~sh->evtchn_mask[i],
1047                        i % 8 == 0 ? "\n   " : " ");
1048
1049         printk("\nlocal cpu%d mask:\n   ", cpu);
1050         for (i = (NR_EVENT_CHANNELS/BITS_PER_LONG)-1; i >= 0; i--)
1051                 printk("%0*lx%s", (int)(sizeof(cpu_evtchn[0])*2),
1052                        cpu_evtchn[i],
1053                        i % 8 == 0 ? "\n   " : " ");
1054
1055         printk("\nlocally unmasked:\n   ");
1056         for (i = ARRAY_SIZE(sh->evtchn_mask)-1; i >= 0; i--) {
1057                 unsigned long pending = sh->evtchn_pending[i]
1058                         & ~sh->evtchn_mask[i]
1059                         & cpu_evtchn[i];
1060                 printk("%0*lx%s", (int)(sizeof(sh->evtchn_mask[0])*2),
1061                        pending, i % 8 == 0 ? "\n   " : " ");
1062         }
1063
1064         printk("\npending list:\n");
1065         for (i = 0; i < NR_EVENT_CHANNELS; i++) {
1066                 if (sync_test_bit(i, sh->evtchn_pending)) {
1067                         int word_idx = i / BITS_PER_LONG;
1068                         printk("  %d: event %d -> irq %d%s%s%s\n",
1069                                cpu_from_evtchn(i), i,
1070                                evtchn_to_irq[i],
1071                                sync_test_bit(word_idx, &v->evtchn_pending_sel)
1072                                              ? "" : " l2-clear",
1073                                !sync_test_bit(i, sh->evtchn_mask)
1074                                              ? "" : " globally-masked",
1075                                sync_test_bit(i, cpu_evtchn)
1076                                              ? "" : " locally-masked");
1077                 }
1078         }
1079
1080         spin_unlock_irqrestore(&debug_lock, flags);
1081
1082         return IRQ_HANDLED;
1083 }
1084
1085 static DEFINE_PER_CPU(unsigned, xed_nesting_count);
1086
1087 /*
1088  * Search the CPUs pending events bitmasks.  For each one found, map
1089  * the event number to an irq, and feed it into do_IRQ() for
1090  * handling.
1091  *
1092  * Xen uses a two-level bitmap to speed searching.  The first level is
1093  * a bitset of words which contain pending event bits.  The second
1094  * level is a bitset of pending events themselves.
1095  */
1096 static void __xen_evtchn_do_upcall(void)
1097 {
1098         int cpu = get_cpu();
1099         struct shared_info *s = HYPERVISOR_shared_info;
1100         struct vcpu_info *vcpu_info = __this_cpu_read(xen_vcpu);
1101         unsigned count;
1102
1103         do {
1104                 unsigned long pending_words;
1105
1106                 vcpu_info->evtchn_upcall_pending = 0;
1107
1108                 if (__this_cpu_inc_return(xed_nesting_count) - 1)
1109                         goto out;
1110
1111 #ifndef CONFIG_X86 /* No need for a barrier -- XCHG is a barrier on x86. */
1112                 /* Clear master flag /before/ clearing selector flag. */
1113                 wmb();
1114 #endif
1115                 pending_words = xchg(&vcpu_info->evtchn_pending_sel, 0);
1116                 while (pending_words != 0) {
1117                         unsigned long pending_bits;
1118                         int word_idx = __ffs(pending_words);
1119                         pending_words &= ~(1UL << word_idx);
1120
1121                         while ((pending_bits = active_evtchns(cpu, s, word_idx)) != 0) {
1122                                 int bit_idx = __ffs(pending_bits);
1123                                 int port = (word_idx * BITS_PER_LONG) + bit_idx;
1124                                 int irq = evtchn_to_irq[port];
1125                                 struct irq_desc *desc;
1126
1127                                 mask_evtchn(port);
1128                                 clear_evtchn(port);
1129
1130                                 if (irq != -1) {
1131                                         desc = irq_to_desc(irq);
1132                                         if (desc)
1133                                                 generic_handle_irq_desc(irq, desc);
1134                                 }
1135                         }
1136                 }
1137
1138                 BUG_ON(!irqs_disabled());
1139
1140                 count = __this_cpu_read(xed_nesting_count);
1141                 __this_cpu_write(xed_nesting_count, 0);
1142         } while (count != 1 || vcpu_info->evtchn_upcall_pending);
1143
1144 out:
1145
1146         put_cpu();
1147 }
1148
1149 void xen_evtchn_do_upcall(struct pt_regs *regs)
1150 {
1151         struct pt_regs *old_regs = set_irq_regs(regs);
1152
1153         exit_idle();
1154         irq_enter();
1155
1156         __xen_evtchn_do_upcall();
1157
1158         irq_exit();
1159         set_irq_regs(old_regs);
1160 }
1161
1162 void xen_hvm_evtchn_do_upcall(void)
1163 {
1164         __xen_evtchn_do_upcall();
1165 }
1166 EXPORT_SYMBOL_GPL(xen_hvm_evtchn_do_upcall);
1167
1168 /* Rebind a new event channel to an existing irq. */
1169 void rebind_evtchn_irq(int evtchn, int irq)
1170 {
1171         struct irq_info *info = info_for_irq(irq);
1172
1173         /* Make sure the irq is masked, since the new event channel
1174            will also be masked. */
1175         disable_irq(irq);
1176
1177         spin_lock(&irq_mapping_update_lock);
1178
1179         /* After resume the irq<->evtchn mappings are all cleared out */
1180         BUG_ON(evtchn_to_irq[evtchn] != -1);
1181         /* Expect irq to have been bound before,
1182            so there should be a proper type */
1183         BUG_ON(info->type == IRQT_UNBOUND);
1184
1185         evtchn_to_irq[evtchn] = irq;
1186         irq_info[irq] = mk_evtchn_info(evtchn);
1187
1188         spin_unlock(&irq_mapping_update_lock);
1189
1190         /* new event channels are always bound to cpu 0 */
1191         irq_set_affinity(irq, cpumask_of(0));
1192
1193         /* Unmask the event channel. */
1194         enable_irq(irq);
1195 }
1196
1197 /* Rebind an evtchn so that it gets delivered to a specific cpu */
1198 static int rebind_irq_to_cpu(unsigned irq, unsigned tcpu)
1199 {
1200         struct evtchn_bind_vcpu bind_vcpu;
1201         int evtchn = evtchn_from_irq(irq);
1202
1203         /* events delivered via platform PCI interrupts are always
1204          * routed to vcpu 0 */
1205         if (!VALID_EVTCHN(evtchn) ||
1206                 (xen_hvm_domain() && !xen_have_vector_callback))
1207                 return -1;
1208
1209         /* Send future instances of this interrupt to other vcpu. */
1210         bind_vcpu.port = evtchn;
1211         bind_vcpu.vcpu = tcpu;
1212
1213         /*
1214          * If this fails, it usually just indicates that we're dealing with a
1215          * virq or IPI channel, which don't actually need to be rebound. Ignore
1216          * it, but don't do the xenlinux-level rebind in that case.
1217          */
1218         if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_vcpu, &bind_vcpu) >= 0)
1219                 bind_evtchn_to_cpu(evtchn, tcpu);
1220
1221         return 0;
1222 }
1223
1224 static int set_affinity_irq(struct irq_data *data, const struct cpumask *dest,
1225                             bool force)
1226 {
1227         unsigned tcpu = cpumask_first(dest);
1228
1229         return rebind_irq_to_cpu(data->irq, tcpu);
1230 }
1231
1232 int resend_irq_on_evtchn(unsigned int irq)
1233 {
1234         int masked, evtchn = evtchn_from_irq(irq);
1235         struct shared_info *s = HYPERVISOR_shared_info;
1236
1237         if (!VALID_EVTCHN(evtchn))
1238                 return 1;
1239
1240         masked = sync_test_and_set_bit(evtchn, s->evtchn_mask);
1241         sync_set_bit(evtchn, s->evtchn_pending);
1242         if (!masked)
1243                 unmask_evtchn(evtchn);
1244
1245         return 1;
1246 }
1247
1248 static void enable_dynirq(struct irq_data *data)
1249 {
1250         int evtchn = evtchn_from_irq(data->irq);
1251
1252         if (VALID_EVTCHN(evtchn))
1253                 unmask_evtchn(evtchn);
1254 }
1255
1256 static void disable_dynirq(struct irq_data *data)
1257 {
1258         int evtchn = evtchn_from_irq(data->irq);
1259
1260         if (VALID_EVTCHN(evtchn))
1261                 mask_evtchn(evtchn);
1262 }
1263
1264 static void ack_dynirq(struct irq_data *data)
1265 {
1266         int evtchn = evtchn_from_irq(data->irq);
1267
1268         move_masked_irq(data->irq);
1269
1270         if (VALID_EVTCHN(evtchn))
1271                 unmask_evtchn(evtchn);
1272 }
1273
1274 static int retrigger_dynirq(struct irq_data *data)
1275 {
1276         int evtchn = evtchn_from_irq(data->irq);
1277         struct shared_info *sh = HYPERVISOR_shared_info;
1278         int ret = 0;
1279
1280         if (VALID_EVTCHN(evtchn)) {
1281                 int masked;
1282
1283                 masked = sync_test_and_set_bit(evtchn, sh->evtchn_mask);
1284                 sync_set_bit(evtchn, sh->evtchn_pending);
1285                 if (!masked)
1286                         unmask_evtchn(evtchn);
1287                 ret = 1;
1288         }
1289
1290         return ret;
1291 }
1292
1293 static void restore_cpu_pirqs(void)
1294 {
1295         int pirq, rc, irq, gsi;
1296         struct physdev_map_pirq map_irq;
1297
1298         for (pirq = 0; pirq < nr_irqs; pirq++) {
1299                 irq = pirq_to_irq[pirq];
1300                 if (irq == -1)
1301                         continue;
1302
1303                 /* save/restore of PT devices doesn't work, so at this point the
1304                  * only devices present are GSI based emulated devices */
1305                 gsi = gsi_from_irq(irq);
1306                 if (!gsi)
1307                         continue;
1308
1309                 map_irq.domid = DOMID_SELF;
1310                 map_irq.type = MAP_PIRQ_TYPE_GSI;
1311                 map_irq.index = gsi;
1312                 map_irq.pirq = pirq;
1313
1314                 rc = HYPERVISOR_physdev_op(PHYSDEVOP_map_pirq, &map_irq);
1315                 if (rc) {
1316                         printk(KERN_WARNING "xen map irq failed gsi=%d irq=%d pirq=%d rc=%d\n",
1317                                         gsi, irq, pirq, rc);
1318                         irq_info[irq] = mk_unbound_info();
1319                         pirq_to_irq[pirq] = -1;
1320                         continue;
1321                 }
1322
1323                 printk(KERN_DEBUG "xen: --> irq=%d, pirq=%d\n", irq, map_irq.pirq);
1324
1325                 __startup_pirq(irq);
1326         }
1327 }
1328
1329 static void restore_cpu_virqs(unsigned int cpu)
1330 {
1331         struct evtchn_bind_virq bind_virq;
1332         int virq, irq, evtchn;
1333
1334         for (virq = 0; virq < NR_VIRQS; virq++) {
1335                 if ((irq = per_cpu(virq_to_irq, cpu)[virq]) == -1)
1336                         continue;
1337
1338                 BUG_ON(virq_from_irq(irq) != virq);
1339
1340                 /* Get a new binding from Xen. */
1341                 bind_virq.virq = virq;
1342                 bind_virq.vcpu = cpu;
1343                 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_virq,
1344                                                 &bind_virq) != 0)
1345                         BUG();
1346                 evtchn = bind_virq.port;
1347
1348                 /* Record the new mapping. */
1349                 evtchn_to_irq[evtchn] = irq;
1350                 irq_info[irq] = mk_virq_info(evtchn, virq);
1351                 bind_evtchn_to_cpu(evtchn, cpu);
1352         }
1353 }
1354
1355 static void restore_cpu_ipis(unsigned int cpu)
1356 {
1357         struct evtchn_bind_ipi bind_ipi;
1358         int ipi, irq, evtchn;
1359
1360         for (ipi = 0; ipi < XEN_NR_IPIS; ipi++) {
1361                 if ((irq = per_cpu(ipi_to_irq, cpu)[ipi]) == -1)
1362                         continue;
1363
1364                 BUG_ON(ipi_from_irq(irq) != ipi);
1365
1366                 /* Get a new binding from Xen. */
1367                 bind_ipi.vcpu = cpu;
1368                 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_ipi,
1369                                                 &bind_ipi) != 0)
1370                         BUG();
1371                 evtchn = bind_ipi.port;
1372
1373                 /* Record the new mapping. */
1374                 evtchn_to_irq[evtchn] = irq;
1375                 irq_info[irq] = mk_ipi_info(evtchn, ipi);
1376                 bind_evtchn_to_cpu(evtchn, cpu);
1377         }
1378 }
1379
1380 /* Clear an irq's pending state, in preparation for polling on it */
1381 void xen_clear_irq_pending(int irq)
1382 {
1383         int evtchn = evtchn_from_irq(irq);
1384
1385         if (VALID_EVTCHN(evtchn))
1386                 clear_evtchn(evtchn);
1387 }
1388 EXPORT_SYMBOL(xen_clear_irq_pending);
1389 void xen_set_irq_pending(int irq)
1390 {
1391         int evtchn = evtchn_from_irq(irq);
1392
1393         if (VALID_EVTCHN(evtchn))
1394                 set_evtchn(evtchn);
1395 }
1396
1397 bool xen_test_irq_pending(int irq)
1398 {
1399         int evtchn = evtchn_from_irq(irq);
1400         bool ret = false;
1401
1402         if (VALID_EVTCHN(evtchn))
1403                 ret = test_evtchn(evtchn);
1404
1405         return ret;
1406 }
1407
1408 /* Poll waiting for an irq to become pending with timeout.  In the usual case,
1409  * the irq will be disabled so it won't deliver an interrupt. */
1410 void xen_poll_irq_timeout(int irq, u64 timeout)
1411 {
1412         evtchn_port_t evtchn = evtchn_from_irq(irq);
1413
1414         if (VALID_EVTCHN(evtchn)) {
1415                 struct sched_poll poll;
1416
1417                 poll.nr_ports = 1;
1418                 poll.timeout = timeout;
1419                 set_xen_guest_handle(poll.ports, &evtchn);
1420
1421                 if (HYPERVISOR_sched_op(SCHEDOP_poll, &poll) != 0)
1422                         BUG();
1423         }
1424 }
1425 EXPORT_SYMBOL(xen_poll_irq_timeout);
1426 /* Poll waiting for an irq to become pending.  In the usual case, the
1427  * irq will be disabled so it won't deliver an interrupt. */
1428 void xen_poll_irq(int irq)
1429 {
1430         xen_poll_irq_timeout(irq, 0 /* no timeout */);
1431 }
1432
1433 void xen_irq_resume(void)
1434 {
1435         unsigned int cpu, irq, evtchn;
1436
1437         init_evtchn_cpu_bindings();
1438
1439         /* New event-channel space is not 'live' yet. */
1440         for (evtchn = 0; evtchn < NR_EVENT_CHANNELS; evtchn++)
1441                 mask_evtchn(evtchn);
1442
1443         /* No IRQ <-> event-channel mappings. */
1444         for (irq = 0; irq < nr_irqs; irq++)
1445                 irq_info[irq].evtchn = 0; /* zap event-channel binding */
1446
1447         for (evtchn = 0; evtchn < NR_EVENT_CHANNELS; evtchn++)
1448                 evtchn_to_irq[evtchn] = -1;
1449
1450         for_each_possible_cpu(cpu) {
1451                 restore_cpu_virqs(cpu);
1452                 restore_cpu_ipis(cpu);
1453         }
1454
1455         restore_cpu_pirqs();
1456 }
1457
1458 static struct irq_chip xen_dynamic_chip __read_mostly = {
1459         .name                   = "xen-dyn",
1460
1461         .irq_disable            = disable_dynirq,
1462         .irq_mask               = disable_dynirq,
1463         .irq_unmask             = enable_dynirq,
1464
1465         .irq_eoi                = ack_dynirq,
1466         .irq_set_affinity       = set_affinity_irq,
1467         .irq_retrigger          = retrigger_dynirq,
1468 };
1469
1470 static struct irq_chip xen_pirq_chip __read_mostly = {
1471         .name                   = "xen-pirq",
1472
1473         .irq_startup            = startup_pirq,
1474         .irq_shutdown           = shutdown_pirq,
1475
1476         .irq_enable             = enable_pirq,
1477         .irq_unmask             = enable_pirq,
1478
1479         .irq_disable            = disable_pirq,
1480         .irq_mask               = disable_pirq,
1481
1482         .irq_ack                = ack_pirq,
1483
1484         .irq_set_affinity       = set_affinity_irq,
1485
1486         .irq_retrigger          = retrigger_dynirq,
1487 };
1488
1489 static struct irq_chip xen_percpu_chip __read_mostly = {
1490         .name                   = "xen-percpu",
1491
1492         .irq_disable            = disable_dynirq,
1493         .irq_mask               = disable_dynirq,
1494         .irq_unmask             = enable_dynirq,
1495
1496         .irq_ack                = ack_dynirq,
1497 };
1498
1499 int xen_set_callback_via(uint64_t via)
1500 {
1501         struct xen_hvm_param a;
1502         a.domid = DOMID_SELF;
1503         a.index = HVM_PARAM_CALLBACK_IRQ;
1504         a.value = via;
1505         return HYPERVISOR_hvm_op(HVMOP_set_param, &a);
1506 }
1507 EXPORT_SYMBOL_GPL(xen_set_callback_via);
1508
1509 #ifdef CONFIG_XEN_PVHVM
1510 /* Vector callbacks are better than PCI interrupts to receive event
1511  * channel notifications because we can receive vector callbacks on any
1512  * vcpu and we don't need PCI support or APIC interactions. */
1513 void xen_callback_vector(void)
1514 {
1515         int rc;
1516         uint64_t callback_via;
1517         if (xen_have_vector_callback) {
1518                 callback_via = HVM_CALLBACK_VECTOR(XEN_HVM_EVTCHN_CALLBACK);
1519                 rc = xen_set_callback_via(callback_via);
1520                 if (rc) {
1521                         printk(KERN_ERR "Request for Xen HVM callback vector"
1522                                         " failed.\n");
1523                         xen_have_vector_callback = 0;
1524                         return;
1525                 }
1526                 printk(KERN_INFO "Xen HVM callback vector for event delivery is "
1527                                 "enabled\n");
1528                 /* in the restore case the vector has already been allocated */
1529                 if (!test_bit(XEN_HVM_EVTCHN_CALLBACK, used_vectors))
1530                         alloc_intr_gate(XEN_HVM_EVTCHN_CALLBACK, xen_hvm_callback_vector);
1531         }
1532 }
1533 #else
1534 void xen_callback_vector(void) {}
1535 #endif
1536
1537 void __init xen_init_IRQ(void)
1538 {
1539         int i;
1540
1541         cpu_evtchn_mask_p = kcalloc(nr_cpu_ids, sizeof(struct cpu_evtchn_s),
1542                                     GFP_KERNEL);
1543         irq_info = kcalloc(nr_irqs, sizeof(*irq_info), GFP_KERNEL);
1544
1545         /* We are using nr_irqs as the maximum number of pirq available but
1546          * that number is actually chosen by Xen and we don't know exactly
1547          * what it is. Be careful choosing high pirq numbers. */
1548         pirq_to_irq = kcalloc(nr_irqs, sizeof(*pirq_to_irq), GFP_KERNEL);
1549         for (i = 0; i < nr_irqs; i++)
1550                 pirq_to_irq[i] = -1;
1551
1552         evtchn_to_irq = kcalloc(NR_EVENT_CHANNELS, sizeof(*evtchn_to_irq),
1553                                     GFP_KERNEL);
1554         for (i = 0; i < NR_EVENT_CHANNELS; i++)
1555                 evtchn_to_irq[i] = -1;
1556
1557         init_evtchn_cpu_bindings();
1558
1559         /* No event channels are 'live' right now. */
1560         for (i = 0; i < NR_EVENT_CHANNELS; i++)
1561                 mask_evtchn(i);
1562
1563         if (xen_hvm_domain()) {
1564                 xen_callback_vector();
1565                 native_init_IRQ();
1566                 /* pci_xen_hvm_init must be called after native_init_IRQ so that
1567                  * __acpi_register_gsi can point at the right function */
1568                 pci_xen_hvm_init();
1569         } else {
1570                 irq_ctx_init(smp_processor_id());
1571                 if (xen_initial_domain())
1572                         xen_setup_pirqs();
1573         }
1574 }