xenbus: don't rely on xen_initial_domain to detect local xenstore
[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 received, 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 static LIST_HEAD(xen_irq_list_head);
60
61 /* IRQ <-> VIRQ mapping. */
62 static DEFINE_PER_CPU(int [NR_VIRQS], virq_to_irq) = {[0 ... NR_VIRQS-1] = -1};
63
64 /* IRQ <-> IPI mapping */
65 static DEFINE_PER_CPU(int [XEN_NR_IPIS], ipi_to_irq) = {[0 ... XEN_NR_IPIS-1] = -1};
66
67 /* Interrupt types. */
68 enum xen_irq_type {
69         IRQT_UNBOUND = 0,
70         IRQT_PIRQ,
71         IRQT_VIRQ,
72         IRQT_IPI,
73         IRQT_EVTCHN
74 };
75
76 /*
77  * Packed IRQ information:
78  * type - enum xen_irq_type
79  * event channel - irq->event channel mapping
80  * cpu - cpu this event channel is bound to
81  * index - type-specific information:
82  *    PIRQ - vector, with MSB being "needs EIO", or physical IRQ of the HVM
83  *           guest, or GSI (real passthrough IRQ) of the device.
84  *    VIRQ - virq number
85  *    IPI - IPI vector
86  *    EVTCHN -
87  */
88 struct irq_info
89 {
90         struct list_head list;
91         enum xen_irq_type type; /* type */
92         unsigned irq;
93         unsigned short evtchn;  /* event channel */
94         unsigned short cpu;     /* cpu bound */
95
96         union {
97                 unsigned short virq;
98                 enum ipi_vector ipi;
99                 struct {
100                         unsigned short pirq;
101                         unsigned short gsi;
102                         unsigned char vector;
103                         unsigned char flags;
104                         uint16_t domid;
105                 } pirq;
106         } u;
107 };
108 #define PIRQ_NEEDS_EOI  (1 << 0)
109 #define PIRQ_SHAREABLE  (1 << 1)
110
111 static int *evtchn_to_irq;
112
113 static DEFINE_PER_CPU(unsigned long [NR_EVENT_CHANNELS/BITS_PER_LONG],
114                       cpu_evtchn_mask);
115
116 /* Xen will never allocate port zero for any purpose. */
117 #define VALID_EVTCHN(chn)       ((chn) != 0)
118
119 static struct irq_chip xen_dynamic_chip;
120 static struct irq_chip xen_percpu_chip;
121 static struct irq_chip xen_pirq_chip;
122 static void enable_dynirq(struct irq_data *data);
123 static void disable_dynirq(struct irq_data *data);
124
125 /* Get info for IRQ */
126 static struct irq_info *info_for_irq(unsigned irq)
127 {
128         return irq_get_handler_data(irq);
129 }
130
131 /* Constructors for packed IRQ information. */
132 static void xen_irq_info_common_init(struct irq_info *info,
133                                      unsigned irq,
134                                      enum xen_irq_type type,
135                                      unsigned short evtchn,
136                                      unsigned short cpu)
137 {
138
139         BUG_ON(info->type != IRQT_UNBOUND && info->type != type);
140
141         info->type = type;
142         info->irq = irq;
143         info->evtchn = evtchn;
144         info->cpu = cpu;
145
146         evtchn_to_irq[evtchn] = irq;
147 }
148
149 static void xen_irq_info_evtchn_init(unsigned irq,
150                                      unsigned short evtchn)
151 {
152         struct irq_info *info = info_for_irq(irq);
153
154         xen_irq_info_common_init(info, irq, IRQT_EVTCHN, evtchn, 0);
155 }
156
157 static void xen_irq_info_ipi_init(unsigned cpu,
158                                   unsigned irq,
159                                   unsigned short evtchn,
160                                   enum ipi_vector ipi)
161 {
162         struct irq_info *info = info_for_irq(irq);
163
164         xen_irq_info_common_init(info, irq, IRQT_IPI, evtchn, 0);
165
166         info->u.ipi = ipi;
167
168         per_cpu(ipi_to_irq, cpu)[ipi] = irq;
169 }
170
171 static void xen_irq_info_virq_init(unsigned cpu,
172                                    unsigned irq,
173                                    unsigned short evtchn,
174                                    unsigned short virq)
175 {
176         struct irq_info *info = info_for_irq(irq);
177
178         xen_irq_info_common_init(info, irq, IRQT_VIRQ, evtchn, 0);
179
180         info->u.virq = virq;
181
182         per_cpu(virq_to_irq, cpu)[virq] = irq;
183 }
184
185 static void xen_irq_info_pirq_init(unsigned irq,
186                                    unsigned short evtchn,
187                                    unsigned short pirq,
188                                    unsigned short gsi,
189                                    unsigned short vector,
190                                    uint16_t domid,
191                                    unsigned char flags)
192 {
193         struct irq_info *info = info_for_irq(irq);
194
195         xen_irq_info_common_init(info, irq, IRQT_PIRQ, evtchn, 0);
196
197         info->u.pirq.pirq = pirq;
198         info->u.pirq.gsi = gsi;
199         info->u.pirq.vector = vector;
200         info->u.pirq.domid = domid;
201         info->u.pirq.flags = flags;
202 }
203
204 /*
205  * Accessors for packed IRQ information.
206  */
207 static unsigned int evtchn_from_irq(unsigned irq)
208 {
209         if (unlikely(WARN(irq < 0 || irq >= nr_irqs, "Invalid irq %d!\n", irq)))
210                 return 0;
211
212         return info_for_irq(irq)->evtchn;
213 }
214
215 unsigned irq_from_evtchn(unsigned int evtchn)
216 {
217         return evtchn_to_irq[evtchn];
218 }
219 EXPORT_SYMBOL_GPL(irq_from_evtchn);
220
221 static enum ipi_vector ipi_from_irq(unsigned irq)
222 {
223         struct irq_info *info = info_for_irq(irq);
224
225         BUG_ON(info == NULL);
226         BUG_ON(info->type != IRQT_IPI);
227
228         return info->u.ipi;
229 }
230
231 static unsigned virq_from_irq(unsigned irq)
232 {
233         struct irq_info *info = info_for_irq(irq);
234
235         BUG_ON(info == NULL);
236         BUG_ON(info->type != IRQT_VIRQ);
237
238         return info->u.virq;
239 }
240
241 static unsigned pirq_from_irq(unsigned irq)
242 {
243         struct irq_info *info = info_for_irq(irq);
244
245         BUG_ON(info == NULL);
246         BUG_ON(info->type != IRQT_PIRQ);
247
248         return info->u.pirq.pirq;
249 }
250
251 static enum xen_irq_type type_from_irq(unsigned irq)
252 {
253         return info_for_irq(irq)->type;
254 }
255
256 static unsigned cpu_from_irq(unsigned irq)
257 {
258         return info_for_irq(irq)->cpu;
259 }
260
261 static unsigned int cpu_from_evtchn(unsigned int evtchn)
262 {
263         int irq = evtchn_to_irq[evtchn];
264         unsigned ret = 0;
265
266         if (irq != -1)
267                 ret = cpu_from_irq(irq);
268
269         return ret;
270 }
271
272 static bool pirq_needs_eoi(unsigned irq)
273 {
274         struct irq_info *info = info_for_irq(irq);
275
276         BUG_ON(info->type != IRQT_PIRQ);
277
278         return info->u.pirq.flags & PIRQ_NEEDS_EOI;
279 }
280
281 static inline unsigned long active_evtchns(unsigned int cpu,
282                                            struct shared_info *sh,
283                                            unsigned int idx)
284 {
285         return (sh->evtchn_pending[idx] &
286                 per_cpu(cpu_evtchn_mask, cpu)[idx] &
287                 ~sh->evtchn_mask[idx]);
288 }
289
290 static void bind_evtchn_to_cpu(unsigned int chn, unsigned int cpu)
291 {
292         int irq = evtchn_to_irq[chn];
293
294         BUG_ON(irq == -1);
295 #ifdef CONFIG_SMP
296         cpumask_copy(irq_to_desc(irq)->irq_data.affinity, cpumask_of(cpu));
297 #endif
298
299         clear_bit(chn, per_cpu(cpu_evtchn_mask, cpu_from_irq(irq)));
300         set_bit(chn, per_cpu(cpu_evtchn_mask, cpu));
301
302         info_for_irq(irq)->cpu = cpu;
303 }
304
305 static void init_evtchn_cpu_bindings(void)
306 {
307         int i;
308 #ifdef CONFIG_SMP
309         struct irq_info *info;
310
311         /* By default all event channels notify CPU#0. */
312         list_for_each_entry(info, &xen_irq_list_head, list) {
313                 struct irq_desc *desc = irq_to_desc(info->irq);
314                 cpumask_copy(desc->irq_data.affinity, cpumask_of(0));
315         }
316 #endif
317
318         for_each_possible_cpu(i)
319                 memset(per_cpu(cpu_evtchn_mask, i),
320                        (i == 0) ? ~0 : 0, sizeof(*per_cpu(cpu_evtchn_mask, i)));
321 }
322
323 static inline void clear_evtchn(int port)
324 {
325         struct shared_info *s = HYPERVISOR_shared_info;
326         sync_clear_bit(port, &s->evtchn_pending[0]);
327 }
328
329 static inline void set_evtchn(int port)
330 {
331         struct shared_info *s = HYPERVISOR_shared_info;
332         sync_set_bit(port, &s->evtchn_pending[0]);
333 }
334
335 static inline int test_evtchn(int port)
336 {
337         struct shared_info *s = HYPERVISOR_shared_info;
338         return sync_test_bit(port, &s->evtchn_pending[0]);
339 }
340
341
342 /**
343  * notify_remote_via_irq - send event to remote end of event channel via irq
344  * @irq: irq of event channel to send event to
345  *
346  * Unlike notify_remote_via_evtchn(), this is safe to use across
347  * save/restore. Notifications on a broken connection are silently
348  * dropped.
349  */
350 void notify_remote_via_irq(int irq)
351 {
352         int evtchn = evtchn_from_irq(irq);
353
354         if (VALID_EVTCHN(evtchn))
355                 notify_remote_via_evtchn(evtchn);
356 }
357 EXPORT_SYMBOL_GPL(notify_remote_via_irq);
358
359 static void mask_evtchn(int port)
360 {
361         struct shared_info *s = HYPERVISOR_shared_info;
362         sync_set_bit(port, &s->evtchn_mask[0]);
363 }
364
365 static void unmask_evtchn(int port)
366 {
367         struct shared_info *s = HYPERVISOR_shared_info;
368         unsigned int cpu = get_cpu();
369
370         BUG_ON(!irqs_disabled());
371
372         /* Slow path (hypercall) if this is a non-local port. */
373         if (unlikely(cpu != cpu_from_evtchn(port))) {
374                 struct evtchn_unmask unmask = { .port = port };
375                 (void)HYPERVISOR_event_channel_op(EVTCHNOP_unmask, &unmask);
376         } else {
377                 struct vcpu_info *vcpu_info = __this_cpu_read(xen_vcpu);
378
379                 sync_clear_bit(port, &s->evtchn_mask[0]);
380
381                 /*
382                  * The following is basically the equivalent of
383                  * 'hw_resend_irq'. Just like a real IO-APIC we 'lose
384                  * the interrupt edge' if the channel is masked.
385                  */
386                 if (sync_test_bit(port, &s->evtchn_pending[0]) &&
387                     !sync_test_and_set_bit(port / BITS_PER_LONG,
388                                            &vcpu_info->evtchn_pending_sel))
389                         vcpu_info->evtchn_upcall_pending = 1;
390         }
391
392         put_cpu();
393 }
394
395 static void xen_irq_init(unsigned irq)
396 {
397         struct irq_info *info;
398 #ifdef CONFIG_SMP
399         struct irq_desc *desc = irq_to_desc(irq);
400
401         /* By default all event channels notify CPU#0. */
402         cpumask_copy(desc->irq_data.affinity, cpumask_of(0));
403 #endif
404
405         info = kzalloc(sizeof(*info), GFP_KERNEL);
406         if (info == NULL)
407                 panic("Unable to allocate metadata for IRQ%d\n", irq);
408
409         info->type = IRQT_UNBOUND;
410
411         irq_set_handler_data(irq, info);
412
413         list_add_tail(&info->list, &xen_irq_list_head);
414 }
415
416 static int __must_check xen_allocate_irq_dynamic(void)
417 {
418         int first = 0;
419         int irq;
420
421 #ifdef CONFIG_X86_IO_APIC
422         /*
423          * For an HVM guest or domain 0 which see "real" (emulated or
424          * actual respectively) GSIs we allocate dynamic IRQs
425          * e.g. those corresponding to event channels or MSIs
426          * etc. from the range above those "real" GSIs to avoid
427          * collisions.
428          */
429         if (xen_initial_domain() || xen_hvm_domain())
430                 first = get_nr_irqs_gsi();
431 #endif
432
433         irq = irq_alloc_desc_from(first, -1);
434
435         xen_irq_init(irq);
436
437         return irq;
438 }
439
440 static int __must_check xen_allocate_irq_gsi(unsigned gsi)
441 {
442         int irq;
443
444         /*
445          * A PV guest has no concept of a GSI (since it has no ACPI
446          * nor access to/knowledge of the physical APICs). Therefore
447          * all IRQs are dynamically allocated from the entire IRQ
448          * space.
449          */
450         if (xen_pv_domain() && !xen_initial_domain())
451                 return xen_allocate_irq_dynamic();
452
453         /* Legacy IRQ descriptors are already allocated by the arch. */
454         if (gsi < NR_IRQS_LEGACY)
455                 irq = gsi;
456         else
457                 irq = irq_alloc_desc_at(gsi, -1);
458
459         xen_irq_init(irq);
460
461         return irq;
462 }
463
464 static void xen_free_irq(unsigned irq)
465 {
466         struct irq_info *info = irq_get_handler_data(irq);
467
468         list_del(&info->list);
469
470         irq_set_handler_data(irq, NULL);
471
472         kfree(info);
473
474         /* Legacy IRQ descriptors are managed by the arch. */
475         if (irq < NR_IRQS_LEGACY)
476                 return;
477
478         irq_free_desc(irq);
479 }
480
481 static void pirq_query_unmask(int irq)
482 {
483         struct physdev_irq_status_query irq_status;
484         struct irq_info *info = info_for_irq(irq);
485
486         BUG_ON(info->type != IRQT_PIRQ);
487
488         irq_status.irq = pirq_from_irq(irq);
489         if (HYPERVISOR_physdev_op(PHYSDEVOP_irq_status_query, &irq_status))
490                 irq_status.flags = 0;
491
492         info->u.pirq.flags &= ~PIRQ_NEEDS_EOI;
493         if (irq_status.flags & XENIRQSTAT_needs_eoi)
494                 info->u.pirq.flags |= PIRQ_NEEDS_EOI;
495 }
496
497 static bool probing_irq(int irq)
498 {
499         struct irq_desc *desc = irq_to_desc(irq);
500
501         return desc && desc->action == NULL;
502 }
503
504 static void eoi_pirq(struct irq_data *data)
505 {
506         int evtchn = evtchn_from_irq(data->irq);
507         struct physdev_eoi eoi = { .irq = pirq_from_irq(data->irq) };
508         int rc = 0;
509
510         irq_move_irq(data);
511
512         if (VALID_EVTCHN(evtchn))
513                 clear_evtchn(evtchn);
514
515         if (pirq_needs_eoi(data->irq)) {
516                 rc = HYPERVISOR_physdev_op(PHYSDEVOP_eoi, &eoi);
517                 WARN_ON(rc);
518         }
519 }
520
521 static void mask_ack_pirq(struct irq_data *data)
522 {
523         disable_dynirq(data);
524         eoi_pirq(data);
525 }
526
527 static unsigned int __startup_pirq(unsigned int irq)
528 {
529         struct evtchn_bind_pirq bind_pirq;
530         struct irq_info *info = info_for_irq(irq);
531         int evtchn = evtchn_from_irq(irq);
532         int rc;
533
534         BUG_ON(info->type != IRQT_PIRQ);
535
536         if (VALID_EVTCHN(evtchn))
537                 goto out;
538
539         bind_pirq.pirq = pirq_from_irq(irq);
540         /* NB. We are happy to share unless we are probing. */
541         bind_pirq.flags = info->u.pirq.flags & PIRQ_SHAREABLE ?
542                                         BIND_PIRQ__WILL_SHARE : 0;
543         rc = HYPERVISOR_event_channel_op(EVTCHNOP_bind_pirq, &bind_pirq);
544         if (rc != 0) {
545                 if (!probing_irq(irq))
546                         printk(KERN_INFO "Failed to obtain physical IRQ %d\n",
547                                irq);
548                 return 0;
549         }
550         evtchn = bind_pirq.port;
551
552         pirq_query_unmask(irq);
553
554         evtchn_to_irq[evtchn] = irq;
555         bind_evtchn_to_cpu(evtchn, 0);
556         info->evtchn = evtchn;
557
558 out:
559         unmask_evtchn(evtchn);
560         eoi_pirq(irq_get_irq_data(irq));
561
562         return 0;
563 }
564
565 static unsigned int startup_pirq(struct irq_data *data)
566 {
567         return __startup_pirq(data->irq);
568 }
569
570 static void shutdown_pirq(struct irq_data *data)
571 {
572         struct evtchn_close close;
573         unsigned int irq = data->irq;
574         struct irq_info *info = info_for_irq(irq);
575         int evtchn = evtchn_from_irq(irq);
576
577         BUG_ON(info->type != IRQT_PIRQ);
578
579         if (!VALID_EVTCHN(evtchn))
580                 return;
581
582         mask_evtchn(evtchn);
583
584         close.port = evtchn;
585         if (HYPERVISOR_event_channel_op(EVTCHNOP_close, &close) != 0)
586                 BUG();
587
588         bind_evtchn_to_cpu(evtchn, 0);
589         evtchn_to_irq[evtchn] = -1;
590         info->evtchn = 0;
591 }
592
593 static void enable_pirq(struct irq_data *data)
594 {
595         startup_pirq(data);
596 }
597
598 static void disable_pirq(struct irq_data *data)
599 {
600         disable_dynirq(data);
601 }
602
603 static int find_irq_by_gsi(unsigned gsi)
604 {
605         struct irq_info *info;
606
607         list_for_each_entry(info, &xen_irq_list_head, list) {
608                 if (info->type != IRQT_PIRQ)
609                         continue;
610
611                 if (info->u.pirq.gsi == gsi)
612                         return info->irq;
613         }
614
615         return -1;
616 }
617
618 int xen_allocate_pirq_gsi(unsigned gsi)
619 {
620         return gsi;
621 }
622
623 /*
624  * Do not make any assumptions regarding the relationship between the
625  * IRQ number returned here and the Xen pirq argument.
626  *
627  * Note: We don't assign an event channel until the irq actually started
628  * up.  Return an existing irq if we've already got one for the gsi.
629  *
630  * Shareable implies level triggered, not shareable implies edge
631  * triggered here.
632  */
633 int xen_bind_pirq_gsi_to_irq(unsigned gsi,
634                              unsigned pirq, int shareable, char *name)
635 {
636         int irq = -1;
637         struct physdev_irq irq_op;
638
639         spin_lock(&irq_mapping_update_lock);
640
641         irq = find_irq_by_gsi(gsi);
642         if (irq != -1) {
643                 printk(KERN_INFO "xen_map_pirq_gsi: returning irq %d for gsi %u\n",
644                        irq, gsi);
645                 goto out;       /* XXX need refcount? */
646         }
647
648         irq = xen_allocate_irq_gsi(gsi);
649         if (irq < 0)
650                 goto out;
651
652         irq_op.irq = irq;
653         irq_op.vector = 0;
654
655         /* Only the privileged domain can do this. For non-priv, the pcifront
656          * driver provides a PCI bus that does the call to do exactly
657          * this in the priv domain. */
658         if (xen_initial_domain() &&
659             HYPERVISOR_physdev_op(PHYSDEVOP_alloc_irq_vector, &irq_op)) {
660                 xen_free_irq(irq);
661                 irq = -ENOSPC;
662                 goto out;
663         }
664
665         xen_irq_info_pirq_init(irq, 0, pirq, gsi, irq_op.vector, DOMID_SELF,
666                                shareable ? PIRQ_SHAREABLE : 0);
667
668         pirq_query_unmask(irq);
669         /* We try to use the handler with the appropriate semantic for the
670          * type of interrupt: if the interrupt is an edge triggered
671          * interrupt we use handle_edge_irq.
672          *
673          * On the other hand if the interrupt is level triggered we use
674          * handle_fasteoi_irq like the native code does for this kind of
675          * interrupts.
676          *
677          * Depending on the Xen version, pirq_needs_eoi might return true
678          * not only for level triggered interrupts but for edge triggered
679          * interrupts too. In any case Xen always honors the eoi mechanism,
680          * not injecting any more pirqs of the same kind if the first one
681          * hasn't received an eoi yet. Therefore using the fasteoi handler
682          * is the right choice either way.
683          */
684         if (shareable)
685                 irq_set_chip_and_handler_name(irq, &xen_pirq_chip,
686                                 handle_fasteoi_irq, name);
687         else
688                 irq_set_chip_and_handler_name(irq, &xen_pirq_chip,
689                                 handle_edge_irq, name);
690
691 out:
692         spin_unlock(&irq_mapping_update_lock);
693
694         return irq;
695 }
696
697 #ifdef CONFIG_PCI_MSI
698 int xen_allocate_pirq_msi(struct pci_dev *dev, struct msi_desc *msidesc)
699 {
700         int rc;
701         struct physdev_get_free_pirq op_get_free_pirq;
702
703         op_get_free_pirq.type = MAP_PIRQ_TYPE_MSI;
704         rc = HYPERVISOR_physdev_op(PHYSDEVOP_get_free_pirq, &op_get_free_pirq);
705
706         WARN_ONCE(rc == -ENOSYS,
707                   "hypervisor does not support the PHYSDEVOP_get_free_pirq interface\n");
708
709         return rc ? -1 : op_get_free_pirq.pirq;
710 }
711
712 int xen_bind_pirq_msi_to_irq(struct pci_dev *dev, struct msi_desc *msidesc,
713                              int pirq, int vector, const char *name,
714                              domid_t domid)
715 {
716         int irq, ret;
717
718         spin_lock(&irq_mapping_update_lock);
719
720         irq = xen_allocate_irq_dynamic();
721         if (irq == -1)
722                 goto out;
723
724         irq_set_chip_and_handler_name(irq, &xen_pirq_chip, handle_edge_irq,
725                         name);
726
727         xen_irq_info_pirq_init(irq, 0, pirq, 0, vector, domid, 0);
728         ret = irq_set_msi_desc(irq, msidesc);
729         if (ret < 0)
730                 goto error_irq;
731 out:
732         spin_unlock(&irq_mapping_update_lock);
733         return irq;
734 error_irq:
735         spin_unlock(&irq_mapping_update_lock);
736         xen_free_irq(irq);
737         return -1;
738 }
739 #endif
740
741 int xen_destroy_irq(int irq)
742 {
743         struct irq_desc *desc;
744         struct physdev_unmap_pirq unmap_irq;
745         struct irq_info *info = info_for_irq(irq);
746         int rc = -ENOENT;
747
748         spin_lock(&irq_mapping_update_lock);
749
750         desc = irq_to_desc(irq);
751         if (!desc)
752                 goto out;
753
754         if (xen_initial_domain()) {
755                 unmap_irq.pirq = info->u.pirq.pirq;
756                 unmap_irq.domid = info->u.pirq.domid;
757                 rc = HYPERVISOR_physdev_op(PHYSDEVOP_unmap_pirq, &unmap_irq);
758                 /* If another domain quits without making the pci_disable_msix
759                  * call, the Xen hypervisor takes care of freeing the PIRQs
760                  * (free_domain_pirqs).
761                  */
762                 if ((rc == -ESRCH && info->u.pirq.domid != DOMID_SELF))
763                         printk(KERN_INFO "domain %d does not have %d anymore\n",
764                                 info->u.pirq.domid, info->u.pirq.pirq);
765                 else if (rc) {
766                         printk(KERN_WARNING "unmap irq failed %d\n", rc);
767                         goto out;
768                 }
769         }
770
771         xen_free_irq(irq);
772
773 out:
774         spin_unlock(&irq_mapping_update_lock);
775         return rc;
776 }
777
778 int xen_irq_from_pirq(unsigned pirq)
779 {
780         int irq;
781
782         struct irq_info *info;
783
784         spin_lock(&irq_mapping_update_lock);
785
786         list_for_each_entry(info, &xen_irq_list_head, list) {
787                 if (info == NULL || info->type != IRQT_PIRQ)
788                         continue;
789                 irq = info->irq;
790                 if (info->u.pirq.pirq == pirq)
791                         goto out;
792         }
793         irq = -1;
794 out:
795         spin_unlock(&irq_mapping_update_lock);
796
797         return irq;
798 }
799
800
801 int xen_pirq_from_irq(unsigned irq)
802 {
803         return pirq_from_irq(irq);
804 }
805 EXPORT_SYMBOL_GPL(xen_pirq_from_irq);
806 int bind_evtchn_to_irq(unsigned int evtchn)
807 {
808         int irq;
809
810         spin_lock(&irq_mapping_update_lock);
811
812         irq = evtchn_to_irq[evtchn];
813
814         if (irq == -1) {
815                 irq = xen_allocate_irq_dynamic();
816                 if (irq == -1)
817                         goto out;
818
819                 irq_set_chip_and_handler_name(irq, &xen_dynamic_chip,
820                                               handle_edge_irq, "event");
821
822                 xen_irq_info_evtchn_init(irq, evtchn);
823         }
824
825 out:
826         spin_unlock(&irq_mapping_update_lock);
827
828         return irq;
829 }
830 EXPORT_SYMBOL_GPL(bind_evtchn_to_irq);
831
832 static int bind_ipi_to_irq(unsigned int ipi, unsigned int cpu)
833 {
834         struct evtchn_bind_ipi bind_ipi;
835         int evtchn, irq;
836
837         spin_lock(&irq_mapping_update_lock);
838
839         irq = per_cpu(ipi_to_irq, cpu)[ipi];
840
841         if (irq == -1) {
842                 irq = xen_allocate_irq_dynamic();
843                 if (irq < 0)
844                         goto out;
845
846                 irq_set_chip_and_handler_name(irq, &xen_percpu_chip,
847                                               handle_percpu_irq, "ipi");
848
849                 bind_ipi.vcpu = cpu;
850                 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_ipi,
851                                                 &bind_ipi) != 0)
852                         BUG();
853                 evtchn = bind_ipi.port;
854
855                 xen_irq_info_ipi_init(cpu, irq, evtchn, ipi);
856
857                 bind_evtchn_to_cpu(evtchn, cpu);
858         }
859
860  out:
861         spin_unlock(&irq_mapping_update_lock);
862         return irq;
863 }
864
865 static int bind_interdomain_evtchn_to_irq(unsigned int remote_domain,
866                                           unsigned int remote_port)
867 {
868         struct evtchn_bind_interdomain bind_interdomain;
869         int err;
870
871         bind_interdomain.remote_dom  = remote_domain;
872         bind_interdomain.remote_port = remote_port;
873
874         err = HYPERVISOR_event_channel_op(EVTCHNOP_bind_interdomain,
875                                           &bind_interdomain);
876
877         return err ? : bind_evtchn_to_irq(bind_interdomain.local_port);
878 }
879
880 static int find_virq(unsigned int virq, unsigned int cpu)
881 {
882         struct evtchn_status status;
883         int port, rc = -ENOENT;
884
885         memset(&status, 0, sizeof(status));
886         for (port = 0; port <= NR_EVENT_CHANNELS; port++) {
887                 status.dom = DOMID_SELF;
888                 status.port = port;
889                 rc = HYPERVISOR_event_channel_op(EVTCHNOP_status, &status);
890                 if (rc < 0)
891                         continue;
892                 if (status.status != EVTCHNSTAT_virq)
893                         continue;
894                 if (status.u.virq == virq && status.vcpu == cpu) {
895                         rc = port;
896                         break;
897                 }
898         }
899         return rc;
900 }
901
902 int bind_virq_to_irq(unsigned int virq, unsigned int cpu)
903 {
904         struct evtchn_bind_virq bind_virq;
905         int evtchn, irq, ret;
906
907         spin_lock(&irq_mapping_update_lock);
908
909         irq = per_cpu(virq_to_irq, cpu)[virq];
910
911         if (irq == -1) {
912                 irq = xen_allocate_irq_dynamic();
913                 if (irq == -1)
914                         goto out;
915
916                 irq_set_chip_and_handler_name(irq, &xen_percpu_chip,
917                                               handle_percpu_irq, "virq");
918
919                 bind_virq.virq = virq;
920                 bind_virq.vcpu = cpu;
921                 ret = HYPERVISOR_event_channel_op(EVTCHNOP_bind_virq,
922                                                 &bind_virq);
923                 if (ret == 0)
924                         evtchn = bind_virq.port;
925                 else {
926                         if (ret == -EEXIST)
927                                 ret = find_virq(virq, cpu);
928                         BUG_ON(ret < 0);
929                         evtchn = ret;
930                 }
931
932                 xen_irq_info_virq_init(cpu, irq, evtchn, virq);
933
934                 bind_evtchn_to_cpu(evtchn, cpu);
935         }
936
937 out:
938         spin_unlock(&irq_mapping_update_lock);
939
940         return irq;
941 }
942
943 static void unbind_from_irq(unsigned int irq)
944 {
945         struct evtchn_close close;
946         int evtchn = evtchn_from_irq(irq);
947
948         spin_lock(&irq_mapping_update_lock);
949
950         if (VALID_EVTCHN(evtchn)) {
951                 close.port = evtchn;
952                 if (HYPERVISOR_event_channel_op(EVTCHNOP_close, &close) != 0)
953                         BUG();
954
955                 switch (type_from_irq(irq)) {
956                 case IRQT_VIRQ:
957                         per_cpu(virq_to_irq, cpu_from_evtchn(evtchn))
958                                 [virq_from_irq(irq)] = -1;
959                         break;
960                 case IRQT_IPI:
961                         per_cpu(ipi_to_irq, cpu_from_evtchn(evtchn))
962                                 [ipi_from_irq(irq)] = -1;
963                         break;
964                 default:
965                         break;
966                 }
967
968                 /* Closed ports are implicitly re-bound to VCPU0. */
969                 bind_evtchn_to_cpu(evtchn, 0);
970
971                 evtchn_to_irq[evtchn] = -1;
972         }
973
974         BUG_ON(info_for_irq(irq)->type == IRQT_UNBOUND);
975
976         xen_free_irq(irq);
977
978         spin_unlock(&irq_mapping_update_lock);
979 }
980
981 int bind_evtchn_to_irqhandler(unsigned int evtchn,
982                               irq_handler_t handler,
983                               unsigned long irqflags,
984                               const char *devname, void *dev_id)
985 {
986         int irq, retval;
987
988         irq = bind_evtchn_to_irq(evtchn);
989         if (irq < 0)
990                 return irq;
991         retval = request_irq(irq, handler, irqflags, devname, dev_id);
992         if (retval != 0) {
993                 unbind_from_irq(irq);
994                 return retval;
995         }
996
997         return irq;
998 }
999 EXPORT_SYMBOL_GPL(bind_evtchn_to_irqhandler);
1000
1001 int bind_interdomain_evtchn_to_irqhandler(unsigned int remote_domain,
1002                                           unsigned int remote_port,
1003                                           irq_handler_t handler,
1004                                           unsigned long irqflags,
1005                                           const char *devname,
1006                                           void *dev_id)
1007 {
1008         int irq, retval;
1009
1010         irq = bind_interdomain_evtchn_to_irq(remote_domain, remote_port);
1011         if (irq < 0)
1012                 return irq;
1013
1014         retval = request_irq(irq, handler, irqflags, devname, dev_id);
1015         if (retval != 0) {
1016                 unbind_from_irq(irq);
1017                 return retval;
1018         }
1019
1020         return irq;
1021 }
1022 EXPORT_SYMBOL_GPL(bind_interdomain_evtchn_to_irqhandler);
1023
1024 int bind_virq_to_irqhandler(unsigned int virq, unsigned int cpu,
1025                             irq_handler_t handler,
1026                             unsigned long irqflags, const char *devname, void *dev_id)
1027 {
1028         int irq, retval;
1029
1030         irq = bind_virq_to_irq(virq, cpu);
1031         if (irq < 0)
1032                 return irq;
1033         retval = request_irq(irq, handler, irqflags, devname, dev_id);
1034         if (retval != 0) {
1035                 unbind_from_irq(irq);
1036                 return retval;
1037         }
1038
1039         return irq;
1040 }
1041 EXPORT_SYMBOL_GPL(bind_virq_to_irqhandler);
1042
1043 int bind_ipi_to_irqhandler(enum ipi_vector ipi,
1044                            unsigned int cpu,
1045                            irq_handler_t handler,
1046                            unsigned long irqflags,
1047                            const char *devname,
1048                            void *dev_id)
1049 {
1050         int irq, retval;
1051
1052         irq = bind_ipi_to_irq(ipi, cpu);
1053         if (irq < 0)
1054                 return irq;
1055
1056         irqflags |= IRQF_NO_SUSPEND | IRQF_FORCE_RESUME;
1057         retval = request_irq(irq, handler, irqflags, devname, dev_id);
1058         if (retval != 0) {
1059                 unbind_from_irq(irq);
1060                 return retval;
1061         }
1062
1063         return irq;
1064 }
1065
1066 void unbind_from_irqhandler(unsigned int irq, void *dev_id)
1067 {
1068         free_irq(irq, dev_id);
1069         unbind_from_irq(irq);
1070 }
1071 EXPORT_SYMBOL_GPL(unbind_from_irqhandler);
1072
1073 void xen_send_IPI_one(unsigned int cpu, enum ipi_vector vector)
1074 {
1075         int irq = per_cpu(ipi_to_irq, cpu)[vector];
1076         BUG_ON(irq < 0);
1077         notify_remote_via_irq(irq);
1078 }
1079
1080 irqreturn_t xen_debug_interrupt(int irq, void *dev_id)
1081 {
1082         struct shared_info *sh = HYPERVISOR_shared_info;
1083         int cpu = smp_processor_id();
1084         unsigned long *cpu_evtchn = per_cpu(cpu_evtchn_mask, cpu);
1085         int i;
1086         unsigned long flags;
1087         static DEFINE_SPINLOCK(debug_lock);
1088         struct vcpu_info *v;
1089
1090         spin_lock_irqsave(&debug_lock, flags);
1091
1092         printk("\nvcpu %d\n  ", cpu);
1093
1094         for_each_online_cpu(i) {
1095                 int pending;
1096                 v = per_cpu(xen_vcpu, i);
1097                 pending = (get_irq_regs() && i == cpu)
1098                         ? xen_irqs_disabled(get_irq_regs())
1099                         : v->evtchn_upcall_mask;
1100                 printk("%d: masked=%d pending=%d event_sel %0*lx\n  ", i,
1101                        pending, v->evtchn_upcall_pending,
1102                        (int)(sizeof(v->evtchn_pending_sel)*2),
1103                        v->evtchn_pending_sel);
1104         }
1105         v = per_cpu(xen_vcpu, cpu);
1106
1107         printk("\npending:\n   ");
1108         for (i = ARRAY_SIZE(sh->evtchn_pending)-1; i >= 0; i--)
1109                 printk("%0*lx%s", (int)sizeof(sh->evtchn_pending[0])*2,
1110                        sh->evtchn_pending[i],
1111                        i % 8 == 0 ? "\n   " : " ");
1112         printk("\nglobal mask:\n   ");
1113         for (i = ARRAY_SIZE(sh->evtchn_mask)-1; i >= 0; i--)
1114                 printk("%0*lx%s",
1115                        (int)(sizeof(sh->evtchn_mask[0])*2),
1116                        sh->evtchn_mask[i],
1117                        i % 8 == 0 ? "\n   " : " ");
1118
1119         printk("\nglobally unmasked:\n   ");
1120         for (i = ARRAY_SIZE(sh->evtchn_mask)-1; i >= 0; i--)
1121                 printk("%0*lx%s", (int)(sizeof(sh->evtchn_mask[0])*2),
1122                        sh->evtchn_pending[i] & ~sh->evtchn_mask[i],
1123                        i % 8 == 0 ? "\n   " : " ");
1124
1125         printk("\nlocal cpu%d mask:\n   ", cpu);
1126         for (i = (NR_EVENT_CHANNELS/BITS_PER_LONG)-1; i >= 0; i--)
1127                 printk("%0*lx%s", (int)(sizeof(cpu_evtchn[0])*2),
1128                        cpu_evtchn[i],
1129                        i % 8 == 0 ? "\n   " : " ");
1130
1131         printk("\nlocally unmasked:\n   ");
1132         for (i = ARRAY_SIZE(sh->evtchn_mask)-1; i >= 0; i--) {
1133                 unsigned long pending = sh->evtchn_pending[i]
1134                         & ~sh->evtchn_mask[i]
1135                         & cpu_evtchn[i];
1136                 printk("%0*lx%s", (int)(sizeof(sh->evtchn_mask[0])*2),
1137                        pending, i % 8 == 0 ? "\n   " : " ");
1138         }
1139
1140         printk("\npending list:\n");
1141         for (i = 0; i < NR_EVENT_CHANNELS; i++) {
1142                 if (sync_test_bit(i, sh->evtchn_pending)) {
1143                         int word_idx = i / BITS_PER_LONG;
1144                         printk("  %d: event %d -> irq %d%s%s%s\n",
1145                                cpu_from_evtchn(i), i,
1146                                evtchn_to_irq[i],
1147                                sync_test_bit(word_idx, &v->evtchn_pending_sel)
1148                                              ? "" : " l2-clear",
1149                                !sync_test_bit(i, sh->evtchn_mask)
1150                                              ? "" : " globally-masked",
1151                                sync_test_bit(i, cpu_evtchn)
1152                                              ? "" : " locally-masked");
1153                 }
1154         }
1155
1156         spin_unlock_irqrestore(&debug_lock, flags);
1157
1158         return IRQ_HANDLED;
1159 }
1160
1161 static DEFINE_PER_CPU(unsigned, xed_nesting_count);
1162 static DEFINE_PER_CPU(unsigned int, current_word_idx);
1163 static DEFINE_PER_CPU(unsigned int, current_bit_idx);
1164
1165 /*
1166  * Mask out the i least significant bits of w
1167  */
1168 #define MASK_LSBS(w, i) (w & ((~0UL) << i))
1169
1170 /*
1171  * Search the CPUs pending events bitmasks.  For each one found, map
1172  * the event number to an irq, and feed it into do_IRQ() for
1173  * handling.
1174  *
1175  * Xen uses a two-level bitmap to speed searching.  The first level is
1176  * a bitset of words which contain pending event bits.  The second
1177  * level is a bitset of pending events themselves.
1178  */
1179 static void __xen_evtchn_do_upcall(void)
1180 {
1181         int start_word_idx, start_bit_idx;
1182         int word_idx, bit_idx;
1183         int i;
1184         int cpu = get_cpu();
1185         struct shared_info *s = HYPERVISOR_shared_info;
1186         struct vcpu_info *vcpu_info = __this_cpu_read(xen_vcpu);
1187         unsigned count;
1188
1189         do {
1190                 unsigned long pending_words;
1191
1192                 vcpu_info->evtchn_upcall_pending = 0;
1193
1194                 if (__this_cpu_inc_return(xed_nesting_count) - 1)
1195                         goto out;
1196
1197 #ifndef CONFIG_X86 /* No need for a barrier -- XCHG is a barrier on x86. */
1198                 /* Clear master flag /before/ clearing selector flag. */
1199                 wmb();
1200 #endif
1201                 pending_words = xchg(&vcpu_info->evtchn_pending_sel, 0);
1202
1203                 start_word_idx = __this_cpu_read(current_word_idx);
1204                 start_bit_idx = __this_cpu_read(current_bit_idx);
1205
1206                 word_idx = start_word_idx;
1207
1208                 for (i = 0; pending_words != 0; i++) {
1209                         unsigned long pending_bits;
1210                         unsigned long words;
1211
1212                         words = MASK_LSBS(pending_words, word_idx);
1213
1214                         /*
1215                          * If we masked out all events, wrap to beginning.
1216                          */
1217                         if (words == 0) {
1218                                 word_idx = 0;
1219                                 bit_idx = 0;
1220                                 continue;
1221                         }
1222                         word_idx = __ffs(words);
1223
1224                         pending_bits = active_evtchns(cpu, s, word_idx);
1225                         bit_idx = 0; /* usually scan entire word from start */
1226                         if (word_idx == start_word_idx) {
1227                                 /* We scan the starting word in two parts */
1228                                 if (i == 0)
1229                                         /* 1st time: start in the middle */
1230                                         bit_idx = start_bit_idx;
1231                                 else
1232                                         /* 2nd time: mask bits done already */
1233                                         bit_idx &= (1UL << start_bit_idx) - 1;
1234                         }
1235
1236                         do {
1237                                 unsigned long bits;
1238                                 int port, irq;
1239                                 struct irq_desc *desc;
1240
1241                                 bits = MASK_LSBS(pending_bits, bit_idx);
1242
1243                                 /* If we masked out all events, move on. */
1244                                 if (bits == 0)
1245                                         break;
1246
1247                                 bit_idx = __ffs(bits);
1248
1249                                 /* Process port. */
1250                                 port = (word_idx * BITS_PER_LONG) + bit_idx;
1251                                 irq = evtchn_to_irq[port];
1252
1253                                 if (irq != -1) {
1254                                         desc = irq_to_desc(irq);
1255                                         if (desc)
1256                                                 generic_handle_irq_desc(irq, desc);
1257                                 }
1258
1259                                 bit_idx = (bit_idx + 1) % BITS_PER_LONG;
1260
1261                                 /* Next caller starts at last processed + 1 */
1262                                 __this_cpu_write(current_word_idx,
1263                                                  bit_idx ? word_idx :
1264                                                  (word_idx+1) % BITS_PER_LONG);
1265                                 __this_cpu_write(current_bit_idx, bit_idx);
1266                         } while (bit_idx != 0);
1267
1268                         /* Scan start_l1i twice; all others once. */
1269                         if ((word_idx != start_word_idx) || (i != 0))
1270                                 pending_words &= ~(1UL << word_idx);
1271
1272                         word_idx = (word_idx + 1) % BITS_PER_LONG;
1273                 }
1274
1275                 BUG_ON(!irqs_disabled());
1276
1277                 count = __this_cpu_read(xed_nesting_count);
1278                 __this_cpu_write(xed_nesting_count, 0);
1279         } while (count != 1 || vcpu_info->evtchn_upcall_pending);
1280
1281 out:
1282
1283         put_cpu();
1284 }
1285
1286 void xen_evtchn_do_upcall(struct pt_regs *regs)
1287 {
1288         struct pt_regs *old_regs = set_irq_regs(regs);
1289
1290         exit_idle();
1291         irq_enter();
1292
1293         __xen_evtchn_do_upcall();
1294
1295         irq_exit();
1296         set_irq_regs(old_regs);
1297 }
1298
1299 void xen_hvm_evtchn_do_upcall(void)
1300 {
1301         __xen_evtchn_do_upcall();
1302 }
1303 EXPORT_SYMBOL_GPL(xen_hvm_evtchn_do_upcall);
1304
1305 /* Rebind a new event channel to an existing irq. */
1306 void rebind_evtchn_irq(int evtchn, int irq)
1307 {
1308         struct irq_info *info = info_for_irq(irq);
1309
1310         /* Make sure the irq is masked, since the new event channel
1311            will also be masked. */
1312         disable_irq(irq);
1313
1314         spin_lock(&irq_mapping_update_lock);
1315
1316         /* After resume the irq<->evtchn mappings are all cleared out */
1317         BUG_ON(evtchn_to_irq[evtchn] != -1);
1318         /* Expect irq to have been bound before,
1319            so there should be a proper type */
1320         BUG_ON(info->type == IRQT_UNBOUND);
1321
1322         xen_irq_info_evtchn_init(irq, evtchn);
1323
1324         spin_unlock(&irq_mapping_update_lock);
1325
1326         /* new event channels are always bound to cpu 0 */
1327         irq_set_affinity(irq, cpumask_of(0));
1328
1329         /* Unmask the event channel. */
1330         enable_irq(irq);
1331 }
1332
1333 /* Rebind an evtchn so that it gets delivered to a specific cpu */
1334 static int rebind_irq_to_cpu(unsigned irq, unsigned tcpu)
1335 {
1336         struct evtchn_bind_vcpu bind_vcpu;
1337         int evtchn = evtchn_from_irq(irq);
1338
1339         if (!VALID_EVTCHN(evtchn))
1340                 return -1;
1341
1342         /*
1343          * Events delivered via platform PCI interrupts are always
1344          * routed to vcpu 0 and hence cannot be rebound.
1345          */
1346         if (xen_hvm_domain() && !xen_have_vector_callback)
1347                 return -1;
1348
1349         /* Send future instances of this interrupt to other vcpu. */
1350         bind_vcpu.port = evtchn;
1351         bind_vcpu.vcpu = tcpu;
1352
1353         /*
1354          * If this fails, it usually just indicates that we're dealing with a
1355          * virq or IPI channel, which don't actually need to be rebound. Ignore
1356          * it, but don't do the xenlinux-level rebind in that case.
1357          */
1358         if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_vcpu, &bind_vcpu) >= 0)
1359                 bind_evtchn_to_cpu(evtchn, tcpu);
1360
1361         return 0;
1362 }
1363
1364 static int set_affinity_irq(struct irq_data *data, const struct cpumask *dest,
1365                             bool force)
1366 {
1367         unsigned tcpu = cpumask_first(dest);
1368
1369         return rebind_irq_to_cpu(data->irq, tcpu);
1370 }
1371
1372 int resend_irq_on_evtchn(unsigned int irq)
1373 {
1374         int masked, evtchn = evtchn_from_irq(irq);
1375         struct shared_info *s = HYPERVISOR_shared_info;
1376
1377         if (!VALID_EVTCHN(evtchn))
1378                 return 1;
1379
1380         masked = sync_test_and_set_bit(evtchn, s->evtchn_mask);
1381         sync_set_bit(evtchn, s->evtchn_pending);
1382         if (!masked)
1383                 unmask_evtchn(evtchn);
1384
1385         return 1;
1386 }
1387
1388 static void enable_dynirq(struct irq_data *data)
1389 {
1390         int evtchn = evtchn_from_irq(data->irq);
1391
1392         if (VALID_EVTCHN(evtchn))
1393                 unmask_evtchn(evtchn);
1394 }
1395
1396 static void disable_dynirq(struct irq_data *data)
1397 {
1398         int evtchn = evtchn_from_irq(data->irq);
1399
1400         if (VALID_EVTCHN(evtchn))
1401                 mask_evtchn(evtchn);
1402 }
1403
1404 static void ack_dynirq(struct irq_data *data)
1405 {
1406         int evtchn = evtchn_from_irq(data->irq);
1407
1408         irq_move_irq(data);
1409
1410         if (VALID_EVTCHN(evtchn))
1411                 clear_evtchn(evtchn);
1412 }
1413
1414 static void mask_ack_dynirq(struct irq_data *data)
1415 {
1416         disable_dynirq(data);
1417         ack_dynirq(data);
1418 }
1419
1420 static int retrigger_dynirq(struct irq_data *data)
1421 {
1422         int evtchn = evtchn_from_irq(data->irq);
1423         struct shared_info *sh = HYPERVISOR_shared_info;
1424         int ret = 0;
1425
1426         if (VALID_EVTCHN(evtchn)) {
1427                 int masked;
1428
1429                 masked = sync_test_and_set_bit(evtchn, sh->evtchn_mask);
1430                 sync_set_bit(evtchn, sh->evtchn_pending);
1431                 if (!masked)
1432                         unmask_evtchn(evtchn);
1433                 ret = 1;
1434         }
1435
1436         return ret;
1437 }
1438
1439 static void restore_pirqs(void)
1440 {
1441         int pirq, rc, irq, gsi;
1442         struct physdev_map_pirq map_irq;
1443         struct irq_info *info;
1444
1445         list_for_each_entry(info, &xen_irq_list_head, list) {
1446                 if (info->type != IRQT_PIRQ)
1447                         continue;
1448
1449                 pirq = info->u.pirq.pirq;
1450                 gsi = info->u.pirq.gsi;
1451                 irq = info->irq;
1452
1453                 /* save/restore of PT devices doesn't work, so at this point the
1454                  * only devices present are GSI based emulated devices */
1455                 if (!gsi)
1456                         continue;
1457
1458                 map_irq.domid = DOMID_SELF;
1459                 map_irq.type = MAP_PIRQ_TYPE_GSI;
1460                 map_irq.index = gsi;
1461                 map_irq.pirq = pirq;
1462
1463                 rc = HYPERVISOR_physdev_op(PHYSDEVOP_map_pirq, &map_irq);
1464                 if (rc) {
1465                         printk(KERN_WARNING "xen map irq failed gsi=%d irq=%d pirq=%d rc=%d\n",
1466                                         gsi, irq, pirq, rc);
1467                         xen_free_irq(irq);
1468                         continue;
1469                 }
1470
1471                 printk(KERN_DEBUG "xen: --> irq=%d, pirq=%d\n", irq, map_irq.pirq);
1472
1473                 __startup_pirq(irq);
1474         }
1475 }
1476
1477 static void restore_cpu_virqs(unsigned int cpu)
1478 {
1479         struct evtchn_bind_virq bind_virq;
1480         int virq, irq, evtchn;
1481
1482         for (virq = 0; virq < NR_VIRQS; virq++) {
1483                 if ((irq = per_cpu(virq_to_irq, cpu)[virq]) == -1)
1484                         continue;
1485
1486                 BUG_ON(virq_from_irq(irq) != virq);
1487
1488                 /* Get a new binding from Xen. */
1489                 bind_virq.virq = virq;
1490                 bind_virq.vcpu = cpu;
1491                 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_virq,
1492                                                 &bind_virq) != 0)
1493                         BUG();
1494                 evtchn = bind_virq.port;
1495
1496                 /* Record the new mapping. */
1497                 xen_irq_info_virq_init(cpu, irq, evtchn, virq);
1498                 bind_evtchn_to_cpu(evtchn, cpu);
1499         }
1500 }
1501
1502 static void restore_cpu_ipis(unsigned int cpu)
1503 {
1504         struct evtchn_bind_ipi bind_ipi;
1505         int ipi, irq, evtchn;
1506
1507         for (ipi = 0; ipi < XEN_NR_IPIS; ipi++) {
1508                 if ((irq = per_cpu(ipi_to_irq, cpu)[ipi]) == -1)
1509                         continue;
1510
1511                 BUG_ON(ipi_from_irq(irq) != ipi);
1512
1513                 /* Get a new binding from Xen. */
1514                 bind_ipi.vcpu = cpu;
1515                 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_ipi,
1516                                                 &bind_ipi) != 0)
1517                         BUG();
1518                 evtchn = bind_ipi.port;
1519
1520                 /* Record the new mapping. */
1521                 xen_irq_info_ipi_init(cpu, irq, evtchn, ipi);
1522                 bind_evtchn_to_cpu(evtchn, cpu);
1523         }
1524 }
1525
1526 /* Clear an irq's pending state, in preparation for polling on it */
1527 void xen_clear_irq_pending(int irq)
1528 {
1529         int evtchn = evtchn_from_irq(irq);
1530
1531         if (VALID_EVTCHN(evtchn))
1532                 clear_evtchn(evtchn);
1533 }
1534 EXPORT_SYMBOL(xen_clear_irq_pending);
1535 void xen_set_irq_pending(int irq)
1536 {
1537         int evtchn = evtchn_from_irq(irq);
1538
1539         if (VALID_EVTCHN(evtchn))
1540                 set_evtchn(evtchn);
1541 }
1542
1543 bool xen_test_irq_pending(int irq)
1544 {
1545         int evtchn = evtchn_from_irq(irq);
1546         bool ret = false;
1547
1548         if (VALID_EVTCHN(evtchn))
1549                 ret = test_evtchn(evtchn);
1550
1551         return ret;
1552 }
1553
1554 /* Poll waiting for an irq to become pending with timeout.  In the usual case,
1555  * the irq will be disabled so it won't deliver an interrupt. */
1556 void xen_poll_irq_timeout(int irq, u64 timeout)
1557 {
1558         evtchn_port_t evtchn = evtchn_from_irq(irq);
1559
1560         if (VALID_EVTCHN(evtchn)) {
1561                 struct sched_poll poll;
1562
1563                 poll.nr_ports = 1;
1564                 poll.timeout = timeout;
1565                 set_xen_guest_handle(poll.ports, &evtchn);
1566
1567                 if (HYPERVISOR_sched_op(SCHEDOP_poll, &poll) != 0)
1568                         BUG();
1569         }
1570 }
1571 EXPORT_SYMBOL(xen_poll_irq_timeout);
1572 /* Poll waiting for an irq to become pending.  In the usual case, the
1573  * irq will be disabled so it won't deliver an interrupt. */
1574 void xen_poll_irq(int irq)
1575 {
1576         xen_poll_irq_timeout(irq, 0 /* no timeout */);
1577 }
1578
1579 /* Check whether the IRQ line is shared with other guests. */
1580 int xen_test_irq_shared(int irq)
1581 {
1582         struct irq_info *info = info_for_irq(irq);
1583         struct physdev_irq_status_query irq_status = { .irq = info->u.pirq.pirq };
1584
1585         if (HYPERVISOR_physdev_op(PHYSDEVOP_irq_status_query, &irq_status))
1586                 return 0;
1587         return !(irq_status.flags & XENIRQSTAT_shared);
1588 }
1589 EXPORT_SYMBOL_GPL(xen_test_irq_shared);
1590
1591 void xen_irq_resume(void)
1592 {
1593         unsigned int cpu, evtchn;
1594         struct irq_info *info;
1595
1596         init_evtchn_cpu_bindings();
1597
1598         /* New event-channel space is not 'live' yet. */
1599         for (evtchn = 0; evtchn < NR_EVENT_CHANNELS; evtchn++)
1600                 mask_evtchn(evtchn);
1601
1602         /* No IRQ <-> event-channel mappings. */
1603         list_for_each_entry(info, &xen_irq_list_head, list)
1604                 info->evtchn = 0; /* zap event-channel binding */
1605
1606         for (evtchn = 0; evtchn < NR_EVENT_CHANNELS; evtchn++)
1607                 evtchn_to_irq[evtchn] = -1;
1608
1609         for_each_possible_cpu(cpu) {
1610                 restore_cpu_virqs(cpu);
1611                 restore_cpu_ipis(cpu);
1612         }
1613
1614         restore_pirqs();
1615 }
1616
1617 static struct irq_chip xen_dynamic_chip __read_mostly = {
1618         .name                   = "xen-dyn",
1619
1620         .irq_disable            = disable_dynirq,
1621         .irq_mask               = disable_dynirq,
1622         .irq_unmask             = enable_dynirq,
1623
1624         .irq_ack                = ack_dynirq,
1625         .irq_mask_ack           = mask_ack_dynirq,
1626
1627         .irq_set_affinity       = set_affinity_irq,
1628         .irq_retrigger          = retrigger_dynirq,
1629 };
1630
1631 static struct irq_chip xen_pirq_chip __read_mostly = {
1632         .name                   = "xen-pirq",
1633
1634         .irq_startup            = startup_pirq,
1635         .irq_shutdown           = shutdown_pirq,
1636         .irq_enable             = enable_pirq,
1637         .irq_disable            = disable_pirq,
1638
1639         .irq_mask               = disable_dynirq,
1640         .irq_unmask             = enable_dynirq,
1641
1642         .irq_ack                = eoi_pirq,
1643         .irq_eoi                = eoi_pirq,
1644         .irq_mask_ack           = mask_ack_pirq,
1645
1646         .irq_set_affinity       = set_affinity_irq,
1647
1648         .irq_retrigger          = retrigger_dynirq,
1649 };
1650
1651 static struct irq_chip xen_percpu_chip __read_mostly = {
1652         .name                   = "xen-percpu",
1653
1654         .irq_disable            = disable_dynirq,
1655         .irq_mask               = disable_dynirq,
1656         .irq_unmask             = enable_dynirq,
1657
1658         .irq_ack                = ack_dynirq,
1659 };
1660
1661 int xen_set_callback_via(uint64_t via)
1662 {
1663         struct xen_hvm_param a;
1664         a.domid = DOMID_SELF;
1665         a.index = HVM_PARAM_CALLBACK_IRQ;
1666         a.value = via;
1667         return HYPERVISOR_hvm_op(HVMOP_set_param, &a);
1668 }
1669 EXPORT_SYMBOL_GPL(xen_set_callback_via);
1670
1671 #ifdef CONFIG_XEN_PVHVM
1672 /* Vector callbacks are better than PCI interrupts to receive event
1673  * channel notifications because we can receive vector callbacks on any
1674  * vcpu and we don't need PCI support or APIC interactions. */
1675 void xen_callback_vector(void)
1676 {
1677         int rc;
1678         uint64_t callback_via;
1679         if (xen_have_vector_callback) {
1680                 callback_via = HVM_CALLBACK_VECTOR(XEN_HVM_EVTCHN_CALLBACK);
1681                 rc = xen_set_callback_via(callback_via);
1682                 if (rc) {
1683                         printk(KERN_ERR "Request for Xen HVM callback vector"
1684                                         " failed.\n");
1685                         xen_have_vector_callback = 0;
1686                         return;
1687                 }
1688                 printk(KERN_INFO "Xen HVM callback vector for event delivery is "
1689                                 "enabled\n");
1690                 /* in the restore case the vector has already been allocated */
1691                 if (!test_bit(XEN_HVM_EVTCHN_CALLBACK, used_vectors))
1692                         alloc_intr_gate(XEN_HVM_EVTCHN_CALLBACK, xen_hvm_callback_vector);
1693         }
1694 }
1695 #else
1696 void xen_callback_vector(void) {}
1697 #endif
1698
1699 void __init xen_init_IRQ(void)
1700 {
1701         int i;
1702
1703         evtchn_to_irq = kcalloc(NR_EVENT_CHANNELS, sizeof(*evtchn_to_irq),
1704                                     GFP_KERNEL);
1705         for (i = 0; i < NR_EVENT_CHANNELS; i++)
1706                 evtchn_to_irq[i] = -1;
1707
1708         init_evtchn_cpu_bindings();
1709
1710         /* No event channels are 'live' right now. */
1711         for (i = 0; i < NR_EVENT_CHANNELS; i++)
1712                 mask_evtchn(i);
1713
1714         if (xen_hvm_domain()) {
1715                 xen_callback_vector();
1716                 native_init_IRQ();
1717                 /* pci_xen_hvm_init must be called after native_init_IRQ so that
1718                  * __acpi_register_gsi can point at the right function */
1719                 pci_xen_hvm_init();
1720         } else {
1721                 irq_ctx_init(smp_processor_id());
1722                 if (xen_initial_domain())
1723                         xen_setup_pirqs();
1724         }
1725 }