81a53eb6cd1dd6ac6a2b4ab2331bf79e5ea08aa7
[firefly-linux-kernel-4.4.55.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)->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->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 get_nr_hw_irqs(void)
380 {
381         int ret = 1;
382
383 #ifdef CONFIG_X86_IO_APIC
384         ret = get_nr_irqs_gsi();
385 #endif
386
387         return ret;
388 }
389
390 static int xen_allocate_irq_dynamic(void)
391 {
392         struct irq_data *data;
393         int irq, res;
394         int bottom = get_nr_hw_irqs();
395         int top = nr_irqs-1;
396
397         if (bottom == nr_irqs)
398                 goto no_irqs;
399
400         /* This loop starts from the top of IRQ space and goes down.
401          * We need this b/c if we have a PCI device in a Xen PV guest
402          * we do not have an IO-APIC (though the backend might have them)
403          * mapped in. To not have a collision of physical IRQs with the Xen
404          * event channels start at the top of the IRQ space for virtual IRQs.
405          */
406         for (irq = top; irq > bottom; irq--) {
407                 data = irq_get_irq_data(irq);
408                 /* only 15->0 have init'd desc; handle irq > 16 */
409                 if (!data)
410                         break;
411                 if (data->chip == &no_irq_chip)
412                         break;
413                 if (data->chip != &xen_dynamic_chip)
414                         continue;
415                 if (irq_info[irq].type == IRQT_UNBOUND)
416                         return irq;
417         }
418
419         if (irq == bottom)
420                 goto no_irqs;
421
422         res = irq_alloc_desc_at(irq, -1);
423
424         if (WARN_ON(res != irq))
425                 return -1;
426
427         return irq;
428
429 no_irqs:
430         panic("No available IRQ to bind to: increase nr_irqs!\n");
431 }
432
433 static bool identity_mapped_irq(unsigned irq)
434 {
435         /* identity map all the hardware irqs */
436         return irq < get_nr_hw_irqs();
437 }
438
439 static int xen_allocate_irq_gsi(unsigned gsi)
440 {
441         int irq;
442
443         if (!identity_mapped_irq(gsi) &&
444             (xen_initial_domain() || !xen_pv_domain()))
445                 return xen_allocate_irq_dynamic();
446
447         /* Legacy IRQ descriptors are already allocated by the arch. */
448         if (gsi < NR_IRQS_LEGACY)
449                 return gsi;
450
451         irq = irq_alloc_desc_at(gsi, -1);
452         if (irq < 0)
453                 panic("Unable to allocate to IRQ%d (%d)\n", gsi, irq);
454
455         return irq;
456 }
457
458 static void xen_free_irq(unsigned irq)
459 {
460         irq_free_desc(irq);
461 }
462
463 static void pirq_unmask_notify(int irq)
464 {
465         struct physdev_eoi eoi = { .irq = pirq_from_irq(irq) };
466
467         if (unlikely(pirq_needs_eoi(irq))) {
468                 int rc = HYPERVISOR_physdev_op(PHYSDEVOP_eoi, &eoi);
469                 WARN_ON(rc);
470         }
471 }
472
473 static void pirq_query_unmask(int irq)
474 {
475         struct physdev_irq_status_query irq_status;
476         struct irq_info *info = info_for_irq(irq);
477
478         BUG_ON(info->type != IRQT_PIRQ);
479
480         irq_status.irq = pirq_from_irq(irq);
481         if (HYPERVISOR_physdev_op(PHYSDEVOP_irq_status_query, &irq_status))
482                 irq_status.flags = 0;
483
484         info->u.pirq.flags &= ~PIRQ_NEEDS_EOI;
485         if (irq_status.flags & XENIRQSTAT_needs_eoi)
486                 info->u.pirq.flags |= PIRQ_NEEDS_EOI;
487 }
488
489 static bool probing_irq(int irq)
490 {
491         struct irq_desc *desc = irq_to_desc(irq);
492
493         return desc && desc->action == NULL;
494 }
495
496 static unsigned int startup_pirq(unsigned int irq)
497 {
498         struct evtchn_bind_pirq bind_pirq;
499         struct irq_info *info = info_for_irq(irq);
500         int evtchn = evtchn_from_irq(irq);
501         int rc;
502
503         BUG_ON(info->type != IRQT_PIRQ);
504
505         if (VALID_EVTCHN(evtchn))
506                 goto out;
507
508         bind_pirq.pirq = pirq_from_irq(irq);
509         /* NB. We are happy to share unless we are probing. */
510         bind_pirq.flags = info->u.pirq.flags & PIRQ_SHAREABLE ?
511                                         BIND_PIRQ__WILL_SHARE : 0;
512         rc = HYPERVISOR_event_channel_op(EVTCHNOP_bind_pirq, &bind_pirq);
513         if (rc != 0) {
514                 if (!probing_irq(irq))
515                         printk(KERN_INFO "Failed to obtain physical IRQ %d\n",
516                                irq);
517                 return 0;
518         }
519         evtchn = bind_pirq.port;
520
521         pirq_query_unmask(irq);
522
523         evtchn_to_irq[evtchn] = irq;
524         bind_evtchn_to_cpu(evtchn, 0);
525         info->evtchn = evtchn;
526
527 out:
528         unmask_evtchn(evtchn);
529         pirq_unmask_notify(irq);
530
531         return 0;
532 }
533
534 static void shutdown_pirq(unsigned int irq)
535 {
536         struct evtchn_close close;
537         struct irq_info *info = info_for_irq(irq);
538         int evtchn = evtchn_from_irq(irq);
539
540         BUG_ON(info->type != IRQT_PIRQ);
541
542         if (!VALID_EVTCHN(evtchn))
543                 return;
544
545         mask_evtchn(evtchn);
546
547         close.port = evtchn;
548         if (HYPERVISOR_event_channel_op(EVTCHNOP_close, &close) != 0)
549                 BUG();
550
551         bind_evtchn_to_cpu(evtchn, 0);
552         evtchn_to_irq[evtchn] = -1;
553         info->evtchn = 0;
554 }
555
556 static void enable_pirq(unsigned int irq)
557 {
558         startup_pirq(irq);
559 }
560
561 static void disable_pirq(unsigned int irq)
562 {
563 }
564
565 static void ack_pirq(unsigned int irq)
566 {
567         int evtchn = evtchn_from_irq(irq);
568
569         move_native_irq(irq);
570
571         if (VALID_EVTCHN(evtchn)) {
572                 mask_evtchn(evtchn);
573                 clear_evtchn(evtchn);
574         }
575 }
576
577 static void end_pirq(unsigned int irq)
578 {
579         int evtchn = evtchn_from_irq(irq);
580         struct irq_desc *desc = irq_to_desc(irq);
581
582         if (WARN_ON(!desc))
583                 return;
584
585         if ((desc->status & (IRQ_DISABLED|IRQ_PENDING)) ==
586             (IRQ_DISABLED|IRQ_PENDING)) {
587                 shutdown_pirq(irq);
588         } else if (VALID_EVTCHN(evtchn)) {
589                 unmask_evtchn(evtchn);
590                 pirq_unmask_notify(irq);
591         }
592 }
593
594 static int find_irq_by_gsi(unsigned gsi)
595 {
596         int irq;
597
598         for (irq = 0; irq < nr_irqs; irq++) {
599                 struct irq_info *info = info_for_irq(irq);
600
601                 if (info == NULL || info->type != IRQT_PIRQ)
602                         continue;
603
604                 if (gsi_from_irq(irq) == gsi)
605                         return irq;
606         }
607
608         return -1;
609 }
610
611 int xen_allocate_pirq(unsigned gsi, int shareable, char *name)
612 {
613         return xen_map_pirq_gsi(gsi, gsi, shareable, name);
614 }
615
616 /* xen_map_pirq_gsi might allocate irqs from the top down, as a
617  * consequence don't assume that the irq number returned has a low value
618  * or can be used as a pirq number unless you know otherwise.
619  *
620  * One notable exception is when xen_map_pirq_gsi is called passing an
621  * hardware gsi as argument, in that case the irq number returned
622  * matches the gsi number passed as second argument.
623  *
624  * Note: We don't assign an event channel until the irq actually started
625  * up.  Return an existing irq if we've already got one for the gsi.
626  */
627 int xen_map_pirq_gsi(unsigned pirq, unsigned gsi, int shareable, char *name)
628 {
629         int irq = 0;
630         struct physdev_irq irq_op;
631
632         spin_lock(&irq_mapping_update_lock);
633
634         if ((pirq > nr_irqs) || (gsi > nr_irqs)) {
635                 printk(KERN_WARNING "xen_map_pirq_gsi: %s %s is incorrect!\n",
636                         pirq > nr_irqs ? "pirq" :"",
637                         gsi > nr_irqs ? "gsi" : "");
638                 goto out;
639         }
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
650         set_irq_chip_and_handler_name(irq, &xen_pirq_chip,
651                                       handle_level_irq, name);
652
653         irq_op.irq = irq;
654         irq_op.vector = 0;
655
656         /* Only the privileged domain can do this. For non-priv, the pcifront
657          * driver provides a PCI bus that does the call to do exactly
658          * this in the priv domain. */
659         if (xen_initial_domain() &&
660             HYPERVISOR_physdev_op(PHYSDEVOP_alloc_irq_vector, &irq_op)) {
661                 xen_free_irq(irq);
662                 irq = -ENOSPC;
663                 goto out;
664         }
665
666         irq_info[irq] = mk_pirq_info(0, pirq, gsi, irq_op.vector);
667         irq_info[irq].u.pirq.flags |= shareable ? PIRQ_SHAREABLE : 0;
668         pirq_to_irq[pirq] = irq;
669
670 out:
671         spin_unlock(&irq_mapping_update_lock);
672
673         return irq;
674 }
675
676 #ifdef CONFIG_PCI_MSI
677 #include <linux/msi.h>
678 #include "../pci/msi.h"
679
680 static int find_unbound_pirq(int type)
681 {
682         int rc, i;
683         struct physdev_get_free_pirq op_get_free_pirq;
684         op_get_free_pirq.type = type;
685
686         rc = HYPERVISOR_physdev_op(PHYSDEVOP_get_free_pirq, &op_get_free_pirq);
687         if (!rc)
688                 return op_get_free_pirq.pirq;
689
690         for (i = 0; i < nr_irqs; i++) {
691                 if (pirq_to_irq[i] < 0)
692                         return i;
693         }
694         return -1;
695 }
696
697 void xen_allocate_pirq_msi(char *name, int *irq, int *pirq, int alloc)
698 {
699         spin_lock(&irq_mapping_update_lock);
700
701         if (alloc & XEN_ALLOC_IRQ) {
702                 *irq = xen_allocate_irq_dynamic();
703                 if (*irq == -1)
704                         goto out;
705         }
706
707         if (alloc & XEN_ALLOC_PIRQ) {
708                 *pirq = find_unbound_pirq(MAP_PIRQ_TYPE_MSI);
709                 if (*pirq == -1)
710                         goto out;
711         }
712
713         set_irq_chip_and_handler_name(*irq, &xen_pirq_chip,
714                                       handle_level_irq, name);
715
716         irq_info[*irq] = mk_pirq_info(0, *pirq, 0, 0);
717         pirq_to_irq[*pirq] = *irq;
718
719 out:
720         spin_unlock(&irq_mapping_update_lock);
721 }
722
723 int xen_create_msi_irq(struct pci_dev *dev, struct msi_desc *msidesc, int type)
724 {
725         int irq = -1;
726         struct physdev_map_pirq map_irq;
727         int rc;
728         int pos;
729         u32 table_offset, bir;
730
731         memset(&map_irq, 0, sizeof(map_irq));
732         map_irq.domid = DOMID_SELF;
733         map_irq.type = MAP_PIRQ_TYPE_MSI;
734         map_irq.index = -1;
735         map_irq.pirq = -1;
736         map_irq.bus = dev->bus->number;
737         map_irq.devfn = dev->devfn;
738
739         if (type == PCI_CAP_ID_MSIX) {
740                 pos = pci_find_capability(dev, PCI_CAP_ID_MSIX);
741
742                 pci_read_config_dword(dev, msix_table_offset_reg(pos),
743                                         &table_offset);
744                 bir = (u8)(table_offset & PCI_MSIX_FLAGS_BIRMASK);
745
746                 map_irq.table_base = pci_resource_start(dev, bir);
747                 map_irq.entry_nr = msidesc->msi_attrib.entry_nr;
748         }
749
750         spin_lock(&irq_mapping_update_lock);
751
752         irq = xen_allocate_irq_dynamic();
753
754         if (irq == -1)
755                 goto out;
756
757         rc = HYPERVISOR_physdev_op(PHYSDEVOP_map_pirq, &map_irq);
758         if (rc) {
759                 printk(KERN_WARNING "xen map irq failed %d\n", rc);
760
761                 xen_free_irq(irq);
762
763                 irq = -1;
764                 goto out;
765         }
766         irq_info[irq] = mk_pirq_info(0, map_irq.pirq, 0, map_irq.index);
767
768         set_irq_chip_and_handler_name(irq, &xen_pirq_chip,
769                         handle_level_irq,
770                         (type == PCI_CAP_ID_MSIX) ? "msi-x":"msi");
771
772 out:
773         spin_unlock(&irq_mapping_update_lock);
774         return irq;
775 }
776 #endif
777
778 int xen_destroy_irq(int irq)
779 {
780         struct irq_desc *desc;
781         struct physdev_unmap_pirq unmap_irq;
782         struct irq_info *info = info_for_irq(irq);
783         int rc = -ENOENT;
784
785         spin_lock(&irq_mapping_update_lock);
786
787         desc = irq_to_desc(irq);
788         if (!desc)
789                 goto out;
790
791         if (xen_initial_domain()) {
792                 unmap_irq.pirq = info->u.pirq.pirq;
793                 unmap_irq.domid = DOMID_SELF;
794                 rc = HYPERVISOR_physdev_op(PHYSDEVOP_unmap_pirq, &unmap_irq);
795                 if (rc) {
796                         printk(KERN_WARNING "unmap irq failed %d\n", rc);
797                         goto out;
798                 }
799                 pirq_to_irq[info->u.pirq.pirq] = -1;
800         }
801         irq_info[irq] = mk_unbound_info();
802
803         xen_free_irq(irq);
804
805 out:
806         spin_unlock(&irq_mapping_update_lock);
807         return rc;
808 }
809
810 int xen_vector_from_irq(unsigned irq)
811 {
812         return vector_from_irq(irq);
813 }
814
815 int xen_gsi_from_irq(unsigned irq)
816 {
817         return gsi_from_irq(irq);
818 }
819
820 int xen_irq_from_pirq(unsigned pirq)
821 {
822         return pirq_to_irq[pirq];
823 }
824
825 int bind_evtchn_to_irq(unsigned int evtchn)
826 {
827         int irq;
828
829         spin_lock(&irq_mapping_update_lock);
830
831         irq = evtchn_to_irq[evtchn];
832
833         if (irq == -1) {
834                 irq = xen_allocate_irq_dynamic();
835
836                 set_irq_chip_and_handler_name(irq, &xen_dynamic_chip,
837                                               handle_fasteoi_irq, "event");
838
839                 evtchn_to_irq[evtchn] = irq;
840                 irq_info[irq] = mk_evtchn_info(evtchn);
841         }
842
843         spin_unlock(&irq_mapping_update_lock);
844
845         return irq;
846 }
847 EXPORT_SYMBOL_GPL(bind_evtchn_to_irq);
848
849 static int bind_ipi_to_irq(unsigned int ipi, unsigned int cpu)
850 {
851         struct evtchn_bind_ipi bind_ipi;
852         int evtchn, irq;
853
854         spin_lock(&irq_mapping_update_lock);
855
856         irq = per_cpu(ipi_to_irq, cpu)[ipi];
857
858         if (irq == -1) {
859                 irq = xen_allocate_irq_dynamic();
860                 if (irq < 0)
861                         goto out;
862
863                 set_irq_chip_and_handler_name(irq, &xen_percpu_chip,
864                                               handle_percpu_irq, "ipi");
865
866                 bind_ipi.vcpu = cpu;
867                 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_ipi,
868                                                 &bind_ipi) != 0)
869                         BUG();
870                 evtchn = bind_ipi.port;
871
872                 evtchn_to_irq[evtchn] = irq;
873                 irq_info[irq] = mk_ipi_info(evtchn, ipi);
874                 per_cpu(ipi_to_irq, cpu)[ipi] = irq;
875
876                 bind_evtchn_to_cpu(evtchn, cpu);
877         }
878
879  out:
880         spin_unlock(&irq_mapping_update_lock);
881         return irq;
882 }
883
884
885 int bind_virq_to_irq(unsigned int virq, unsigned int cpu)
886 {
887         struct evtchn_bind_virq bind_virq;
888         int evtchn, irq;
889
890         spin_lock(&irq_mapping_update_lock);
891
892         irq = per_cpu(virq_to_irq, cpu)[virq];
893
894         if (irq == -1) {
895                 irq = xen_allocate_irq_dynamic();
896
897                 set_irq_chip_and_handler_name(irq, &xen_percpu_chip,
898                                               handle_percpu_irq, "virq");
899
900                 bind_virq.virq = virq;
901                 bind_virq.vcpu = cpu;
902                 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_virq,
903                                                 &bind_virq) != 0)
904                         BUG();
905                 evtchn = bind_virq.port;
906
907                 evtchn_to_irq[evtchn] = irq;
908                 irq_info[irq] = mk_virq_info(evtchn, virq);
909
910                 per_cpu(virq_to_irq, cpu)[virq] = irq;
911
912                 bind_evtchn_to_cpu(evtchn, cpu);
913         }
914
915         spin_unlock(&irq_mapping_update_lock);
916
917         return irq;
918 }
919
920 static void unbind_from_irq(unsigned int irq)
921 {
922         struct evtchn_close close;
923         int evtchn = evtchn_from_irq(irq);
924
925         spin_lock(&irq_mapping_update_lock);
926
927         if (VALID_EVTCHN(evtchn)) {
928                 close.port = evtchn;
929                 if (HYPERVISOR_event_channel_op(EVTCHNOP_close, &close) != 0)
930                         BUG();
931
932                 switch (type_from_irq(irq)) {
933                 case IRQT_VIRQ:
934                         per_cpu(virq_to_irq, cpu_from_evtchn(evtchn))
935                                 [virq_from_irq(irq)] = -1;
936                         break;
937                 case IRQT_IPI:
938                         per_cpu(ipi_to_irq, cpu_from_evtchn(evtchn))
939                                 [ipi_from_irq(irq)] = -1;
940                         break;
941                 default:
942                         break;
943                 }
944
945                 /* Closed ports are implicitly re-bound to VCPU0. */
946                 bind_evtchn_to_cpu(evtchn, 0);
947
948                 evtchn_to_irq[evtchn] = -1;
949         }
950
951         if (irq_info[irq].type != IRQT_UNBOUND) {
952                 irq_info[irq] = mk_unbound_info();
953
954                 xen_free_irq(irq);
955         }
956
957         spin_unlock(&irq_mapping_update_lock);
958 }
959
960 int bind_evtchn_to_irqhandler(unsigned int evtchn,
961                               irq_handler_t handler,
962                               unsigned long irqflags,
963                               const char *devname, void *dev_id)
964 {
965         unsigned int irq;
966         int retval;
967
968         irq = bind_evtchn_to_irq(evtchn);
969         retval = request_irq(irq, handler, irqflags, devname, dev_id);
970         if (retval != 0) {
971                 unbind_from_irq(irq);
972                 return retval;
973         }
974
975         return irq;
976 }
977 EXPORT_SYMBOL_GPL(bind_evtchn_to_irqhandler);
978
979 int bind_virq_to_irqhandler(unsigned int virq, unsigned int cpu,
980                             irq_handler_t handler,
981                             unsigned long irqflags, const char *devname, void *dev_id)
982 {
983         unsigned int irq;
984         int retval;
985
986         irq = bind_virq_to_irq(virq, cpu);
987         retval = request_irq(irq, handler, irqflags, devname, dev_id);
988         if (retval != 0) {
989                 unbind_from_irq(irq);
990                 return retval;
991         }
992
993         return irq;
994 }
995 EXPORT_SYMBOL_GPL(bind_virq_to_irqhandler);
996
997 int bind_ipi_to_irqhandler(enum ipi_vector ipi,
998                            unsigned int cpu,
999                            irq_handler_t handler,
1000                            unsigned long irqflags,
1001                            const char *devname,
1002                            void *dev_id)
1003 {
1004         int irq, retval;
1005
1006         irq = bind_ipi_to_irq(ipi, cpu);
1007         if (irq < 0)
1008                 return irq;
1009
1010         irqflags |= IRQF_NO_SUSPEND;
1011         retval = request_irq(irq, handler, irqflags, devname, dev_id);
1012         if (retval != 0) {
1013                 unbind_from_irq(irq);
1014                 return retval;
1015         }
1016
1017         return irq;
1018 }
1019
1020 void unbind_from_irqhandler(unsigned int irq, void *dev_id)
1021 {
1022         free_irq(irq, dev_id);
1023         unbind_from_irq(irq);
1024 }
1025 EXPORT_SYMBOL_GPL(unbind_from_irqhandler);
1026
1027 void xen_send_IPI_one(unsigned int cpu, enum ipi_vector vector)
1028 {
1029         int irq = per_cpu(ipi_to_irq, cpu)[vector];
1030         BUG_ON(irq < 0);
1031         notify_remote_via_irq(irq);
1032 }
1033
1034 irqreturn_t xen_debug_interrupt(int irq, void *dev_id)
1035 {
1036         struct shared_info *sh = HYPERVISOR_shared_info;
1037         int cpu = smp_processor_id();
1038         unsigned long *cpu_evtchn = cpu_evtchn_mask(cpu);
1039         int i;
1040         unsigned long flags;
1041         static DEFINE_SPINLOCK(debug_lock);
1042         struct vcpu_info *v;
1043
1044         spin_lock_irqsave(&debug_lock, flags);
1045
1046         printk("\nvcpu %d\n  ", cpu);
1047
1048         for_each_online_cpu(i) {
1049                 int pending;
1050                 v = per_cpu(xen_vcpu, i);
1051                 pending = (get_irq_regs() && i == cpu)
1052                         ? xen_irqs_disabled(get_irq_regs())
1053                         : v->evtchn_upcall_mask;
1054                 printk("%d: masked=%d pending=%d event_sel %0*lx\n  ", i,
1055                        pending, v->evtchn_upcall_pending,
1056                        (int)(sizeof(v->evtchn_pending_sel)*2),
1057                        v->evtchn_pending_sel);
1058         }
1059         v = per_cpu(xen_vcpu, cpu);
1060
1061         printk("\npending:\n   ");
1062         for (i = ARRAY_SIZE(sh->evtchn_pending)-1; i >= 0; i--)
1063                 printk("%0*lx%s", (int)sizeof(sh->evtchn_pending[0])*2,
1064                        sh->evtchn_pending[i],
1065                        i % 8 == 0 ? "\n   " : " ");
1066         printk("\nglobal mask:\n   ");
1067         for (i = ARRAY_SIZE(sh->evtchn_mask)-1; i >= 0; i--)
1068                 printk("%0*lx%s",
1069                        (int)(sizeof(sh->evtchn_mask[0])*2),
1070                        sh->evtchn_mask[i],
1071                        i % 8 == 0 ? "\n   " : " ");
1072
1073         printk("\nglobally unmasked:\n   ");
1074         for (i = ARRAY_SIZE(sh->evtchn_mask)-1; i >= 0; i--)
1075                 printk("%0*lx%s", (int)(sizeof(sh->evtchn_mask[0])*2),
1076                        sh->evtchn_pending[i] & ~sh->evtchn_mask[i],
1077                        i % 8 == 0 ? "\n   " : " ");
1078
1079         printk("\nlocal cpu%d mask:\n   ", cpu);
1080         for (i = (NR_EVENT_CHANNELS/BITS_PER_LONG)-1; i >= 0; i--)
1081                 printk("%0*lx%s", (int)(sizeof(cpu_evtchn[0])*2),
1082                        cpu_evtchn[i],
1083                        i % 8 == 0 ? "\n   " : " ");
1084
1085         printk("\nlocally unmasked:\n   ");
1086         for (i = ARRAY_SIZE(sh->evtchn_mask)-1; i >= 0; i--) {
1087                 unsigned long pending = sh->evtchn_pending[i]
1088                         & ~sh->evtchn_mask[i]
1089                         & cpu_evtchn[i];
1090                 printk("%0*lx%s", (int)(sizeof(sh->evtchn_mask[0])*2),
1091                        pending, i % 8 == 0 ? "\n   " : " ");
1092         }
1093
1094         printk("\npending list:\n");
1095         for (i = 0; i < NR_EVENT_CHANNELS; i++) {
1096                 if (sync_test_bit(i, sh->evtchn_pending)) {
1097                         int word_idx = i / BITS_PER_LONG;
1098                         printk("  %d: event %d -> irq %d%s%s%s\n",
1099                                cpu_from_evtchn(i), i,
1100                                evtchn_to_irq[i],
1101                                sync_test_bit(word_idx, &v->evtchn_pending_sel)
1102                                              ? "" : " l2-clear",
1103                                !sync_test_bit(i, sh->evtchn_mask)
1104                                              ? "" : " globally-masked",
1105                                sync_test_bit(i, cpu_evtchn)
1106                                              ? "" : " locally-masked");
1107                 }
1108         }
1109
1110         spin_unlock_irqrestore(&debug_lock, flags);
1111
1112         return IRQ_HANDLED;
1113 }
1114
1115 static DEFINE_PER_CPU(unsigned, xed_nesting_count);
1116
1117 /*
1118  * Search the CPUs pending events bitmasks.  For each one found, map
1119  * the event number to an irq, and feed it into do_IRQ() for
1120  * handling.
1121  *
1122  * Xen uses a two-level bitmap to speed searching.  The first level is
1123  * a bitset of words which contain pending event bits.  The second
1124  * level is a bitset of pending events themselves.
1125  */
1126 static void __xen_evtchn_do_upcall(void)
1127 {
1128         int cpu = get_cpu();
1129         struct shared_info *s = HYPERVISOR_shared_info;
1130         struct vcpu_info *vcpu_info = __this_cpu_read(xen_vcpu);
1131         unsigned count;
1132
1133         do {
1134                 unsigned long pending_words;
1135
1136                 vcpu_info->evtchn_upcall_pending = 0;
1137
1138                 if (__this_cpu_inc_return(xed_nesting_count) - 1)
1139                         goto out;
1140
1141 #ifndef CONFIG_X86 /* No need for a barrier -- XCHG is a barrier on x86. */
1142                 /* Clear master flag /before/ clearing selector flag. */
1143                 wmb();
1144 #endif
1145                 pending_words = xchg(&vcpu_info->evtchn_pending_sel, 0);
1146                 while (pending_words != 0) {
1147                         unsigned long pending_bits;
1148                         int word_idx = __ffs(pending_words);
1149                         pending_words &= ~(1UL << word_idx);
1150
1151                         while ((pending_bits = active_evtchns(cpu, s, word_idx)) != 0) {
1152                                 int bit_idx = __ffs(pending_bits);
1153                                 int port = (word_idx * BITS_PER_LONG) + bit_idx;
1154                                 int irq = evtchn_to_irq[port];
1155                                 struct irq_desc *desc;
1156
1157                                 mask_evtchn(port);
1158                                 clear_evtchn(port);
1159
1160                                 if (irq != -1) {
1161                                         desc = irq_to_desc(irq);
1162                                         if (desc)
1163                                                 generic_handle_irq_desc(irq, desc);
1164                                 }
1165                         }
1166                 }
1167
1168                 BUG_ON(!irqs_disabled());
1169
1170                 count = __this_cpu_read(xed_nesting_count);
1171                 __this_cpu_write(xed_nesting_count, 0);
1172         } while (count != 1 || vcpu_info->evtchn_upcall_pending);
1173
1174 out:
1175
1176         put_cpu();
1177 }
1178
1179 void xen_evtchn_do_upcall(struct pt_regs *regs)
1180 {
1181         struct pt_regs *old_regs = set_irq_regs(regs);
1182
1183         exit_idle();
1184         irq_enter();
1185
1186         __xen_evtchn_do_upcall();
1187
1188         irq_exit();
1189         set_irq_regs(old_regs);
1190 }
1191
1192 void xen_hvm_evtchn_do_upcall(void)
1193 {
1194         __xen_evtchn_do_upcall();
1195 }
1196 EXPORT_SYMBOL_GPL(xen_hvm_evtchn_do_upcall);
1197
1198 /* Rebind a new event channel to an existing irq. */
1199 void rebind_evtchn_irq(int evtchn, int irq)
1200 {
1201         struct irq_info *info = info_for_irq(irq);
1202
1203         /* Make sure the irq is masked, since the new event channel
1204            will also be masked. */
1205         disable_irq(irq);
1206
1207         spin_lock(&irq_mapping_update_lock);
1208
1209         /* After resume the irq<->evtchn mappings are all cleared out */
1210         BUG_ON(evtchn_to_irq[evtchn] != -1);
1211         /* Expect irq to have been bound before,
1212            so there should be a proper type */
1213         BUG_ON(info->type == IRQT_UNBOUND);
1214
1215         evtchn_to_irq[evtchn] = irq;
1216         irq_info[irq] = mk_evtchn_info(evtchn);
1217
1218         spin_unlock(&irq_mapping_update_lock);
1219
1220         /* new event channels are always bound to cpu 0 */
1221         irq_set_affinity(irq, cpumask_of(0));
1222
1223         /* Unmask the event channel. */
1224         enable_irq(irq);
1225 }
1226
1227 /* Rebind an evtchn so that it gets delivered to a specific cpu */
1228 static int rebind_irq_to_cpu(unsigned irq, unsigned tcpu)
1229 {
1230         struct evtchn_bind_vcpu bind_vcpu;
1231         int evtchn = evtchn_from_irq(irq);
1232
1233         /* events delivered via platform PCI interrupts are always
1234          * routed to vcpu 0 */
1235         if (!VALID_EVTCHN(evtchn) ||
1236                 (xen_hvm_domain() && !xen_have_vector_callback))
1237                 return -1;
1238
1239         /* Send future instances of this interrupt to other vcpu. */
1240         bind_vcpu.port = evtchn;
1241         bind_vcpu.vcpu = tcpu;
1242
1243         /*
1244          * If this fails, it usually just indicates that we're dealing with a
1245          * virq or IPI channel, which don't actually need to be rebound. Ignore
1246          * it, but don't do the xenlinux-level rebind in that case.
1247          */
1248         if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_vcpu, &bind_vcpu) >= 0)
1249                 bind_evtchn_to_cpu(evtchn, tcpu);
1250
1251         return 0;
1252 }
1253
1254 static int set_affinity_irq(unsigned irq, const struct cpumask *dest)
1255 {
1256         unsigned tcpu = cpumask_first(dest);
1257
1258         return rebind_irq_to_cpu(irq, tcpu);
1259 }
1260
1261 int resend_irq_on_evtchn(unsigned int irq)
1262 {
1263         int masked, evtchn = evtchn_from_irq(irq);
1264         struct shared_info *s = HYPERVISOR_shared_info;
1265
1266         if (!VALID_EVTCHN(evtchn))
1267                 return 1;
1268
1269         masked = sync_test_and_set_bit(evtchn, s->evtchn_mask);
1270         sync_set_bit(evtchn, s->evtchn_pending);
1271         if (!masked)
1272                 unmask_evtchn(evtchn);
1273
1274         return 1;
1275 }
1276
1277 static void enable_dynirq(unsigned int irq)
1278 {
1279         int evtchn = evtchn_from_irq(irq);
1280
1281         if (VALID_EVTCHN(evtchn))
1282                 unmask_evtchn(evtchn);
1283 }
1284
1285 static void disable_dynirq(unsigned int irq)
1286 {
1287         int evtchn = evtchn_from_irq(irq);
1288
1289         if (VALID_EVTCHN(evtchn))
1290                 mask_evtchn(evtchn);
1291 }
1292
1293 static void ack_dynirq(unsigned int irq)
1294 {
1295         int evtchn = evtchn_from_irq(irq);
1296
1297         move_masked_irq(irq);
1298
1299         if (VALID_EVTCHN(evtchn))
1300                 unmask_evtchn(evtchn);
1301 }
1302
1303 static int retrigger_dynirq(unsigned int irq)
1304 {
1305         int evtchn = evtchn_from_irq(irq);
1306         struct shared_info *sh = HYPERVISOR_shared_info;
1307         int ret = 0;
1308
1309         if (VALID_EVTCHN(evtchn)) {
1310                 int masked;
1311
1312                 masked = sync_test_and_set_bit(evtchn, sh->evtchn_mask);
1313                 sync_set_bit(evtchn, sh->evtchn_pending);
1314                 if (!masked)
1315                         unmask_evtchn(evtchn);
1316                 ret = 1;
1317         }
1318
1319         return ret;
1320 }
1321
1322 static void restore_cpu_pirqs(void)
1323 {
1324         int pirq, rc, irq, gsi;
1325         struct physdev_map_pirq map_irq;
1326
1327         for (pirq = 0; pirq < nr_irqs; pirq++) {
1328                 irq = pirq_to_irq[pirq];
1329                 if (irq == -1)
1330                         continue;
1331
1332                 /* save/restore of PT devices doesn't work, so at this point the
1333                  * only devices present are GSI based emulated devices */
1334                 gsi = gsi_from_irq(irq);
1335                 if (!gsi)
1336                         continue;
1337
1338                 map_irq.domid = DOMID_SELF;
1339                 map_irq.type = MAP_PIRQ_TYPE_GSI;
1340                 map_irq.index = gsi;
1341                 map_irq.pirq = pirq;
1342
1343                 rc = HYPERVISOR_physdev_op(PHYSDEVOP_map_pirq, &map_irq);
1344                 if (rc) {
1345                         printk(KERN_WARNING "xen map irq failed gsi=%d irq=%d pirq=%d rc=%d\n",
1346                                         gsi, irq, pirq, rc);
1347                         irq_info[irq] = mk_unbound_info();
1348                         pirq_to_irq[pirq] = -1;
1349                         continue;
1350                 }
1351
1352                 printk(KERN_DEBUG "xen: --> irq=%d, pirq=%d\n", irq, map_irq.pirq);
1353
1354                 startup_pirq(irq);
1355         }
1356 }
1357
1358 static void restore_cpu_virqs(unsigned int cpu)
1359 {
1360         struct evtchn_bind_virq bind_virq;
1361         int virq, irq, evtchn;
1362
1363         for (virq = 0; virq < NR_VIRQS; virq++) {
1364                 if ((irq = per_cpu(virq_to_irq, cpu)[virq]) == -1)
1365                         continue;
1366
1367                 BUG_ON(virq_from_irq(irq) != virq);
1368
1369                 /* Get a new binding from Xen. */
1370                 bind_virq.virq = virq;
1371                 bind_virq.vcpu = cpu;
1372                 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_virq,
1373                                                 &bind_virq) != 0)
1374                         BUG();
1375                 evtchn = bind_virq.port;
1376
1377                 /* Record the new mapping. */
1378                 evtchn_to_irq[evtchn] = irq;
1379                 irq_info[irq] = mk_virq_info(evtchn, virq);
1380                 bind_evtchn_to_cpu(evtchn, cpu);
1381         }
1382 }
1383
1384 static void restore_cpu_ipis(unsigned int cpu)
1385 {
1386         struct evtchn_bind_ipi bind_ipi;
1387         int ipi, irq, evtchn;
1388
1389         for (ipi = 0; ipi < XEN_NR_IPIS; ipi++) {
1390                 if ((irq = per_cpu(ipi_to_irq, cpu)[ipi]) == -1)
1391                         continue;
1392
1393                 BUG_ON(ipi_from_irq(irq) != ipi);
1394
1395                 /* Get a new binding from Xen. */
1396                 bind_ipi.vcpu = cpu;
1397                 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_ipi,
1398                                                 &bind_ipi) != 0)
1399                         BUG();
1400                 evtchn = bind_ipi.port;
1401
1402                 /* Record the new mapping. */
1403                 evtchn_to_irq[evtchn] = irq;
1404                 irq_info[irq] = mk_ipi_info(evtchn, ipi);
1405                 bind_evtchn_to_cpu(evtchn, cpu);
1406         }
1407 }
1408
1409 /* Clear an irq's pending state, in preparation for polling on it */
1410 void xen_clear_irq_pending(int irq)
1411 {
1412         int evtchn = evtchn_from_irq(irq);
1413
1414         if (VALID_EVTCHN(evtchn))
1415                 clear_evtchn(evtchn);
1416 }
1417 EXPORT_SYMBOL(xen_clear_irq_pending);
1418 void xen_set_irq_pending(int irq)
1419 {
1420         int evtchn = evtchn_from_irq(irq);
1421
1422         if (VALID_EVTCHN(evtchn))
1423                 set_evtchn(evtchn);
1424 }
1425
1426 bool xen_test_irq_pending(int irq)
1427 {
1428         int evtchn = evtchn_from_irq(irq);
1429         bool ret = false;
1430
1431         if (VALID_EVTCHN(evtchn))
1432                 ret = test_evtchn(evtchn);
1433
1434         return ret;
1435 }
1436
1437 /* Poll waiting for an irq to become pending with timeout.  In the usual case,
1438  * the irq will be disabled so it won't deliver an interrupt. */
1439 void xen_poll_irq_timeout(int irq, u64 timeout)
1440 {
1441         evtchn_port_t evtchn = evtchn_from_irq(irq);
1442
1443         if (VALID_EVTCHN(evtchn)) {
1444                 struct sched_poll poll;
1445
1446                 poll.nr_ports = 1;
1447                 poll.timeout = timeout;
1448                 set_xen_guest_handle(poll.ports, &evtchn);
1449
1450                 if (HYPERVISOR_sched_op(SCHEDOP_poll, &poll) != 0)
1451                         BUG();
1452         }
1453 }
1454 EXPORT_SYMBOL(xen_poll_irq_timeout);
1455 /* Poll waiting for an irq to become pending.  In the usual case, the
1456  * irq will be disabled so it won't deliver an interrupt. */
1457 void xen_poll_irq(int irq)
1458 {
1459         xen_poll_irq_timeout(irq, 0 /* no timeout */);
1460 }
1461
1462 void xen_irq_resume(void)
1463 {
1464         unsigned int cpu, irq, evtchn;
1465         struct irq_desc *desc;
1466
1467         init_evtchn_cpu_bindings();
1468
1469         /* New event-channel space is not 'live' yet. */
1470         for (evtchn = 0; evtchn < NR_EVENT_CHANNELS; evtchn++)
1471                 mask_evtchn(evtchn);
1472
1473         /* No IRQ <-> event-channel mappings. */
1474         for (irq = 0; irq < nr_irqs; irq++)
1475                 irq_info[irq].evtchn = 0; /* zap event-channel binding */
1476
1477         for (evtchn = 0; evtchn < NR_EVENT_CHANNELS; evtchn++)
1478                 evtchn_to_irq[evtchn] = -1;
1479
1480         for_each_possible_cpu(cpu) {
1481                 restore_cpu_virqs(cpu);
1482                 restore_cpu_ipis(cpu);
1483         }
1484
1485         /*
1486          * Unmask any IRQF_NO_SUSPEND IRQs which are enabled. These
1487          * are not handled by the IRQ core.
1488          */
1489         for_each_irq_desc(irq, desc) {
1490                 if (!desc->action || !(desc->action->flags & IRQF_NO_SUSPEND))
1491                         continue;
1492                 if (desc->status & IRQ_DISABLED)
1493                         continue;
1494
1495                 evtchn = evtchn_from_irq(irq);
1496                 if (evtchn == -1)
1497                         continue;
1498
1499                 unmask_evtchn(evtchn);
1500         }
1501
1502         restore_cpu_pirqs();
1503 }
1504
1505 static struct irq_chip xen_dynamic_chip __read_mostly = {
1506         .name           = "xen-dyn",
1507
1508         .disable        = disable_dynirq,
1509         .mask           = disable_dynirq,
1510         .unmask         = enable_dynirq,
1511
1512         .eoi            = ack_dynirq,
1513         .set_affinity   = set_affinity_irq,
1514         .retrigger      = retrigger_dynirq,
1515 };
1516
1517 static struct irq_chip xen_pirq_chip __read_mostly = {
1518         .name           = "xen-pirq",
1519
1520         .startup        = startup_pirq,
1521         .shutdown       = shutdown_pirq,
1522
1523         .enable         = enable_pirq,
1524         .unmask         = enable_pirq,
1525
1526         .disable        = disable_pirq,
1527         .mask           = disable_pirq,
1528
1529         .ack            = ack_pirq,
1530         .end            = end_pirq,
1531
1532         .set_affinity   = set_affinity_irq,
1533
1534         .retrigger      = retrigger_dynirq,
1535 };
1536
1537 static struct irq_chip xen_percpu_chip __read_mostly = {
1538         .name           = "xen-percpu",
1539
1540         .disable        = disable_dynirq,
1541         .mask           = disable_dynirq,
1542         .unmask         = enable_dynirq,
1543
1544         .ack            = ack_dynirq,
1545 };
1546
1547 int xen_set_callback_via(uint64_t via)
1548 {
1549         struct xen_hvm_param a;
1550         a.domid = DOMID_SELF;
1551         a.index = HVM_PARAM_CALLBACK_IRQ;
1552         a.value = via;
1553         return HYPERVISOR_hvm_op(HVMOP_set_param, &a);
1554 }
1555 EXPORT_SYMBOL_GPL(xen_set_callback_via);
1556
1557 #ifdef CONFIG_XEN_PVHVM
1558 /* Vector callbacks are better than PCI interrupts to receive event
1559  * channel notifications because we can receive vector callbacks on any
1560  * vcpu and we don't need PCI support or APIC interactions. */
1561 void xen_callback_vector(void)
1562 {
1563         int rc;
1564         uint64_t callback_via;
1565         if (xen_have_vector_callback) {
1566                 callback_via = HVM_CALLBACK_VECTOR(XEN_HVM_EVTCHN_CALLBACK);
1567                 rc = xen_set_callback_via(callback_via);
1568                 if (rc) {
1569                         printk(KERN_ERR "Request for Xen HVM callback vector"
1570                                         " failed.\n");
1571                         xen_have_vector_callback = 0;
1572                         return;
1573                 }
1574                 printk(KERN_INFO "Xen HVM callback vector for event delivery is "
1575                                 "enabled\n");
1576                 /* in the restore case the vector has already been allocated */
1577                 if (!test_bit(XEN_HVM_EVTCHN_CALLBACK, used_vectors))
1578                         alloc_intr_gate(XEN_HVM_EVTCHN_CALLBACK, xen_hvm_callback_vector);
1579         }
1580 }
1581 #else
1582 void xen_callback_vector(void) {}
1583 #endif
1584
1585 void __init xen_init_IRQ(void)
1586 {
1587         int i;
1588
1589         cpu_evtchn_mask_p = kcalloc(nr_cpu_ids, sizeof(struct cpu_evtchn_s),
1590                                     GFP_KERNEL);
1591         irq_info = kcalloc(nr_irqs, sizeof(*irq_info), GFP_KERNEL);
1592
1593         /* We are using nr_irqs as the maximum number of pirq available but
1594          * that number is actually chosen by Xen and we don't know exactly
1595          * what it is. Be careful choosing high pirq numbers. */
1596         pirq_to_irq = kcalloc(nr_irqs, sizeof(*pirq_to_irq), GFP_KERNEL);
1597         for (i = 0; i < nr_irqs; i++)
1598                 pirq_to_irq[i] = -1;
1599
1600         evtchn_to_irq = kcalloc(NR_EVENT_CHANNELS, sizeof(*evtchn_to_irq),
1601                                     GFP_KERNEL);
1602         for (i = 0; i < NR_EVENT_CHANNELS; i++)
1603                 evtchn_to_irq[i] = -1;
1604
1605         init_evtchn_cpu_bindings();
1606
1607         /* No event channels are 'live' right now. */
1608         for (i = 0; i < NR_EVENT_CHANNELS; i++)
1609                 mask_evtchn(i);
1610
1611         if (xen_hvm_domain()) {
1612                 xen_callback_vector();
1613                 native_init_IRQ();
1614                 /* pci_xen_hvm_init must be called after native_init_IRQ so that
1615                  * __acpi_register_gsi can point at the right function */
1616                 pci_xen_hvm_init();
1617         } else {
1618                 irq_ctx_init(smp_processor_id());
1619                 if (xen_initial_domain())
1620                         xen_setup_pirqs();
1621         }
1622 }