x86/xen: allow userspace access during hypercalls
[muen/linux.git] / drivers / xen / events / events_base.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 #define pr_fmt(fmt) "xen:" KBUILD_MODNAME ": " fmt
25
26 #include <linux/linkage.h>
27 #include <linux/interrupt.h>
28 #include <linux/irq.h>
29 #include <linux/moduleparam.h>
30 #include <linux/string.h>
31 #include <linux/bootmem.h>
32 #include <linux/slab.h>
33 #include <linux/irqnr.h>
34 #include <linux/pci.h>
35
36 #ifdef CONFIG_X86
37 #include <asm/desc.h>
38 #include <asm/ptrace.h>
39 #include <asm/irq.h>
40 #include <asm/io_apic.h>
41 #include <asm/i8259.h>
42 #include <asm/xen/pci.h>
43 #endif
44 #include <asm/sync_bitops.h>
45 #include <asm/xen/hypercall.h>
46 #include <asm/xen/hypervisor.h>
47 #include <xen/page.h>
48
49 #include <xen/xen.h>
50 #include <xen/hvm.h>
51 #include <xen/xen-ops.h>
52 #include <xen/events.h>
53 #include <xen/interface/xen.h>
54 #include <xen/interface/event_channel.h>
55 #include <xen/interface/hvm/hvm_op.h>
56 #include <xen/interface/hvm/params.h>
57 #include <xen/interface/physdev.h>
58 #include <xen/interface/sched.h>
59 #include <xen/interface/vcpu.h>
60 #include <asm/hw_irq.h>
61
62 #include "events_internal.h"
63
64 const struct evtchn_ops *evtchn_ops;
65
66 /*
67  * This lock protects updates to the following mapping and reference-count
68  * arrays. The lock does not need to be acquired to read the mapping tables.
69  */
70 static DEFINE_MUTEX(irq_mapping_update_lock);
71
72 static LIST_HEAD(xen_irq_list_head);
73
74 /* IRQ <-> VIRQ mapping. */
75 static DEFINE_PER_CPU(int [NR_VIRQS], virq_to_irq) = {[0 ... NR_VIRQS-1] = -1};
76
77 /* IRQ <-> IPI mapping */
78 static DEFINE_PER_CPU(int [XEN_NR_IPIS], ipi_to_irq) = {[0 ... XEN_NR_IPIS-1] = -1};
79
80 int **evtchn_to_irq;
81 #ifdef CONFIG_X86
82 static unsigned long *pirq_eoi_map;
83 #endif
84 static bool (*pirq_needs_eoi)(unsigned irq);
85
86 #define EVTCHN_ROW(e)  (e / (PAGE_SIZE/sizeof(**evtchn_to_irq)))
87 #define EVTCHN_COL(e)  (e % (PAGE_SIZE/sizeof(**evtchn_to_irq)))
88 #define EVTCHN_PER_ROW (PAGE_SIZE / sizeof(**evtchn_to_irq))
89
90 /* Xen will never allocate port zero for any purpose. */
91 #define VALID_EVTCHN(chn)       ((chn) != 0)
92
93 static struct irq_chip xen_dynamic_chip;
94 static struct irq_chip xen_percpu_chip;
95 static struct irq_chip xen_pirq_chip;
96 static void enable_dynirq(struct irq_data *data);
97 static void disable_dynirq(struct irq_data *data);
98
99 static void clear_evtchn_to_irq_row(unsigned row)
100 {
101         unsigned col;
102
103         for (col = 0; col < EVTCHN_PER_ROW; col++)
104                 evtchn_to_irq[row][col] = -1;
105 }
106
107 static void clear_evtchn_to_irq_all(void)
108 {
109         unsigned row;
110
111         for (row = 0; row < EVTCHN_ROW(xen_evtchn_max_channels()); row++) {
112                 if (evtchn_to_irq[row] == NULL)
113                         continue;
114                 clear_evtchn_to_irq_row(row);
115         }
116 }
117
118 static int set_evtchn_to_irq(unsigned evtchn, unsigned irq)
119 {
120         unsigned row;
121         unsigned col;
122
123         if (evtchn >= xen_evtchn_max_channels())
124                 return -EINVAL;
125
126         row = EVTCHN_ROW(evtchn);
127         col = EVTCHN_COL(evtchn);
128
129         if (evtchn_to_irq[row] == NULL) {
130                 /* Unallocated irq entries return -1 anyway */
131                 if (irq == -1)
132                         return 0;
133
134                 evtchn_to_irq[row] = (int *)get_zeroed_page(GFP_KERNEL);
135                 if (evtchn_to_irq[row] == NULL)
136                         return -ENOMEM;
137
138                 clear_evtchn_to_irq_row(row);
139         }
140
141         evtchn_to_irq[EVTCHN_ROW(evtchn)][EVTCHN_COL(evtchn)] = irq;
142         return 0;
143 }
144
145 int get_evtchn_to_irq(unsigned evtchn)
146 {
147         if (evtchn >= xen_evtchn_max_channels())
148                 return -1;
149         if (evtchn_to_irq[EVTCHN_ROW(evtchn)] == NULL)
150                 return -1;
151         return evtchn_to_irq[EVTCHN_ROW(evtchn)][EVTCHN_COL(evtchn)];
152 }
153
154 /* Get info for IRQ */
155 struct irq_info *info_for_irq(unsigned irq)
156 {
157         return irq_get_handler_data(irq);
158 }
159
160 /* Constructors for packed IRQ information. */
161 static int xen_irq_info_common_setup(struct irq_info *info,
162                                      unsigned irq,
163                                      enum xen_irq_type type,
164                                      unsigned evtchn,
165                                      unsigned short cpu)
166 {
167         int ret;
168
169         BUG_ON(info->type != IRQT_UNBOUND && info->type != type);
170
171         info->type = type;
172         info->irq = irq;
173         info->evtchn = evtchn;
174         info->cpu = cpu;
175
176         ret = set_evtchn_to_irq(evtchn, irq);
177         if (ret < 0)
178                 return ret;
179
180         irq_clear_status_flags(irq, IRQ_NOREQUEST|IRQ_NOAUTOEN);
181
182         return xen_evtchn_port_setup(info);
183 }
184
185 static int xen_irq_info_evtchn_setup(unsigned irq,
186                                      unsigned evtchn)
187 {
188         struct irq_info *info = info_for_irq(irq);
189
190         return xen_irq_info_common_setup(info, irq, IRQT_EVTCHN, evtchn, 0);
191 }
192
193 static int xen_irq_info_ipi_setup(unsigned cpu,
194                                   unsigned irq,
195                                   unsigned evtchn,
196                                   enum ipi_vector ipi)
197 {
198         struct irq_info *info = info_for_irq(irq);
199
200         info->u.ipi = ipi;
201
202         per_cpu(ipi_to_irq, cpu)[ipi] = irq;
203
204         return xen_irq_info_common_setup(info, irq, IRQT_IPI, evtchn, 0);
205 }
206
207 static int xen_irq_info_virq_setup(unsigned cpu,
208                                    unsigned irq,
209                                    unsigned evtchn,
210                                    unsigned virq)
211 {
212         struct irq_info *info = info_for_irq(irq);
213
214         info->u.virq = virq;
215
216         per_cpu(virq_to_irq, cpu)[virq] = irq;
217
218         return xen_irq_info_common_setup(info, irq, IRQT_VIRQ, evtchn, 0);
219 }
220
221 static int xen_irq_info_pirq_setup(unsigned irq,
222                                    unsigned evtchn,
223                                    unsigned pirq,
224                                    unsigned gsi,
225                                    uint16_t domid,
226                                    unsigned char flags)
227 {
228         struct irq_info *info = info_for_irq(irq);
229
230         info->u.pirq.pirq = pirq;
231         info->u.pirq.gsi = gsi;
232         info->u.pirq.domid = domid;
233         info->u.pirq.flags = flags;
234
235         return xen_irq_info_common_setup(info, irq, IRQT_PIRQ, evtchn, 0);
236 }
237
238 static void xen_irq_info_cleanup(struct irq_info *info)
239 {
240         set_evtchn_to_irq(info->evtchn, -1);
241         info->evtchn = 0;
242 }
243
244 /*
245  * Accessors for packed IRQ information.
246  */
247 unsigned int evtchn_from_irq(unsigned irq)
248 {
249         if (unlikely(WARN(irq >= nr_irqs, "Invalid irq %d!\n", irq)))
250                 return 0;
251
252         return info_for_irq(irq)->evtchn;
253 }
254
255 unsigned irq_from_evtchn(unsigned int evtchn)
256 {
257         return get_evtchn_to_irq(evtchn);
258 }
259 EXPORT_SYMBOL_GPL(irq_from_evtchn);
260
261 int irq_from_virq(unsigned int cpu, unsigned int virq)
262 {
263         return per_cpu(virq_to_irq, cpu)[virq];
264 }
265
266 static enum ipi_vector ipi_from_irq(unsigned irq)
267 {
268         struct irq_info *info = info_for_irq(irq);
269
270         BUG_ON(info == NULL);
271         BUG_ON(info->type != IRQT_IPI);
272
273         return info->u.ipi;
274 }
275
276 static unsigned virq_from_irq(unsigned irq)
277 {
278         struct irq_info *info = info_for_irq(irq);
279
280         BUG_ON(info == NULL);
281         BUG_ON(info->type != IRQT_VIRQ);
282
283         return info->u.virq;
284 }
285
286 static unsigned pirq_from_irq(unsigned irq)
287 {
288         struct irq_info *info = info_for_irq(irq);
289
290         BUG_ON(info == NULL);
291         BUG_ON(info->type != IRQT_PIRQ);
292
293         return info->u.pirq.pirq;
294 }
295
296 static enum xen_irq_type type_from_irq(unsigned irq)
297 {
298         return info_for_irq(irq)->type;
299 }
300
301 unsigned cpu_from_irq(unsigned irq)
302 {
303         return info_for_irq(irq)->cpu;
304 }
305
306 unsigned int cpu_from_evtchn(unsigned int evtchn)
307 {
308         int irq = get_evtchn_to_irq(evtchn);
309         unsigned ret = 0;
310
311         if (irq != -1)
312                 ret = cpu_from_irq(irq);
313
314         return ret;
315 }
316
317 #ifdef CONFIG_X86
318 static bool pirq_check_eoi_map(unsigned irq)
319 {
320         return test_bit(pirq_from_irq(irq), pirq_eoi_map);
321 }
322 #endif
323
324 static bool pirq_needs_eoi_flag(unsigned irq)
325 {
326         struct irq_info *info = info_for_irq(irq);
327         BUG_ON(info->type != IRQT_PIRQ);
328
329         return info->u.pirq.flags & PIRQ_NEEDS_EOI;
330 }
331
332 static void bind_evtchn_to_cpu(unsigned int chn, unsigned int cpu)
333 {
334         int irq = get_evtchn_to_irq(chn);
335         struct irq_info *info = info_for_irq(irq);
336
337         BUG_ON(irq == -1);
338 #ifdef CONFIG_SMP
339         cpumask_copy(irq_get_affinity_mask(irq), cpumask_of(cpu));
340 #endif
341         xen_evtchn_port_bind_to_cpu(info, cpu);
342
343         info->cpu = cpu;
344 }
345
346 static void xen_evtchn_mask_all(void)
347 {
348         unsigned int evtchn;
349
350         for (evtchn = 0; evtchn < xen_evtchn_nr_channels(); evtchn++)
351                 mask_evtchn(evtchn);
352 }
353
354 /**
355  * notify_remote_via_irq - send event to remote end of event channel via irq
356  * @irq: irq of event channel to send event to
357  *
358  * Unlike notify_remote_via_evtchn(), this is safe to use across
359  * save/restore. Notifications on a broken connection are silently
360  * dropped.
361  */
362 void notify_remote_via_irq(int irq)
363 {
364         int evtchn = evtchn_from_irq(irq);
365
366         if (VALID_EVTCHN(evtchn))
367                 notify_remote_via_evtchn(evtchn);
368 }
369 EXPORT_SYMBOL_GPL(notify_remote_via_irq);
370
371 static void xen_irq_init(unsigned irq)
372 {
373         struct irq_info *info;
374 #ifdef CONFIG_SMP
375         /* By default all event channels notify CPU#0. */
376         cpumask_copy(irq_get_affinity_mask(irq), cpumask_of(0));
377 #endif
378
379         info = kzalloc(sizeof(*info), GFP_KERNEL);
380         if (info == NULL)
381                 panic("Unable to allocate metadata for IRQ%d\n", irq);
382
383         info->type = IRQT_UNBOUND;
384         info->refcnt = -1;
385
386         irq_set_handler_data(irq, info);
387
388         list_add_tail(&info->list, &xen_irq_list_head);
389 }
390
391 static int __must_check xen_allocate_irqs_dynamic(int nvec)
392 {
393         int i, irq = irq_alloc_descs(-1, 0, nvec, -1);
394
395         if (irq >= 0) {
396                 for (i = 0; i < nvec; i++)
397                         xen_irq_init(irq + i);
398         }
399
400         return irq;
401 }
402
403 static inline int __must_check xen_allocate_irq_dynamic(void)
404 {
405
406         return xen_allocate_irqs_dynamic(1);
407 }
408
409 static int __must_check xen_allocate_irq_gsi(unsigned gsi)
410 {
411         int irq;
412
413         /*
414          * A PV guest has no concept of a GSI (since it has no ACPI
415          * nor access to/knowledge of the physical APICs). Therefore
416          * all IRQs are dynamically allocated from the entire IRQ
417          * space.
418          */
419         if (xen_pv_domain() && !xen_initial_domain())
420                 return xen_allocate_irq_dynamic();
421
422         /* Legacy IRQ descriptors are already allocated by the arch. */
423         if (gsi < nr_legacy_irqs())
424                 irq = gsi;
425         else
426                 irq = irq_alloc_desc_at(gsi, -1);
427
428         xen_irq_init(irq);
429
430         return irq;
431 }
432
433 static void xen_free_irq(unsigned irq)
434 {
435         struct irq_info *info = irq_get_handler_data(irq);
436
437         if (WARN_ON(!info))
438                 return;
439
440         list_del(&info->list);
441
442         irq_set_handler_data(irq, NULL);
443
444         WARN_ON(info->refcnt > 0);
445
446         kfree(info);
447
448         /* Legacy IRQ descriptors are managed by the arch. */
449         if (irq < nr_legacy_irqs())
450                 return;
451
452         irq_free_desc(irq);
453 }
454
455 static void xen_evtchn_close(unsigned int port)
456 {
457         struct evtchn_close close;
458
459         close.port = port;
460         if (HYPERVISOR_event_channel_op(EVTCHNOP_close, &close) != 0)
461                 BUG();
462 }
463
464 static void pirq_query_unmask(int irq)
465 {
466         struct physdev_irq_status_query irq_status;
467         struct irq_info *info = info_for_irq(irq);
468
469         BUG_ON(info->type != IRQT_PIRQ);
470
471         irq_status.irq = pirq_from_irq(irq);
472         if (HYPERVISOR_physdev_op(PHYSDEVOP_irq_status_query, &irq_status))
473                 irq_status.flags = 0;
474
475         info->u.pirq.flags &= ~PIRQ_NEEDS_EOI;
476         if (irq_status.flags & XENIRQSTAT_needs_eoi)
477                 info->u.pirq.flags |= PIRQ_NEEDS_EOI;
478 }
479
480 static void eoi_pirq(struct irq_data *data)
481 {
482         int evtchn = evtchn_from_irq(data->irq);
483         struct physdev_eoi eoi = { .irq = pirq_from_irq(data->irq) };
484         int rc = 0;
485
486         if (!VALID_EVTCHN(evtchn))
487                 return;
488
489         if (unlikely(irqd_is_setaffinity_pending(data)) &&
490             likely(!irqd_irq_disabled(data))) {
491                 int masked = test_and_set_mask(evtchn);
492
493                 clear_evtchn(evtchn);
494
495                 irq_move_masked_irq(data);
496
497                 if (!masked)
498                         unmask_evtchn(evtchn);
499         } else
500                 clear_evtchn(evtchn);
501
502         if (pirq_needs_eoi(data->irq)) {
503                 rc = HYPERVISOR_physdev_op(PHYSDEVOP_eoi, &eoi);
504                 WARN_ON(rc);
505         }
506 }
507
508 static void mask_ack_pirq(struct irq_data *data)
509 {
510         disable_dynirq(data);
511         eoi_pirq(data);
512 }
513
514 static unsigned int __startup_pirq(unsigned int irq)
515 {
516         struct evtchn_bind_pirq bind_pirq;
517         struct irq_info *info = info_for_irq(irq);
518         int evtchn = evtchn_from_irq(irq);
519         int rc;
520
521         BUG_ON(info->type != IRQT_PIRQ);
522
523         if (VALID_EVTCHN(evtchn))
524                 goto out;
525
526         bind_pirq.pirq = pirq_from_irq(irq);
527         /* NB. We are happy to share unless we are probing. */
528         bind_pirq.flags = info->u.pirq.flags & PIRQ_SHAREABLE ?
529                                         BIND_PIRQ__WILL_SHARE : 0;
530         rc = HYPERVISOR_event_channel_op(EVTCHNOP_bind_pirq, &bind_pirq);
531         if (rc != 0) {
532                 pr_warn("Failed to obtain physical IRQ %d\n", irq);
533                 return 0;
534         }
535         evtchn = bind_pirq.port;
536
537         pirq_query_unmask(irq);
538
539         rc = set_evtchn_to_irq(evtchn, irq);
540         if (rc)
541                 goto err;
542
543         info->evtchn = evtchn;
544         bind_evtchn_to_cpu(evtchn, 0);
545
546         rc = xen_evtchn_port_setup(info);
547         if (rc)
548                 goto err;
549
550 out:
551         unmask_evtchn(evtchn);
552         eoi_pirq(irq_get_irq_data(irq));
553
554         return 0;
555
556 err:
557         pr_err("irq%d: Failed to set port to irq mapping (%d)\n", irq, rc);
558         xen_evtchn_close(evtchn);
559         return 0;
560 }
561
562 static unsigned int startup_pirq(struct irq_data *data)
563 {
564         return __startup_pirq(data->irq);
565 }
566
567 static void shutdown_pirq(struct irq_data *data)
568 {
569         unsigned int irq = data->irq;
570         struct irq_info *info = info_for_irq(irq);
571         unsigned evtchn = evtchn_from_irq(irq);
572
573         BUG_ON(info->type != IRQT_PIRQ);
574
575         if (!VALID_EVTCHN(evtchn))
576                 return;
577
578         mask_evtchn(evtchn);
579         xen_evtchn_close(evtchn);
580         xen_irq_info_cleanup(info);
581 }
582
583 static void enable_pirq(struct irq_data *data)
584 {
585         startup_pirq(data);
586 }
587
588 static void disable_pirq(struct irq_data *data)
589 {
590         disable_dynirq(data);
591 }
592
593 int xen_irq_from_gsi(unsigned gsi)
594 {
595         struct irq_info *info;
596
597         list_for_each_entry(info, &xen_irq_list_head, list) {
598                 if (info->type != IRQT_PIRQ)
599                         continue;
600
601                 if (info->u.pirq.gsi == gsi)
602                         return info->irq;
603         }
604
605         return -1;
606 }
607 EXPORT_SYMBOL_GPL(xen_irq_from_gsi);
608
609 static void __unbind_from_irq(unsigned int irq)
610 {
611         int evtchn = evtchn_from_irq(irq);
612         struct irq_info *info = irq_get_handler_data(irq);
613
614         if (info->refcnt > 0) {
615                 info->refcnt--;
616                 if (info->refcnt != 0)
617                         return;
618         }
619
620         if (VALID_EVTCHN(evtchn)) {
621                 unsigned int cpu = cpu_from_irq(irq);
622
623                 xen_evtchn_close(evtchn);
624
625                 switch (type_from_irq(irq)) {
626                 case IRQT_VIRQ:
627                         per_cpu(virq_to_irq, cpu)[virq_from_irq(irq)] = -1;
628                         break;
629                 case IRQT_IPI:
630                         per_cpu(ipi_to_irq, cpu)[ipi_from_irq(irq)] = -1;
631                         break;
632                 default:
633                         break;
634                 }
635
636                 xen_irq_info_cleanup(info);
637         }
638
639         BUG_ON(info_for_irq(irq)->type == IRQT_UNBOUND);
640
641         xen_free_irq(irq);
642 }
643
644 /*
645  * Do not make any assumptions regarding the relationship between the
646  * IRQ number returned here and the Xen pirq argument.
647  *
648  * Note: We don't assign an event channel until the irq actually started
649  * up.  Return an existing irq if we've already got one for the gsi.
650  *
651  * Shareable implies level triggered, not shareable implies edge
652  * triggered here.
653  */
654 int xen_bind_pirq_gsi_to_irq(unsigned gsi,
655                              unsigned pirq, int shareable, char *name)
656 {
657         int irq = -1;
658         struct physdev_irq irq_op;
659         int ret;
660
661         mutex_lock(&irq_mapping_update_lock);
662
663         irq = xen_irq_from_gsi(gsi);
664         if (irq != -1) {
665                 pr_info("%s: returning irq %d for gsi %u\n",
666                         __func__, irq, gsi);
667                 goto out;
668         }
669
670         irq = xen_allocate_irq_gsi(gsi);
671         if (irq < 0)
672                 goto out;
673
674         irq_op.irq = irq;
675         irq_op.vector = 0;
676
677         /* Only the privileged domain can do this. For non-priv, the pcifront
678          * driver provides a PCI bus that does the call to do exactly
679          * this in the priv domain. */
680         if (xen_initial_domain() &&
681             HYPERVISOR_physdev_op(PHYSDEVOP_alloc_irq_vector, &irq_op)) {
682                 xen_free_irq(irq);
683                 irq = -ENOSPC;
684                 goto out;
685         }
686
687         ret = xen_irq_info_pirq_setup(irq, 0, pirq, gsi, DOMID_SELF,
688                                shareable ? PIRQ_SHAREABLE : 0);
689         if (ret < 0) {
690                 __unbind_from_irq(irq);
691                 irq = ret;
692                 goto out;
693         }
694
695         pirq_query_unmask(irq);
696         /* We try to use the handler with the appropriate semantic for the
697          * type of interrupt: if the interrupt is an edge triggered
698          * interrupt we use handle_edge_irq.
699          *
700          * On the other hand if the interrupt is level triggered we use
701          * handle_fasteoi_irq like the native code does for this kind of
702          * interrupts.
703          *
704          * Depending on the Xen version, pirq_needs_eoi might return true
705          * not only for level triggered interrupts but for edge triggered
706          * interrupts too. In any case Xen always honors the eoi mechanism,
707          * not injecting any more pirqs of the same kind if the first one
708          * hasn't received an eoi yet. Therefore using the fasteoi handler
709          * is the right choice either way.
710          */
711         if (shareable)
712                 irq_set_chip_and_handler_name(irq, &xen_pirq_chip,
713                                 handle_fasteoi_irq, name);
714         else
715                 irq_set_chip_and_handler_name(irq, &xen_pirq_chip,
716                                 handle_edge_irq, name);
717
718 out:
719         mutex_unlock(&irq_mapping_update_lock);
720
721         return irq;
722 }
723
724 #ifdef CONFIG_PCI_MSI
725 int xen_allocate_pirq_msi(struct pci_dev *dev, struct msi_desc *msidesc)
726 {
727         int rc;
728         struct physdev_get_free_pirq op_get_free_pirq;
729
730         op_get_free_pirq.type = MAP_PIRQ_TYPE_MSI;
731         rc = HYPERVISOR_physdev_op(PHYSDEVOP_get_free_pirq, &op_get_free_pirq);
732
733         WARN_ONCE(rc == -ENOSYS,
734                   "hypervisor does not support the PHYSDEVOP_get_free_pirq interface\n");
735
736         return rc ? -1 : op_get_free_pirq.pirq;
737 }
738
739 int xen_bind_pirq_msi_to_irq(struct pci_dev *dev, struct msi_desc *msidesc,
740                              int pirq, int nvec, const char *name, domid_t domid)
741 {
742         int i, irq, ret;
743
744         mutex_lock(&irq_mapping_update_lock);
745
746         irq = xen_allocate_irqs_dynamic(nvec);
747         if (irq < 0)
748                 goto out;
749
750         for (i = 0; i < nvec; i++) {
751                 irq_set_chip_and_handler_name(irq + i, &xen_pirq_chip, handle_edge_irq, name);
752
753                 ret = xen_irq_info_pirq_setup(irq + i, 0, pirq + i, 0, domid,
754                                               i == 0 ? 0 : PIRQ_MSI_GROUP);
755                 if (ret < 0)
756                         goto error_irq;
757         }
758
759         ret = irq_set_msi_desc(irq, msidesc);
760         if (ret < 0)
761                 goto error_irq;
762 out:
763         mutex_unlock(&irq_mapping_update_lock);
764         return irq;
765 error_irq:
766         for (; i >= 0; i--)
767                 __unbind_from_irq(irq + i);
768         mutex_unlock(&irq_mapping_update_lock);
769         return ret;
770 }
771 #endif
772
773 int xen_destroy_irq(int irq)
774 {
775         struct physdev_unmap_pirq unmap_irq;
776         struct irq_info *info = info_for_irq(irq);
777         int rc = -ENOENT;
778
779         mutex_lock(&irq_mapping_update_lock);
780
781         /*
782          * If trying to remove a vector in a MSI group different
783          * than the first one skip the PIRQ unmap unless this vector
784          * is the first one in the group.
785          */
786         if (xen_initial_domain() && !(info->u.pirq.flags & PIRQ_MSI_GROUP)) {
787                 unmap_irq.pirq = info->u.pirq.pirq;
788                 unmap_irq.domid = info->u.pirq.domid;
789                 rc = HYPERVISOR_physdev_op(PHYSDEVOP_unmap_pirq, &unmap_irq);
790                 /* If another domain quits without making the pci_disable_msix
791                  * call, the Xen hypervisor takes care of freeing the PIRQs
792                  * (free_domain_pirqs).
793                  */
794                 if ((rc == -ESRCH && info->u.pirq.domid != DOMID_SELF))
795                         pr_info("domain %d does not have %d anymore\n",
796                                 info->u.pirq.domid, info->u.pirq.pirq);
797                 else if (rc) {
798                         pr_warn("unmap irq failed %d\n", rc);
799                         goto out;
800                 }
801         }
802
803         xen_free_irq(irq);
804
805 out:
806         mutex_unlock(&irq_mapping_update_lock);
807         return rc;
808 }
809
810 int xen_irq_from_pirq(unsigned pirq)
811 {
812         int irq;
813
814         struct irq_info *info;
815
816         mutex_lock(&irq_mapping_update_lock);
817
818         list_for_each_entry(info, &xen_irq_list_head, list) {
819                 if (info->type != IRQT_PIRQ)
820                         continue;
821                 irq = info->irq;
822                 if (info->u.pirq.pirq == pirq)
823                         goto out;
824         }
825         irq = -1;
826 out:
827         mutex_unlock(&irq_mapping_update_lock);
828
829         return irq;
830 }
831
832
833 int xen_pirq_from_irq(unsigned irq)
834 {
835         return pirq_from_irq(irq);
836 }
837 EXPORT_SYMBOL_GPL(xen_pirq_from_irq);
838
839 int bind_evtchn_to_irq(unsigned int evtchn)
840 {
841         int irq;
842         int ret;
843
844         if (evtchn >= xen_evtchn_max_channels())
845                 return -ENOMEM;
846
847         mutex_lock(&irq_mapping_update_lock);
848
849         irq = get_evtchn_to_irq(evtchn);
850
851         if (irq == -1) {
852                 irq = xen_allocate_irq_dynamic();
853                 if (irq < 0)
854                         goto out;
855
856                 irq_set_chip_and_handler_name(irq, &xen_dynamic_chip,
857                                               handle_edge_irq, "event");
858
859                 ret = xen_irq_info_evtchn_setup(irq, evtchn);
860                 if (ret < 0) {
861                         __unbind_from_irq(irq);
862                         irq = ret;
863                         goto out;
864                 }
865                 /* New interdomain events are bound to VCPU 0. */
866                 bind_evtchn_to_cpu(evtchn, 0);
867         } else {
868                 struct irq_info *info = info_for_irq(irq);
869                 WARN_ON(info == NULL || info->type != IRQT_EVTCHN);
870         }
871
872 out:
873         mutex_unlock(&irq_mapping_update_lock);
874
875         return irq;
876 }
877 EXPORT_SYMBOL_GPL(bind_evtchn_to_irq);
878
879 static int bind_ipi_to_irq(unsigned int ipi, unsigned int cpu)
880 {
881         struct evtchn_bind_ipi bind_ipi;
882         int evtchn, irq;
883         int ret;
884
885         mutex_lock(&irq_mapping_update_lock);
886
887         irq = per_cpu(ipi_to_irq, cpu)[ipi];
888
889         if (irq == -1) {
890                 irq = xen_allocate_irq_dynamic();
891                 if (irq < 0)
892                         goto out;
893
894                 irq_set_chip_and_handler_name(irq, &xen_percpu_chip,
895                                               handle_percpu_irq, "ipi");
896
897                 bind_ipi.vcpu = xen_vcpu_nr(cpu);
898                 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_ipi,
899                                                 &bind_ipi) != 0)
900                         BUG();
901                 evtchn = bind_ipi.port;
902
903                 ret = xen_irq_info_ipi_setup(cpu, irq, evtchn, ipi);
904                 if (ret < 0) {
905                         __unbind_from_irq(irq);
906                         irq = ret;
907                         goto out;
908                 }
909                 bind_evtchn_to_cpu(evtchn, cpu);
910         } else {
911                 struct irq_info *info = info_for_irq(irq);
912                 WARN_ON(info == NULL || info->type != IRQT_IPI);
913         }
914
915  out:
916         mutex_unlock(&irq_mapping_update_lock);
917         return irq;
918 }
919
920 int bind_interdomain_evtchn_to_irq(unsigned int remote_domain,
921                                    unsigned int remote_port)
922 {
923         struct evtchn_bind_interdomain bind_interdomain;
924         int err;
925
926         bind_interdomain.remote_dom  = remote_domain;
927         bind_interdomain.remote_port = remote_port;
928
929         err = HYPERVISOR_event_channel_op(EVTCHNOP_bind_interdomain,
930                                           &bind_interdomain);
931
932         return err ? : bind_evtchn_to_irq(bind_interdomain.local_port);
933 }
934 EXPORT_SYMBOL_GPL(bind_interdomain_evtchn_to_irq);
935
936 static int find_virq(unsigned int virq, unsigned int cpu)
937 {
938         struct evtchn_status status;
939         int port, rc = -ENOENT;
940
941         memset(&status, 0, sizeof(status));
942         for (port = 0; port < xen_evtchn_max_channels(); port++) {
943                 status.dom = DOMID_SELF;
944                 status.port = port;
945                 rc = HYPERVISOR_event_channel_op(EVTCHNOP_status, &status);
946                 if (rc < 0)
947                         continue;
948                 if (status.status != EVTCHNSTAT_virq)
949                         continue;
950                 if (status.u.virq == virq && status.vcpu == xen_vcpu_nr(cpu)) {
951                         rc = port;
952                         break;
953                 }
954         }
955         return rc;
956 }
957
958 /**
959  * xen_evtchn_nr_channels - number of usable event channel ports
960  *
961  * This may be less than the maximum supported by the current
962  * hypervisor ABI. Use xen_evtchn_max_channels() for the maximum
963  * supported.
964  */
965 unsigned xen_evtchn_nr_channels(void)
966 {
967         return evtchn_ops->nr_channels();
968 }
969 EXPORT_SYMBOL_GPL(xen_evtchn_nr_channels);
970
971 int bind_virq_to_irq(unsigned int virq, unsigned int cpu, bool percpu)
972 {
973         struct evtchn_bind_virq bind_virq;
974         int evtchn, irq, ret;
975
976         mutex_lock(&irq_mapping_update_lock);
977
978         irq = per_cpu(virq_to_irq, cpu)[virq];
979
980         if (irq == -1) {
981                 irq = xen_allocate_irq_dynamic();
982                 if (irq < 0)
983                         goto out;
984
985                 if (percpu)
986                         irq_set_chip_and_handler_name(irq, &xen_percpu_chip,
987                                                       handle_percpu_irq, "virq");
988                 else
989                         irq_set_chip_and_handler_name(irq, &xen_dynamic_chip,
990                                                       handle_edge_irq, "virq");
991
992                 bind_virq.virq = virq;
993                 bind_virq.vcpu = xen_vcpu_nr(cpu);
994                 ret = HYPERVISOR_event_channel_op(EVTCHNOP_bind_virq,
995                                                 &bind_virq);
996                 if (ret == 0)
997                         evtchn = bind_virq.port;
998                 else {
999                         if (ret == -EEXIST)
1000                                 ret = find_virq(virq, cpu);
1001                         BUG_ON(ret < 0);
1002                         evtchn = ret;
1003                 }
1004
1005                 ret = xen_irq_info_virq_setup(cpu, irq, evtchn, virq);
1006                 if (ret < 0) {
1007                         __unbind_from_irq(irq);
1008                         irq = ret;
1009                         goto out;
1010                 }
1011
1012                 bind_evtchn_to_cpu(evtchn, cpu);
1013         } else {
1014                 struct irq_info *info = info_for_irq(irq);
1015                 WARN_ON(info == NULL || info->type != IRQT_VIRQ);
1016         }
1017
1018 out:
1019         mutex_unlock(&irq_mapping_update_lock);
1020
1021         return irq;
1022 }
1023
1024 static void unbind_from_irq(unsigned int irq)
1025 {
1026         mutex_lock(&irq_mapping_update_lock);
1027         __unbind_from_irq(irq);
1028         mutex_unlock(&irq_mapping_update_lock);
1029 }
1030
1031 int bind_evtchn_to_irqhandler(unsigned int evtchn,
1032                               irq_handler_t handler,
1033                               unsigned long irqflags,
1034                               const char *devname, void *dev_id)
1035 {
1036         int irq, retval;
1037
1038         irq = bind_evtchn_to_irq(evtchn);
1039         if (irq < 0)
1040                 return irq;
1041         retval = request_irq(irq, handler, irqflags, devname, dev_id);
1042         if (retval != 0) {
1043                 unbind_from_irq(irq);
1044                 return retval;
1045         }
1046
1047         return irq;
1048 }
1049 EXPORT_SYMBOL_GPL(bind_evtchn_to_irqhandler);
1050
1051 int bind_interdomain_evtchn_to_irqhandler(unsigned int remote_domain,
1052                                           unsigned int remote_port,
1053                                           irq_handler_t handler,
1054                                           unsigned long irqflags,
1055                                           const char *devname,
1056                                           void *dev_id)
1057 {
1058         int irq, retval;
1059
1060         irq = bind_interdomain_evtchn_to_irq(remote_domain, remote_port);
1061         if (irq < 0)
1062                 return irq;
1063
1064         retval = request_irq(irq, handler, irqflags, devname, dev_id);
1065         if (retval != 0) {
1066                 unbind_from_irq(irq);
1067                 return retval;
1068         }
1069
1070         return irq;
1071 }
1072 EXPORT_SYMBOL_GPL(bind_interdomain_evtchn_to_irqhandler);
1073
1074 int bind_virq_to_irqhandler(unsigned int virq, unsigned int cpu,
1075                             irq_handler_t handler,
1076                             unsigned long irqflags, const char *devname, void *dev_id)
1077 {
1078         int irq, retval;
1079
1080         irq = bind_virq_to_irq(virq, cpu, irqflags & IRQF_PERCPU);
1081         if (irq < 0)
1082                 return irq;
1083         retval = request_irq(irq, handler, irqflags, devname, dev_id);
1084         if (retval != 0) {
1085                 unbind_from_irq(irq);
1086                 return retval;
1087         }
1088
1089         return irq;
1090 }
1091 EXPORT_SYMBOL_GPL(bind_virq_to_irqhandler);
1092
1093 int bind_ipi_to_irqhandler(enum ipi_vector ipi,
1094                            unsigned int cpu,
1095                            irq_handler_t handler,
1096                            unsigned long irqflags,
1097                            const char *devname,
1098                            void *dev_id)
1099 {
1100         int irq, retval;
1101
1102         irq = bind_ipi_to_irq(ipi, cpu);
1103         if (irq < 0)
1104                 return irq;
1105
1106         irqflags |= IRQF_NO_SUSPEND | IRQF_FORCE_RESUME | IRQF_EARLY_RESUME;
1107         retval = request_irq(irq, handler, irqflags, devname, dev_id);
1108         if (retval != 0) {
1109                 unbind_from_irq(irq);
1110                 return retval;
1111         }
1112
1113         return irq;
1114 }
1115
1116 void unbind_from_irqhandler(unsigned int irq, void *dev_id)
1117 {
1118         struct irq_info *info = irq_get_handler_data(irq);
1119
1120         if (WARN_ON(!info))
1121                 return;
1122         free_irq(irq, dev_id);
1123         unbind_from_irq(irq);
1124 }
1125 EXPORT_SYMBOL_GPL(unbind_from_irqhandler);
1126
1127 /**
1128  * xen_set_irq_priority() - set an event channel priority.
1129  * @irq:irq bound to an event channel.
1130  * @priority: priority between XEN_IRQ_PRIORITY_MAX and XEN_IRQ_PRIORITY_MIN.
1131  */
1132 int xen_set_irq_priority(unsigned irq, unsigned priority)
1133 {
1134         struct evtchn_set_priority set_priority;
1135
1136         set_priority.port = evtchn_from_irq(irq);
1137         set_priority.priority = priority;
1138
1139         return HYPERVISOR_event_channel_op(EVTCHNOP_set_priority,
1140                                            &set_priority);
1141 }
1142 EXPORT_SYMBOL_GPL(xen_set_irq_priority);
1143
1144 int evtchn_make_refcounted(unsigned int evtchn)
1145 {
1146         int irq = get_evtchn_to_irq(evtchn);
1147         struct irq_info *info;
1148
1149         if (irq == -1)
1150                 return -ENOENT;
1151
1152         info = irq_get_handler_data(irq);
1153
1154         if (!info)
1155                 return -ENOENT;
1156
1157         WARN_ON(info->refcnt != -1);
1158
1159         info->refcnt = 1;
1160
1161         return 0;
1162 }
1163 EXPORT_SYMBOL_GPL(evtchn_make_refcounted);
1164
1165 int evtchn_get(unsigned int evtchn)
1166 {
1167         int irq;
1168         struct irq_info *info;
1169         int err = -ENOENT;
1170
1171         if (evtchn >= xen_evtchn_max_channels())
1172                 return -EINVAL;
1173
1174         mutex_lock(&irq_mapping_update_lock);
1175
1176         irq = get_evtchn_to_irq(evtchn);
1177         if (irq == -1)
1178                 goto done;
1179
1180         info = irq_get_handler_data(irq);
1181
1182         if (!info)
1183                 goto done;
1184
1185         err = -EINVAL;
1186         if (info->refcnt <= 0)
1187                 goto done;
1188
1189         info->refcnt++;
1190         err = 0;
1191  done:
1192         mutex_unlock(&irq_mapping_update_lock);
1193
1194         return err;
1195 }
1196 EXPORT_SYMBOL_GPL(evtchn_get);
1197
1198 void evtchn_put(unsigned int evtchn)
1199 {
1200         int irq = get_evtchn_to_irq(evtchn);
1201         if (WARN_ON(irq == -1))
1202                 return;
1203         unbind_from_irq(irq);
1204 }
1205 EXPORT_SYMBOL_GPL(evtchn_put);
1206
1207 void xen_send_IPI_one(unsigned int cpu, enum ipi_vector vector)
1208 {
1209         int irq;
1210
1211 #ifdef CONFIG_X86
1212         if (unlikely(vector == XEN_NMI_VECTOR)) {
1213                 int rc =  HYPERVISOR_vcpu_op(VCPUOP_send_nmi, xen_vcpu_nr(cpu),
1214                                              NULL);
1215                 if (rc < 0)
1216                         printk(KERN_WARNING "Sending nmi to CPU%d failed (rc:%d)\n", cpu, rc);
1217                 return;
1218         }
1219 #endif
1220         irq = per_cpu(ipi_to_irq, cpu)[vector];
1221         BUG_ON(irq < 0);
1222         notify_remote_via_irq(irq);
1223 }
1224
1225 static DEFINE_PER_CPU(unsigned, xed_nesting_count);
1226
1227 static void __xen_evtchn_do_upcall(void)
1228 {
1229         struct vcpu_info *vcpu_info = __this_cpu_read(xen_vcpu);
1230         int cpu = get_cpu();
1231         unsigned count;
1232
1233         do {
1234                 vcpu_info->evtchn_upcall_pending = 0;
1235
1236                 if (__this_cpu_inc_return(xed_nesting_count) - 1)
1237                         goto out;
1238
1239                 xen_evtchn_handle_events(cpu);
1240
1241                 BUG_ON(!irqs_disabled());
1242
1243                 count = __this_cpu_read(xed_nesting_count);
1244                 __this_cpu_write(xed_nesting_count, 0);
1245         } while (count != 1 || vcpu_info->evtchn_upcall_pending);
1246
1247 out:
1248
1249         put_cpu();
1250 }
1251
1252 void xen_evtchn_do_upcall(struct pt_regs *regs)
1253 {
1254         struct pt_regs *old_regs = set_irq_regs(regs);
1255
1256         irq_enter();
1257 #ifdef CONFIG_X86
1258         inc_irq_stat(irq_hv_callback_count);
1259 #endif
1260
1261         __xen_evtchn_do_upcall();
1262
1263         irq_exit();
1264         set_irq_regs(old_regs);
1265 }
1266
1267 void xen_hvm_evtchn_do_upcall(void)
1268 {
1269         __xen_evtchn_do_upcall();
1270 }
1271 EXPORT_SYMBOL_GPL(xen_hvm_evtchn_do_upcall);
1272
1273 /* Rebind a new event channel to an existing irq. */
1274 void rebind_evtchn_irq(int evtchn, int irq)
1275 {
1276         struct irq_info *info = info_for_irq(irq);
1277
1278         if (WARN_ON(!info))
1279                 return;
1280
1281         /* Make sure the irq is masked, since the new event channel
1282            will also be masked. */
1283         disable_irq(irq);
1284
1285         mutex_lock(&irq_mapping_update_lock);
1286
1287         /* After resume the irq<->evtchn mappings are all cleared out */
1288         BUG_ON(get_evtchn_to_irq(evtchn) != -1);
1289         /* Expect irq to have been bound before,
1290            so there should be a proper type */
1291         BUG_ON(info->type == IRQT_UNBOUND);
1292
1293         (void)xen_irq_info_evtchn_setup(irq, evtchn);
1294
1295         mutex_unlock(&irq_mapping_update_lock);
1296
1297         bind_evtchn_to_cpu(evtchn, info->cpu);
1298         /* This will be deferred until interrupt is processed */
1299         irq_set_affinity(irq, cpumask_of(info->cpu));
1300
1301         /* Unmask the event channel. */
1302         enable_irq(irq);
1303 }
1304
1305 /* Rebind an evtchn so that it gets delivered to a specific cpu */
1306 int xen_rebind_evtchn_to_cpu(int evtchn, unsigned tcpu)
1307 {
1308         struct evtchn_bind_vcpu bind_vcpu;
1309         int masked;
1310
1311         if (!VALID_EVTCHN(evtchn))
1312                 return -1;
1313
1314         if (!xen_support_evtchn_rebind())
1315                 return -1;
1316
1317         /* Send future instances of this interrupt to other vcpu. */
1318         bind_vcpu.port = evtchn;
1319         bind_vcpu.vcpu = xen_vcpu_nr(tcpu);
1320
1321         /*
1322          * Mask the event while changing the VCPU binding to prevent
1323          * it being delivered on an unexpected VCPU.
1324          */
1325         masked = test_and_set_mask(evtchn);
1326
1327         /*
1328          * If this fails, it usually just indicates that we're dealing with a
1329          * virq or IPI channel, which don't actually need to be rebound. Ignore
1330          * it, but don't do the xenlinux-level rebind in that case.
1331          */
1332         if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_vcpu, &bind_vcpu) >= 0)
1333                 bind_evtchn_to_cpu(evtchn, tcpu);
1334
1335         if (!masked)
1336                 unmask_evtchn(evtchn);
1337
1338         return 0;
1339 }
1340 EXPORT_SYMBOL_GPL(xen_rebind_evtchn_to_cpu);
1341
1342 static int set_affinity_irq(struct irq_data *data, const struct cpumask *dest,
1343                             bool force)
1344 {
1345         unsigned tcpu = cpumask_first_and(dest, cpu_online_mask);
1346
1347         return xen_rebind_evtchn_to_cpu(evtchn_from_irq(data->irq), tcpu);
1348 }
1349
1350 static void enable_dynirq(struct irq_data *data)
1351 {
1352         int evtchn = evtchn_from_irq(data->irq);
1353
1354         if (VALID_EVTCHN(evtchn))
1355                 unmask_evtchn(evtchn);
1356 }
1357
1358 static void disable_dynirq(struct irq_data *data)
1359 {
1360         int evtchn = evtchn_from_irq(data->irq);
1361
1362         if (VALID_EVTCHN(evtchn))
1363                 mask_evtchn(evtchn);
1364 }
1365
1366 static void ack_dynirq(struct irq_data *data)
1367 {
1368         int evtchn = evtchn_from_irq(data->irq);
1369
1370         if (!VALID_EVTCHN(evtchn))
1371                 return;
1372
1373         if (unlikely(irqd_is_setaffinity_pending(data)) &&
1374             likely(!irqd_irq_disabled(data))) {
1375                 int masked = test_and_set_mask(evtchn);
1376
1377                 clear_evtchn(evtchn);
1378
1379                 irq_move_masked_irq(data);
1380
1381                 if (!masked)
1382                         unmask_evtchn(evtchn);
1383         } else
1384                 clear_evtchn(evtchn);
1385 }
1386
1387 static void mask_ack_dynirq(struct irq_data *data)
1388 {
1389         disable_dynirq(data);
1390         ack_dynirq(data);
1391 }
1392
1393 static int retrigger_dynirq(struct irq_data *data)
1394 {
1395         unsigned int evtchn = evtchn_from_irq(data->irq);
1396         int masked;
1397
1398         if (!VALID_EVTCHN(evtchn))
1399                 return 0;
1400
1401         masked = test_and_set_mask(evtchn);
1402         set_evtchn(evtchn);
1403         if (!masked)
1404                 unmask_evtchn(evtchn);
1405
1406         return 1;
1407 }
1408
1409 static void restore_pirqs(void)
1410 {
1411         int pirq, rc, irq, gsi;
1412         struct physdev_map_pirq map_irq;
1413         struct irq_info *info;
1414
1415         list_for_each_entry(info, &xen_irq_list_head, list) {
1416                 if (info->type != IRQT_PIRQ)
1417                         continue;
1418
1419                 pirq = info->u.pirq.pirq;
1420                 gsi = info->u.pirq.gsi;
1421                 irq = info->irq;
1422
1423                 /* save/restore of PT devices doesn't work, so at this point the
1424                  * only devices present are GSI based emulated devices */
1425                 if (!gsi)
1426                         continue;
1427
1428                 map_irq.domid = DOMID_SELF;
1429                 map_irq.type = MAP_PIRQ_TYPE_GSI;
1430                 map_irq.index = gsi;
1431                 map_irq.pirq = pirq;
1432
1433                 rc = HYPERVISOR_physdev_op(PHYSDEVOP_map_pirq, &map_irq);
1434                 if (rc) {
1435                         pr_warn("xen map irq failed gsi=%d irq=%d pirq=%d rc=%d\n",
1436                                 gsi, irq, pirq, rc);
1437                         xen_free_irq(irq);
1438                         continue;
1439                 }
1440
1441                 printk(KERN_DEBUG "xen: --> irq=%d, pirq=%d\n", irq, map_irq.pirq);
1442
1443                 __startup_pirq(irq);
1444         }
1445 }
1446
1447 static void restore_cpu_virqs(unsigned int cpu)
1448 {
1449         struct evtchn_bind_virq bind_virq;
1450         int virq, irq, evtchn;
1451
1452         for (virq = 0; virq < NR_VIRQS; virq++) {
1453                 if ((irq = per_cpu(virq_to_irq, cpu)[virq]) == -1)
1454                         continue;
1455
1456                 BUG_ON(virq_from_irq(irq) != virq);
1457
1458                 /* Get a new binding from Xen. */
1459                 bind_virq.virq = virq;
1460                 bind_virq.vcpu = xen_vcpu_nr(cpu);
1461                 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_virq,
1462                                                 &bind_virq) != 0)
1463                         BUG();
1464                 evtchn = bind_virq.port;
1465
1466                 /* Record the new mapping. */
1467                 (void)xen_irq_info_virq_setup(cpu, irq, evtchn, virq);
1468                 bind_evtchn_to_cpu(evtchn, cpu);
1469         }
1470 }
1471
1472 static void restore_cpu_ipis(unsigned int cpu)
1473 {
1474         struct evtchn_bind_ipi bind_ipi;
1475         int ipi, irq, evtchn;
1476
1477         for (ipi = 0; ipi < XEN_NR_IPIS; ipi++) {
1478                 if ((irq = per_cpu(ipi_to_irq, cpu)[ipi]) == -1)
1479                         continue;
1480
1481                 BUG_ON(ipi_from_irq(irq) != ipi);
1482
1483                 /* Get a new binding from Xen. */
1484                 bind_ipi.vcpu = xen_vcpu_nr(cpu);
1485                 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_ipi,
1486                                                 &bind_ipi) != 0)
1487                         BUG();
1488                 evtchn = bind_ipi.port;
1489
1490                 /* Record the new mapping. */
1491                 (void)xen_irq_info_ipi_setup(cpu, irq, evtchn, ipi);
1492                 bind_evtchn_to_cpu(evtchn, cpu);
1493         }
1494 }
1495
1496 /* Clear an irq's pending state, in preparation for polling on it */
1497 void xen_clear_irq_pending(int irq)
1498 {
1499         int evtchn = evtchn_from_irq(irq);
1500
1501         if (VALID_EVTCHN(evtchn))
1502                 clear_evtchn(evtchn);
1503 }
1504 EXPORT_SYMBOL(xen_clear_irq_pending);
1505 void xen_set_irq_pending(int irq)
1506 {
1507         int evtchn = evtchn_from_irq(irq);
1508
1509         if (VALID_EVTCHN(evtchn))
1510                 set_evtchn(evtchn);
1511 }
1512
1513 bool xen_test_irq_pending(int irq)
1514 {
1515         int evtchn = evtchn_from_irq(irq);
1516         bool ret = false;
1517
1518         if (VALID_EVTCHN(evtchn))
1519                 ret = test_evtchn(evtchn);
1520
1521         return ret;
1522 }
1523
1524 /* Poll waiting for an irq to become pending with timeout.  In the usual case,
1525  * the irq will be disabled so it won't deliver an interrupt. */
1526 void xen_poll_irq_timeout(int irq, u64 timeout)
1527 {
1528         evtchn_port_t evtchn = evtchn_from_irq(irq);
1529
1530         if (VALID_EVTCHN(evtchn)) {
1531                 struct sched_poll poll;
1532
1533                 poll.nr_ports = 1;
1534                 poll.timeout = timeout;
1535                 set_xen_guest_handle(poll.ports, &evtchn);
1536
1537                 if (HYPERVISOR_sched_op(SCHEDOP_poll, &poll) != 0)
1538                         BUG();
1539         }
1540 }
1541 EXPORT_SYMBOL(xen_poll_irq_timeout);
1542 /* Poll waiting for an irq to become pending.  In the usual case, the
1543  * irq will be disabled so it won't deliver an interrupt. */
1544 void xen_poll_irq(int irq)
1545 {
1546         xen_poll_irq_timeout(irq, 0 /* no timeout */);
1547 }
1548
1549 /* Check whether the IRQ line is shared with other guests. */
1550 int xen_test_irq_shared(int irq)
1551 {
1552         struct irq_info *info = info_for_irq(irq);
1553         struct physdev_irq_status_query irq_status;
1554
1555         if (WARN_ON(!info))
1556                 return -ENOENT;
1557
1558         irq_status.irq = info->u.pirq.pirq;
1559
1560         if (HYPERVISOR_physdev_op(PHYSDEVOP_irq_status_query, &irq_status))
1561                 return 0;
1562         return !(irq_status.flags & XENIRQSTAT_shared);
1563 }
1564 EXPORT_SYMBOL_GPL(xen_test_irq_shared);
1565
1566 void xen_irq_resume(void)
1567 {
1568         unsigned int cpu;
1569         struct irq_info *info;
1570
1571         /* New event-channel space is not 'live' yet. */
1572         xen_evtchn_mask_all();
1573         xen_evtchn_resume();
1574
1575         /* No IRQ <-> event-channel mappings. */
1576         list_for_each_entry(info, &xen_irq_list_head, list)
1577                 info->evtchn = 0; /* zap event-channel binding */
1578
1579         clear_evtchn_to_irq_all();
1580
1581         for_each_possible_cpu(cpu) {
1582                 restore_cpu_virqs(cpu);
1583                 restore_cpu_ipis(cpu);
1584         }
1585
1586         restore_pirqs();
1587 }
1588
1589 static struct irq_chip xen_dynamic_chip __read_mostly = {
1590         .name                   = "xen-dyn",
1591
1592         .irq_disable            = disable_dynirq,
1593         .irq_mask               = disable_dynirq,
1594         .irq_unmask             = enable_dynirq,
1595
1596         .irq_ack                = ack_dynirq,
1597         .irq_mask_ack           = mask_ack_dynirq,
1598
1599         .irq_set_affinity       = set_affinity_irq,
1600         .irq_retrigger          = retrigger_dynirq,
1601 };
1602
1603 static struct irq_chip xen_pirq_chip __read_mostly = {
1604         .name                   = "xen-pirq",
1605
1606         .irq_startup            = startup_pirq,
1607         .irq_shutdown           = shutdown_pirq,
1608         .irq_enable             = enable_pirq,
1609         .irq_disable            = disable_pirq,
1610
1611         .irq_mask               = disable_dynirq,
1612         .irq_unmask             = enable_dynirq,
1613
1614         .irq_ack                = eoi_pirq,
1615         .irq_eoi                = eoi_pirq,
1616         .irq_mask_ack           = mask_ack_pirq,
1617
1618         .irq_set_affinity       = set_affinity_irq,
1619
1620         .irq_retrigger          = retrigger_dynirq,
1621 };
1622
1623 static struct irq_chip xen_percpu_chip __read_mostly = {
1624         .name                   = "xen-percpu",
1625
1626         .irq_disable            = disable_dynirq,
1627         .irq_mask               = disable_dynirq,
1628         .irq_unmask             = enable_dynirq,
1629
1630         .irq_ack                = ack_dynirq,
1631 };
1632
1633 int xen_set_callback_via(uint64_t via)
1634 {
1635         struct xen_hvm_param a;
1636         a.domid = DOMID_SELF;
1637         a.index = HVM_PARAM_CALLBACK_IRQ;
1638         a.value = via;
1639         return HYPERVISOR_hvm_op(HVMOP_set_param, &a);
1640 }
1641 EXPORT_SYMBOL_GPL(xen_set_callback_via);
1642
1643 #ifdef CONFIG_XEN_PVHVM
1644 /* Vector callbacks are better than PCI interrupts to receive event
1645  * channel notifications because we can receive vector callbacks on any
1646  * vcpu and we don't need PCI support or APIC interactions. */
1647 void xen_callback_vector(void)
1648 {
1649         int rc;
1650         uint64_t callback_via;
1651
1652         if (xen_have_vector_callback) {
1653                 callback_via = HVM_CALLBACK_VECTOR(HYPERVISOR_CALLBACK_VECTOR);
1654                 rc = xen_set_callback_via(callback_via);
1655                 if (rc) {
1656                         pr_err("Request for Xen HVM callback vector failed\n");
1657                         xen_have_vector_callback = 0;
1658                         return;
1659                 }
1660                 pr_info("Xen HVM callback vector for event delivery is enabled\n");
1661                 /* in the restore case the vector has already been allocated */
1662                 if (!test_bit(HYPERVISOR_CALLBACK_VECTOR, used_vectors))
1663                         alloc_intr_gate(HYPERVISOR_CALLBACK_VECTOR,
1664                                         xen_hvm_callback_vector);
1665         }
1666 }
1667 #else
1668 void xen_callback_vector(void) {}
1669 #endif
1670
1671 #undef MODULE_PARAM_PREFIX
1672 #define MODULE_PARAM_PREFIX "xen."
1673
1674 static bool fifo_events = true;
1675 module_param(fifo_events, bool, 0);
1676
1677 void __init xen_init_IRQ(void)
1678 {
1679         int ret = -EINVAL;
1680
1681         if (fifo_events)
1682                 ret = xen_evtchn_fifo_init();
1683         if (ret < 0)
1684                 xen_evtchn_2l_init();
1685
1686         evtchn_to_irq = kcalloc(EVTCHN_ROW(xen_evtchn_max_channels()),
1687                                 sizeof(*evtchn_to_irq), GFP_KERNEL);
1688         BUG_ON(!evtchn_to_irq);
1689
1690         /* No event channels are 'live' right now. */
1691         xen_evtchn_mask_all();
1692
1693         pirq_needs_eoi = pirq_needs_eoi_flag;
1694
1695 #ifdef CONFIG_X86
1696         if (xen_pv_domain()) {
1697                 irq_ctx_init(smp_processor_id());
1698                 if (xen_initial_domain())
1699                         pci_xen_initial_domain();
1700         }
1701         if (xen_feature(XENFEAT_hvm_callback_vector))
1702                 xen_callback_vector();
1703
1704         if (xen_hvm_domain()) {
1705                 native_init_IRQ();
1706                 /* pci_xen_hvm_init must be called after native_init_IRQ so that
1707                  * __acpi_register_gsi can point at the right function */
1708                 pci_xen_hvm_init();
1709         } else {
1710                 int rc;
1711                 struct physdev_pirq_eoi_gmfn eoi_gmfn;
1712
1713                 pirq_eoi_map = (void *)__get_free_page(GFP_KERNEL|__GFP_ZERO);
1714                 eoi_gmfn.gmfn = virt_to_gfn(pirq_eoi_map);
1715                 rc = HYPERVISOR_physdev_op(PHYSDEVOP_pirq_eoi_gmfn_v2, &eoi_gmfn);
1716                 if (rc != 0) {
1717                         free_page((unsigned long) pirq_eoi_map);
1718                         pirq_eoi_map = NULL;
1719                 } else
1720                         pirq_needs_eoi = pirq_check_eoi_map;
1721         }
1722 #endif
1723 }