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