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