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