drivers/xen/events.c

   1 /*
   2  * Xen event channels
   3  *
   4  * Xen models interrupts with abstract event channels.  Because each
   5  * domain gets 1024 event channels, but NR_IRQ is not that large, we
   6  * must dynamically map irqs<->event channels.  The event channels
   7  * interface with the rest of the kernel by defining a xen interrupt
   8  * chip.  When an event is recieved, it is mapped to an irq and sent
   9  * through the normal interrupt processing path.
  10  *
  11  * There are four kinds of events which can be mapped to an event
  12  * channel:
  13  *
  14  * 1. Inter-domain notifications.  This includes all the virtual
  15  *    device events, since they're driven by front-ends in another domain
  16  *    (typically dom0).
  17  * 2. VIRQs, typically used for timers.  These are per-cpu events.
  18  * 3. IPIs.
  19  * 4. PIRQs - Hardware interrupts.
  20  *
  21  * Jeremy Fitzhardinge <jeremy@xensource.com>, XenSource Inc, 2007
  22  */
  23
  24 #include <linux/linkage.h>
  25 #include <linux/interrupt.h>
  26 #include <linux/irq.h>
  27 #include <linux/module.h>
  28 #include <linux/string.h>
  29 #include <linux/bootmem.h>
  30 #include <linux/slab.h>
  31 #include <linux/irqnr.h>
  32 #include <linux/pci.h>
  33
  34 #include <asm/desc.h>
  35 #include <asm/ptrace.h>
  36 #include <asm/irq.h>
  37 #include <asm/idle.h>
  38 #include <asm/io_apic.h>
  39 #include <asm/sync_bitops.h>
  40 #include <asm/xen/pci.h>
  41 #include <asm/xen/hypercall.h>
  42 #include <asm/xen/hypervisor.h>
  43
  44 #include <xen/xen.h>
  45 #include <xen/hvm.h>
  46 #include <xen/xen-ops.h>
  47 #include <xen/events.h>
  48 #include <xen/interface/xen.h>
  49 #include <xen/interface/event_channel.h>
  50 #include <xen/interface/hvm/hvm_op.h>
  51 #include <xen/interface/hvm/params.h>
  52
  53 /*
  54  * This lock protects updates to the following mapping and reference-count
  55  * arrays. The lock does not need to be acquired to read the mapping tables.
  56  */
  57 static DEFINE_SPINLOCK(irq_mapping_update_lock);
  58
  59 /* IRQ <-> VIRQ mapping. */
  60 static DEFINE_PER_CPU(int [NR_VIRQS], virq_to_irq) = {[0 ... NR_VIRQS-1] = -1};
  61
  62 /* IRQ <-> IPI mapping */
  63 static DEFINE_PER_CPU(int [XEN_NR_IPIS], ipi_to_irq) = {[0 ... XEN_NR_IPIS-1] = -1};
  64
  65 /* Interrupt types. */
  66 enum xen_irq_type {
  67         IRQT_UNBOUND = 0,
  68         IRQT_PIRQ,
  69         IRQT_VIRQ,
  70         IRQT_IPI,
  71         IRQT_EVTCHN
  72 };
  73
  74 /*
  75  * Packed IRQ information:
  76  * type - enum xen_irq_type
  77  * event channel - irq->event channel mapping
  78  * cpu - cpu this event channel is bound to
  79  * index - type-specific information:
  80  *    PIRQ - vector, with MSB being "needs EIO", or physical IRQ of the HVM
  81  *           guest, or GSI (real passthrough IRQ) of the device.
  82  *    VIRQ - virq number
  83  *    IPI - IPI vector
  84  *    EVTCHN -
  85  */
  86 struct irq_info
  87 {
  88         enum xen_irq_type type; /* type */
  89         unsigned short evtchn;  /* event channel */
  90         unsigned short cpu;     /* cpu bound */
  91
  92         union {
  93                 unsigned short virq;
  94                 enum ipi_vector ipi;
  95                 struct {
  96                         unsigned short pirq;
  97                         unsigned short gsi;
  98                         unsigned char vector;
  99                         unsigned char flags;
 100                 } pirq;
 101         } u;
 102 };
 103 #define PIRQ_NEEDS_EOI  (1 << 0)
 104 #define PIRQ_SHAREABLE  (1 << 1)
 105
 106 static struct irq_info *irq_info;
 107 static int *pirq_to_irq;
 108 static int nr_pirqs;
 109
 110 static int *evtchn_to_irq;
 111 struct cpu_evtchn_s {
 112         unsigned long bits[NR_EVENT_CHANNELS/BITS_PER_LONG];
 113 };
 114
 115 static __initdata struct cpu_evtchn_s init_evtchn_mask = {
 116         .bits[0 ... (NR_EVENT_CHANNELS/BITS_PER_LONG)-1] = ~0ul,
 117 };
 118 static struct cpu_evtchn_s *cpu_evtchn_mask_p = &init_evtchn_mask;
 119
 120 static inline unsigned long *cpu_evtchn_mask(int cpu)
 121 {
 122         return cpu_evtchn_mask_p[cpu].bits;
 123 }
 124
 125 /* Xen will never allocate port zero for any purpose. */
 126 #define VALID_EVTCHN(chn)       ((chn) != 0)
 127
 128 static struct irq_chip xen_dynamic_chip;
 129 static struct irq_chip xen_percpu_chip;
 130 static struct irq_chip xen_pirq_chip;
 131
 132 /* Constructor for packed IRQ information. */
 133 static struct irq_info mk_unbound_info(void)
 134 {
 135         return (struct irq_info) { .type = IRQT_UNBOUND };
 136 }
 137
 138 static struct irq_info mk_evtchn_info(unsigned short evtchn)
 139 {
 140         return (struct irq_info) { .type = IRQT_EVTCHN, .evtchn = evtchn,
 141                         .cpu = 0 };
 142 }
 143
 144 static struct irq_info mk_ipi_info(unsigned short evtchn, enum ipi_vector ipi)
 145 {
 146         return (struct irq_info) { .type = IRQT_IPI, .evtchn = evtchn,
 147                         .cpu = 0, .u.ipi = ipi };
 148 }
 149
 150 static struct irq_info mk_virq_info(unsigned short evtchn, unsigned short virq)
 151 {
 152         return (struct irq_info) { .type = IRQT_VIRQ, .evtchn = evtchn,
 153                         .cpu = 0, .u.virq = virq };
 154 }
 155
 156 static struct irq_info mk_pirq_info(unsigned short evtchn, unsigned short pirq,
 157                                     unsigned short gsi, unsigned short vector)
 158 {
 159         return (struct irq_info) { .type = IRQT_PIRQ, .evtchn = evtchn,
 160                         .cpu = 0,
 161                         .u.pirq = { .pirq = pirq, .gsi = gsi, .vector = vector } };
 162 }
 163
 164 /*
 165  * Accessors for packed IRQ information.
 166  */
 167 static struct irq_info *info_for_irq(unsigned irq)
 168 {
 169         return &irq_info[irq];
 170 }
 171
 172 static unsigned int evtchn_from_irq(unsigned irq)
 173 {
 174         return info_for_irq(irq)->evtchn;
 175 }
 176
 177 unsigned irq_from_evtchn(unsigned int evtchn)
 178 {
 179         return evtchn_to_irq[evtchn];
 180 }
 181 EXPORT_SYMBOL_GPL(irq_from_evtchn);
 182
 183 static enum ipi_vector ipi_from_irq(unsigned irq)
 184 {
 185         struct irq_info *info = info_for_irq(irq);
 186
 187         BUG_ON(info == NULL);
 188         BUG_ON(info->type != IRQT_IPI);
 189
 190         return info->u.ipi;
 191 }
 192
 193 static unsigned virq_from_irq(unsigned irq)
 194 {
 195         struct irq_info *info = info_for_irq(irq);
 196
 197         BUG_ON(info == NULL);
 198         BUG_ON(info->type != IRQT_VIRQ);
 199
 200         return info->u.virq;
 201 }
 202
 203 static unsigned pirq_from_irq(unsigned irq)
 204 {
 205         struct irq_info *info = info_for_irq(irq);
 206
 207         BUG_ON(info == NULL);
 208         BUG_ON(info->type != IRQT_PIRQ);
 209
 210         return info->u.pirq.pirq;
 211 }
 212
 213 static unsigned gsi_from_irq(unsigned irq)
 214 {
 215         struct irq_info *info = info_for_irq(irq);
 216
 217         BUG_ON(info == NULL);
 218         BUG_ON(info->type != IRQT_PIRQ);
 219
 220         return info->u.pirq.gsi;
 221 }
 222
 223 static unsigned vector_from_irq(unsigned irq)
 224 {
 225         struct irq_info *info = info_for_irq(irq);
 226
 227         BUG_ON(info == NULL);
 228         BUG_ON(info->type != IRQT_PIRQ);
 229
 230         return info->u.pirq.vector;
 231 }
 232
 233 static enum xen_irq_type type_from_irq(unsigned irq)
 234 {
 235         return info_for_irq(irq)->type;
 236 }
 237
 238 static unsigned cpu_from_irq(unsigned irq)
 239 {
 240         return info_for_irq(irq)->cpu;
 241 }
 242
 243 static unsigned int cpu_from_evtchn(unsigned int evtchn)
 244 {
 245         int irq = evtchn_to_irq[evtchn];
 246         unsigned ret = 0;
 247
 248         if (irq != -1)
 249                 ret = cpu_from_irq(irq);
 250
 251         return ret;
 252 }
 253
 254 static bool pirq_needs_eoi(unsigned irq)
 255 {
 256         struct irq_info *info = info_for_irq(irq);
 257
 258         BUG_ON(info->type != IRQT_PIRQ);
 259
 260         return info->u.pirq.flags & PIRQ_NEEDS_EOI;
 261 }
 262
 263 static inline unsigned long active_evtchns(unsigned int cpu,
 264                                            struct shared_info *sh,
 265                                            unsigned int idx)
 266 {
 267         return (sh->evtchn_pending[idx] &
 268                 cpu_evtchn_mask(cpu)[idx] &
 269                 ~sh->evtchn_mask[idx]);
 270 }
 271
 272 static void bind_evtchn_to_cpu(unsigned int chn, unsigned int cpu)
 273 {
 274         int irq = evtchn_to_irq[chn];
 275
 276         BUG_ON(irq == -1);
 277 #ifdef CONFIG_SMP
 278         cpumask_copy(irq_to_desc(irq)->affinity, cpumask_of(cpu));
 279 #endif
 280
 281         __clear_bit(chn, cpu_evtchn_mask(cpu_from_irq(irq)));
 282         __set_bit(chn, cpu_evtchn_mask(cpu));
 283
 284         irq_info[irq].cpu = cpu;
 285 }
 286
 287 static void init_evtchn_cpu_bindings(void)
 288 {
 289 #ifdef CONFIG_SMP
 290         struct irq_desc *desc;
 291         int i;
 292
 293         /* By default all event channels notify CPU#0. */
 294         for_each_irq_desc(i, desc) {
 295                 cpumask_copy(desc->affinity, cpumask_of(0));
 296         }
 297 #endif
 298
 299         memset(cpu_evtchn_mask(0), ~0, sizeof(cpu_evtchn_mask(0)));
 300 }
 301
 302 static inline void clear_evtchn(int port)
 303 {
 304         struct shared_info *s = HYPERVISOR_shared_info;
 305         sync_clear_bit(port, &s->evtchn_pending[0]);
 306 }
 307
 308 static inline void set_evtchn(int port)
 309 {
 310         struct shared_info *s = HYPERVISOR_shared_info;
 311         sync_set_bit(port, &s->evtchn_pending[0]);
 312 }
 313
 314 static inline int test_evtchn(int port)
 315 {
 316         struct shared_info *s = HYPERVISOR_shared_info;
 317         return sync_test_bit(port, &s->evtchn_pending[0]);
 318 }
 319
 320
 321 /**
 322  * notify_remote_via_irq - send event to remote end of event channel via irq
 323  * @irq: irq of event channel to send event to
 324  *
 325  * Unlike notify_remote_via_evtchn(), this is safe to use across
 326  * save/restore. Notifications on a broken connection are silently
 327  * dropped.
 328  */
 329 void notify_remote_via_irq(int irq)
 330 {
 331         int evtchn = evtchn_from_irq(irq);
 332
 333         if (VALID_EVTCHN(evtchn))
 334                 notify_remote_via_evtchn(evtchn);
 335 }
 336 EXPORT_SYMBOL_GPL(notify_remote_via_irq);
 337
 338 static void mask_evtchn(int port)
 339 {
 340         struct shared_info *s = HYPERVISOR_shared_info;
 341         sync_set_bit(port, &s->evtchn_mask[0]);
 342 }
 343
 344 static void unmask_evtchn(int port)
 345 {
 346         struct shared_info *s = HYPERVISOR_shared_info;
 347         unsigned int cpu = get_cpu();
 348
 349         BUG_ON(!irqs_disabled());
 350
 351         /* Slow path (hypercall) if this is a non-local port. */
 352         if (unlikely(cpu != cpu_from_evtchn(port))) {
 353                 struct evtchn_unmask unmask = { .port = port };
 354                 (void)HYPERVISOR_event_channel_op(EVTCHNOP_unmask, &unmask);
 355         } else {
 356                 struct vcpu_info *vcpu_info = __get_cpu_var(xen_vcpu);
 357
 358                 sync_clear_bit(port, &s->evtchn_mask[0]);
 359
 360                 /*
 361                  * The following is basically the equivalent of
 362                  * 'hw_resend_irq'. Just like a real IO-APIC we 'lose
 363                  * the interrupt edge' if the channel is masked.
 364                  */
 365                 if (sync_test_bit(port, &s->evtchn_pending[0]) &&
 366                     !sync_test_and_set_bit(port / BITS_PER_LONG,
 367                                            &vcpu_info->evtchn_pending_sel))
 368                         vcpu_info->evtchn_upcall_pending = 1;
 369         }
 370
 371         put_cpu();
 372 }
 373
 374 static int get_nr_hw_irqs(void)
 375 {
 376         int ret = 1;
 377
 378 #ifdef CONFIG_X86_IO_APIC
 379         ret = get_nr_irqs_gsi();
 380 #endif
 381
 382         return ret;
 383 }
 384
 385 /* callers of this function should make sure that PHYSDEVOP_get_nr_pirqs
 386  * succeeded otherwise nr_pirqs won't hold the right value */
 387 static int find_unbound_pirq(void)
 388 {
 389         int i;
 390         for (i = nr_pirqs-1; i >= 0; i--) {
 391                 if (pirq_to_irq[i] < 0)
 392                         return i;
 393         }
 394         return -1;
 395 }
 396
 397 static int find_unbound_irq(void)
 398 {
 399         struct irq_data *data;
 400         int irq, res;
 401         int start = get_nr_hw_irqs();
 402
 403         if (start == nr_irqs)
 404                 goto no_irqs;
 405
 406         /* nr_irqs is a magic value. Must not use it.*/
 407         for (irq = nr_irqs-1; irq > start; irq--) {
 408                 data = irq_get_irq_data(irq);
 409                 /* only 0->15 have init'd desc; handle irq > 16 */
 410                 if (!data)
 411                         break;
 412                 if (data->chip == &no_irq_chip)
 413                         break;
 414                 if (data->chip != &xen_dynamic_chip)
 415                         continue;
 416                 if (irq_info[irq].type == IRQT_UNBOUND)
 417                         return irq;
 418         }
 419
 420         if (irq == start)
 421                 goto no_irqs;
 422
 423         res = irq_alloc_desc_at(irq, 0);
 424
 425         if (WARN_ON(res != irq))
 426                 return -1;
 427
 428         return irq;
 429
 430 no_irqs:
 431         panic("No available IRQ to bind to: increase nr_irqs!\n");
 432 }
 433
 434 static bool identity_mapped_irq(unsigned irq)
 435 {
 436         /* identity map all the hardware irqs */
 437         return irq < get_nr_hw_irqs();
 438 }
 439
 440 static void pirq_unmask_notify(int irq)
 441 {
 442         struct physdev_eoi eoi = { .irq = pirq_from_irq(irq) };
 443
 444         if (unlikely(pirq_needs_eoi(irq))) {
 445                 int rc = HYPERVISOR_physdev_op(PHYSDEVOP_eoi, &eoi);
 446                 WARN_ON(rc);
 447         }
 448 }
 449
 450 static void pirq_query_unmask(int irq)
 451 {
 452         struct physdev_irq_status_query irq_status;
 453         struct irq_info *info = info_for_irq(irq);
 454
 455         BUG_ON(info->type != IRQT_PIRQ);
 456
 457         irq_status.irq = pirq_from_irq(irq);
 458         if (HYPERVISOR_physdev_op(PHYSDEVOP_irq_status_query, &irq_status))
 459                 irq_status.flags = 0;
 460
 461         info->u.pirq.flags &= ~PIRQ_NEEDS_EOI;
 462         if (irq_status.flags & XENIRQSTAT_needs_eoi)
 463                 info->u.pirq.flags |= PIRQ_NEEDS_EOI;
 464 }
 465
 466 static bool probing_irq(int irq)
 467 {
 468         struct irq_desc *desc = irq_to_desc(irq);
 469
 470         return desc && desc->action == NULL;
 471 }
 472
 473 static unsigned int startup_pirq(unsigned int irq)
 474 {
 475         struct evtchn_bind_pirq bind_pirq;
 476         struct irq_info *info = info_for_irq(irq);
 477         int evtchn = evtchn_from_irq(irq);
 478         int rc;
 479
 480         BUG_ON(info->type != IRQT_PIRQ);
 481
 482         if (VALID_EVTCHN(evtchn))
 483                 goto out;
 484
 485         bind_pirq.pirq = pirq_from_irq(irq);
 486         /* NB. We are happy to share unless we are probing. */
 487         bind_pirq.flags = info->u.pirq.flags & PIRQ_SHAREABLE ?
 488                                         BIND_PIRQ__WILL_SHARE : 0;
 489         rc = HYPERVISOR_event_channel_op(EVTCHNOP_bind_pirq, &bind_pirq);
 490         if (rc != 0) {
 491                 if (!probing_irq(irq))
 492                         printk(KERN_INFO "Failed to obtain physical IRQ %d\n",
 493                                irq);
 494                 return 0;
 495         }
 496         evtchn = bind_pirq.port;
 497
 498         pirq_query_unmask(irq);
 499
 500         evtchn_to_irq[evtchn] = irq;
 501         bind_evtchn_to_cpu(evtchn, 0);
 502         info->evtchn = evtchn;
 503
 504 out:
 505         unmask_evtchn(evtchn);
 506         pirq_unmask_notify(irq);
 507
 508         return 0;
 509 }
 510
 511 static void shutdown_pirq(unsigned int irq)
 512 {
 513         struct evtchn_close close;
 514         struct irq_info *info = info_for_irq(irq);
 515         int evtchn = evtchn_from_irq(irq);
 516
 517         BUG_ON(info->type != IRQT_PIRQ);
 518
 519         if (!VALID_EVTCHN(evtchn))
 520                 return;
 521
 522         mask_evtchn(evtchn);
 523
 524         close.port = evtchn;
 525         if (HYPERVISOR_event_channel_op(EVTCHNOP_close, &close) != 0)
 526                 BUG();
 527
 528         bind_evtchn_to_cpu(evtchn, 0);
 529         evtchn_to_irq[evtchn] = -1;
 530         info->evtchn = 0;
 531 }
 532
 533 static void enable_pirq(unsigned int irq)
 534 {
 535         startup_pirq(irq);
 536 }
 537
 538 static void disable_pirq(unsigned int irq)
 539 {
 540 }
 541
 542 static void ack_pirq(unsigned int irq)
 543 {
 544         int evtchn = evtchn_from_irq(irq);
 545
 546         move_native_irq(irq);
 547
 548         if (VALID_EVTCHN(evtchn)) {
 549                 mask_evtchn(evtchn);
 550                 clear_evtchn(evtchn);
 551         }
 552 }
 553
 554 static void end_pirq(unsigned int irq)
 555 {
 556         int evtchn = evtchn_from_irq(irq);
 557         struct irq_desc *desc = irq_to_desc(irq);
 558
 559         if (WARN_ON(!desc))
 560                 return;
 561
 562         if ((desc->status & (IRQ_DISABLED|IRQ_PENDING)) ==
 563             (IRQ_DISABLED|IRQ_PENDING)) {
 564                 shutdown_pirq(irq);
 565         } else if (VALID_EVTCHN(evtchn)) {
 566                 unmask_evtchn(evtchn);
 567                 pirq_unmask_notify(irq);
 568         }
 569 }
 570
 571 static int find_irq_by_gsi(unsigned gsi)
 572 {
 573         int irq;
 574
 575         for (irq = 0; irq < nr_irqs; irq++) {
 576                 struct irq_info *info = info_for_irq(irq);
 577
 578                 if (info == NULL || info->type != IRQT_PIRQ)
 579                         continue;
 580
 581                 if (gsi_from_irq(irq) == gsi)
 582                         return irq;
 583         }
 584
 585         return -1;
 586 }
 587
 588 int xen_allocate_pirq(unsigned gsi, int shareable, char *name)
 589 {
 590         return xen_map_pirq_gsi(gsi, gsi, shareable, name);
 591 }
 592
 593 /* xen_map_pirq_gsi might allocate irqs from the top down, as a
 594  * consequence don't assume that the irq number returned has a low value
 595  * or can be used as a pirq number unless you know otherwise.
 596  *
 597  * One notable exception is when xen_map_pirq_gsi is called passing an
 598  * hardware gsi as argument, in that case the irq number returned
 599  * matches the gsi number passed as second argument.
 600  *
 601  * Note: We don't assign an event channel until the irq actually started
 602  * up.  Return an existing irq if we've already got one for the gsi.
 603  */
 604 int xen_map_pirq_gsi(unsigned pirq, unsigned gsi, int shareable, char *name)
 605 {
 606         int irq = 0;
 607         struct physdev_irq irq_op;
 608
 609         spin_lock(&irq_mapping_update_lock);
 610
 611         if ((pirq > nr_pirqs) || (gsi > nr_irqs)) {
 612                 printk(KERN_WARNING "xen_map_pirq_gsi: %s %s is incorrect!\n",
 613                         pirq > nr_pirqs ? "nr_pirqs" :"",
 614                         gsi > nr_irqs ? "nr_irqs" : "");
 615                 goto out;
 616         }
 617
 618         irq = find_irq_by_gsi(gsi);
 619         if (irq != -1) {
 620                 printk(KERN_INFO "xen_map_pirq_gsi: returning irq %d for gsi %u\n",
 621                        irq, gsi);
 622                 goto out;       /* XXX need refcount? */
 623         }
 624
 625         /* If we are a PV guest, we don't have GSIs (no ACPI passed). Therefore
 626          * we are using the !xen_initial_domain() to drop in the function.*/
 627         if (identity_mapped_irq(gsi) || (!xen_initial_domain() &&
 628                                 xen_pv_domain())) {
 629                 irq = gsi;
 630                 irq_alloc_desc_at(irq, 0);
 631         } else
 632                 irq = find_unbound_irq();
 633
 634         set_irq_chip_and_handler_name(irq, &xen_pirq_chip,
 635                                       handle_level_irq, name);
 636
 637         irq_op.irq = irq;
 638         irq_op.vector = 0;
 639
 640         /* Only the privileged domain can do this. For non-priv, the pcifront
 641          * driver provides a PCI bus that does the call to do exactly
 642          * this in the priv domain. */
 643         if (xen_initial_domain() &&
 644             HYPERVISOR_physdev_op(PHYSDEVOP_alloc_irq_vector, &irq_op)) {
 645                 irq_free_desc(irq);
 646                 irq = -ENOSPC;
 647                 goto out;
 648         }
 649
 650         irq_info[irq] = mk_pirq_info(0, pirq, gsi, irq_op.vector);
 651         irq_info[irq].u.pirq.flags |= shareable ? PIRQ_SHAREABLE : 0;
 652         pirq_to_irq[pirq] = irq;
 653
 654 out:
 655         spin_unlock(&irq_mapping_update_lock);
 656
 657         return irq;
 658 }
 659
 660 #ifdef CONFIG_PCI_MSI
 661 #include <linux/msi.h>
 662 #include "../pci/msi.h"
 663
 664 void xen_allocate_pirq_msi(char *name, int *irq, int *pirq)
 665 {
 666         spin_lock(&irq_mapping_update_lock);
 667
 668         *irq = find_unbound_irq();
 669         if (*irq == -1)
 670                 goto out;
 671
 672         *pirq = find_unbound_pirq();
 673         if (*pirq == -1)
 674                 goto out;
 675
 676         set_irq_chip_and_handler_name(*irq, &xen_pirq_chip,
 677                                       handle_level_irq, name);
 678
 679         irq_info[*irq] = mk_pirq_info(0, *pirq, 0, 0);
 680         pirq_to_irq[*pirq] = *irq;
 681
 682 out:
 683         spin_unlock(&irq_mapping_update_lock);
 684 }
 685
 686 int xen_create_msi_irq(struct pci_dev *dev, struct msi_desc *msidesc, int type)
 687 {
 688         int irq = -1;
 689         struct physdev_map_pirq map_irq;
 690         int rc;
 691         int pos;
 692         u32 table_offset, bir;
 693
 694         memset(&map_irq, 0, sizeof(map_irq));
 695         map_irq.domid = DOMID_SELF;
 696         map_irq.type = MAP_PIRQ_TYPE_MSI;
 697         map_irq.index = -1;
 698         map_irq.pirq = -1;
 699         map_irq.bus = dev->bus->number;
 700         map_irq.devfn = dev->devfn;
 701
 702         if (type == PCI_CAP_ID_MSIX) {
 703                 pos = pci_find_capability(dev, PCI_CAP_ID_MSIX);
 704
 705                 pci_read_config_dword(dev, msix_table_offset_reg(pos),
 706                                         &table_offset);
 707                 bir = (u8)(table_offset & PCI_MSIX_FLAGS_BIRMASK);
 708
 709                 map_irq.table_base = pci_resource_start(dev, bir);
 710                 map_irq.entry_nr = msidesc->msi_attrib.entry_nr;
 711         }
 712
 713         spin_lock(&irq_mapping_update_lock);
 714
 715         irq = find_unbound_irq();
 716
 717         if (irq == -1)
 718                 goto out;
 719
 720         rc = HYPERVISOR_physdev_op(PHYSDEVOP_map_pirq, &map_irq);
 721         if (rc) {
 722                 printk(KERN_WARNING "xen map irq failed %d\n", rc);
 723
 724                 irq_free_desc(irq);
 725
 726                 irq = -1;
 727                 goto out;
 728         }
 729         irq_info[irq] = mk_pirq_info(0, map_irq.pirq, 0, map_irq.index);
 730
 731         set_irq_chip_and_handler_name(irq, &xen_pirq_chip,
 732                         handle_level_irq,
 733                         (type == PCI_CAP_ID_MSIX) ? "msi-x":"msi");
 734
 735 out:
 736         spin_unlock(&irq_mapping_update_lock);
 737         return irq;
 738 }
 739 #endif
 740
 741 int xen_destroy_irq(int irq)
 742 {
 743         struct irq_desc *desc;
 744         struct physdev_unmap_pirq unmap_irq;
 745         struct irq_info *info = info_for_irq(irq);
 746         int rc = -ENOENT;
 747
 748         spin_lock(&irq_mapping_update_lock);
 749
 750         desc = irq_to_desc(irq);
 751         if (!desc)
 752                 goto out;
 753
 754         if (xen_initial_domain()) {
 755                 unmap_irq.pirq = info->u.pirq.gsi;
 756                 unmap_irq.domid = DOMID_SELF;
 757                 rc = HYPERVISOR_physdev_op(PHYSDEVOP_unmap_pirq, &unmap_irq);
 758                 if (rc) {
 759                         printk(KERN_WARNING "unmap irq failed %d\n", rc);
 760                         goto out;
 761                 }
 762         }
 763         irq_info[irq] = mk_unbound_info();
 764
 765         irq_free_desc(irq);
 766
 767 out:
 768         spin_unlock(&irq_mapping_update_lock);
 769         return rc;
 770 }
 771
 772 int xen_vector_from_irq(unsigned irq)
 773 {
 774         return vector_from_irq(irq);
 775 }
 776
 777 int xen_gsi_from_irq(unsigned irq)
 778 {
 779         return gsi_from_irq(irq);
 780 }
 781
 782 int bind_evtchn_to_irq(unsigned int evtchn)
 783 {
 784         int irq;
 785
 786         spin_lock(&irq_mapping_update_lock);
 787
 788         irq = evtchn_to_irq[evtchn];
 789
 790         if (irq == -1) {
 791                 irq = find_unbound_irq();
 792
 793                 set_irq_chip_and_handler_name(irq, &xen_dynamic_chip,
 794                                               handle_edge_irq, "event");
 795
 796                 evtchn_to_irq[evtchn] = irq;
 797                 irq_info[irq] = mk_evtchn_info(evtchn);
 798         }
 799
 800         spin_unlock(&irq_mapping_update_lock);
 801
 802         return irq;
 803 }
 804 EXPORT_SYMBOL_GPL(bind_evtchn_to_irq);
 805
 806 static int bind_ipi_to_irq(unsigned int ipi, unsigned int cpu)
 807 {
 808         struct evtchn_bind_ipi bind_ipi;
 809         int evtchn, irq;
 810
 811         spin_lock(&irq_mapping_update_lock);
 812
 813         irq = per_cpu(ipi_to_irq, cpu)[ipi];
 814
 815         if (irq == -1) {
 816                 irq = find_unbound_irq();
 817                 if (irq < 0)
 818                         goto out;
 819
 820                 set_irq_chip_and_handler_name(irq, &xen_percpu_chip,
 821                                               handle_percpu_irq, "ipi");
 822
 823                 bind_ipi.vcpu = cpu;
 824                 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_ipi,
 825                                                 &bind_ipi) != 0)
 826                         BUG();
 827                 evtchn = bind_ipi.port;
 828
 829                 evtchn_to_irq[evtchn] = irq;
 830                 irq_info[irq] = mk_ipi_info(evtchn, ipi);
 831                 per_cpu(ipi_to_irq, cpu)[ipi] = irq;
 832
 833                 bind_evtchn_to_cpu(evtchn, cpu);
 834         }
 835
 836  out:
 837         spin_unlock(&irq_mapping_update_lock);
 838         return irq;
 839 }
 840
 841
 842 static int bind_virq_to_irq(unsigned int virq, unsigned int cpu)
 843 {
 844         struct evtchn_bind_virq bind_virq;
 845         int evtchn, irq;
 846
 847         spin_lock(&irq_mapping_update_lock);
 848
 849         irq = per_cpu(virq_to_irq, cpu)[virq];
 850
 851         if (irq == -1) {
 852                 bind_virq.virq = virq;
 853                 bind_virq.vcpu = cpu;
 854                 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_virq,
 855                                                 &bind_virq) != 0)
 856                         BUG();
 857                 evtchn = bind_virq.port;
 858
 859                 irq = find_unbound_irq();
 860
 861                 set_irq_chip_and_handler_name(irq, &xen_percpu_chip,
 862                                               handle_percpu_irq, "virq");
 863
 864                 evtchn_to_irq[evtchn] = irq;
 865                 irq_info[irq] = mk_virq_info(evtchn, virq);
 866
 867                 per_cpu(virq_to_irq, cpu)[virq] = irq;
 868
 869                 bind_evtchn_to_cpu(evtchn, cpu);
 870         }
 871
 872         spin_unlock(&irq_mapping_update_lock);
 873
 874         return irq;
 875 }
 876
 877 static void unbind_from_irq(unsigned int irq)
 878 {
 879         struct evtchn_close close;
 880         int evtchn = evtchn_from_irq(irq);
 881
 882         spin_lock(&irq_mapping_update_lock);
 883
 884         if (VALID_EVTCHN(evtchn)) {
 885                 close.port = evtchn;
 886                 if (HYPERVISOR_event_channel_op(EVTCHNOP_close, &close) != 0)
 887                         BUG();
 888
 889                 switch (type_from_irq(irq)) {
 890                 case IRQT_VIRQ:
 891                         per_cpu(virq_to_irq, cpu_from_evtchn(evtchn))
 892                                 [virq_from_irq(irq)] = -1;
 893                         break;
 894                 case IRQT_IPI:
 895                         per_cpu(ipi_to_irq, cpu_from_evtchn(evtchn))
 896                                 [ipi_from_irq(irq)] = -1;
 897                         break;
 898                 default:
 899                         break;
 900                 }
 901
 902                 /* Closed ports are implicitly re-bound to VCPU0. */
 903                 bind_evtchn_to_cpu(evtchn, 0);
 904
 905                 evtchn_to_irq[evtchn] = -1;
 906         }
 907
 908         if (irq_info[irq].type != IRQT_UNBOUND) {
 909                 irq_info[irq] = mk_unbound_info();
 910
 911                 irq_free_desc(irq);
 912         }
 913
 914         spin_unlock(&irq_mapping_update_lock);
 915 }
 916
 917 int bind_evtchn_to_irqhandler(unsigned int evtchn,
 918                               irq_handler_t handler,
 919                               unsigned long irqflags,
 920                               const char *devname, void *dev_id)
 921 {
 922         unsigned int irq;
 923         int retval;
 924
 925         irq = bind_evtchn_to_irq(evtchn);
 926         retval = request_irq(irq, handler, irqflags, devname, dev_id);
 927         if (retval != 0) {
 928                 unbind_from_irq(irq);
 929                 return retval;
 930         }
 931
 932         return irq;
 933 }
 934 EXPORT_SYMBOL_GPL(bind_evtchn_to_irqhandler);
 935
 936 int bind_virq_to_irqhandler(unsigned int virq, unsigned int cpu,
 937                             irq_handler_t handler,
 938                             unsigned long irqflags, const char *devname, void *dev_id)
 939 {
 940         unsigned int irq;
 941         int retval;
 942
 943         irq = bind_virq_to_irq(virq, cpu);
 944         retval = request_irq(irq, handler, irqflags, devname, dev_id);
 945         if (retval != 0) {
 946                 unbind_from_irq(irq);
 947                 return retval;
 948         }
 949
 950         return irq;
 951 }
 952 EXPORT_SYMBOL_GPL(bind_virq_to_irqhandler);
 953
 954 int bind_ipi_to_irqhandler(enum ipi_vector ipi,
 955                            unsigned int cpu,
 956                            irq_handler_t handler,
 957                            unsigned long irqflags,
 958                            const char *devname,
 959                            void *dev_id)
 960 {
 961         int irq, retval;
 962
 963         irq = bind_ipi_to_irq(ipi, cpu);
 964         if (irq < 0)
 965                 return irq;
 966
 967         irqflags |= IRQF_NO_SUSPEND;
 968         retval = request_irq(irq, handler, irqflags, devname, dev_id);
 969         if (retval != 0) {
 970                 unbind_from_irq(irq);
 971                 return retval;
 972         }
 973
 974         return irq;
 975 }
 976
 977 void unbind_from_irqhandler(unsigned int irq, void *dev_id)
 978 {
 979         free_irq(irq, dev_id);
 980         unbind_from_irq(irq);
 981 }
 982 EXPORT_SYMBOL_GPL(unbind_from_irqhandler);
 983
 984 void xen_send_IPI_one(unsigned int cpu, enum ipi_vector vector)
 985 {
 986         int irq = per_cpu(ipi_to_irq, cpu)[vector];
 987         BUG_ON(irq < 0);
 988         notify_remote_via_irq(irq);
 989 }
 990
 991 irqreturn_t xen_debug_interrupt(int irq, void *dev_id)
 992 {
 993         struct shared_info *sh = HYPERVISOR_shared_info;
 994         int cpu = smp_processor_id();
 995         int i;
 996         unsigned long flags;
 997         static DEFINE_SPINLOCK(debug_lock);
 998
 999         spin_lock_irqsave(&debug_lock, flags);
1000
1001         printk("vcpu %d\n  ", cpu);
1002
1003         for_each_online_cpu(i) {
1004                 struct vcpu_info *v = per_cpu(xen_vcpu, i);
1005                 printk("%d: masked=%d pending=%d event_sel %08lx\n  ", i,
1006                         (get_irq_regs() && i == cpu) ? xen_irqs_disabled(get_irq_regs()) : v->evtchn_upcall_mask,
1007                         v->evtchn_upcall_pending,
1008                         v->evtchn_pending_sel);
1009         }
1010         printk("pending:\n   ");
1011         for(i = ARRAY_SIZE(sh->evtchn_pending)-1; i >= 0; i--)
1012                 printk("%08lx%s", sh->evtchn_pending[i],
1013                         i % 8 == 0 ? "\n   " : " ");
1014         printk("\nmasks:\n   ");
1015         for(i = ARRAY_SIZE(sh->evtchn_mask)-1; i >= 0; i--)
1016                 printk("%08lx%s", sh->evtchn_mask[i],
1017                         i % 8 == 0 ? "\n   " : " ");
1018
1019         printk("\nunmasked:\n   ");
1020         for(i = ARRAY_SIZE(sh->evtchn_mask)-1; i >= 0; i--)
1021                 printk("%08lx%s", sh->evtchn_pending[i] & ~sh->evtchn_mask[i],
1022                         i % 8 == 0 ? "\n   " : " ");
1023
1024         printk("\npending list:\n");
1025         for(i = 0; i < NR_EVENT_CHANNELS; i++) {
1026                 if (sync_test_bit(i, sh->evtchn_pending)) {
1027                         printk("  %d: event %d -> irq %d\n",
1028                                cpu_from_evtchn(i), i,
1029                                evtchn_to_irq[i]);
1030                 }
1031         }
1032
1033         spin_unlock_irqrestore(&debug_lock, flags);
1034
1035         return IRQ_HANDLED;
1036 }
1037
1038 static DEFINE_PER_CPU(unsigned, xed_nesting_count);
1039
1040 /*
1041  * Search the CPUs pending events bitmasks.  For each one found, map
1042  * the event number to an irq, and feed it into do_IRQ() for
1043  * handling.
1044  *
1045  * Xen uses a two-level bitmap to speed searching.  The first level is
1046  * a bitset of words which contain pending event bits.  The second
1047  * level is a bitset of pending events themselves.
1048  */
1049 static void __xen_evtchn_do_upcall(void)
1050 {
1051         int cpu = get_cpu();
1052         struct shared_info *s = HYPERVISOR_shared_info;
1053         struct vcpu_info *vcpu_info = __get_cpu_var(xen_vcpu);
1054         unsigned count;
1055
1056         do {
1057                 unsigned long pending_words;
1058
1059                 vcpu_info->evtchn_upcall_pending = 0;
1060
1061                 if (__get_cpu_var(xed_nesting_count)++)
1062                         goto out;
1063
1064 #ifndef CONFIG_X86 /* No need for a barrier -- XCHG is a barrier on x86. */
1065                 /* Clear master flag /before/ clearing selector flag. */
1066                 wmb();
1067 #endif
1068                 pending_words = xchg(&vcpu_info->evtchn_pending_sel, 0);
1069                 while (pending_words != 0) {
1070                         unsigned long pending_bits;
1071                         int word_idx = __ffs(pending_words);
1072                         pending_words &= ~(1UL << word_idx);
1073
1074                         while ((pending_bits = active_evtchns(cpu, s, word_idx)) != 0) {
1075                                 int bit_idx = __ffs(pending_bits);
1076                                 int port = (word_idx * BITS_PER_LONG) + bit_idx;
1077                                 int irq = evtchn_to_irq[port];
1078                                 struct irq_desc *desc;
1079
1080                                 if (irq != -1) {
1081                                         desc = irq_to_desc(irq);
1082                                         if (desc)
1083                                                 generic_handle_irq_desc(irq, desc);
1084                                 }
1085                         }
1086                 }
1087
1088                 BUG_ON(!irqs_disabled());
1089
1090                 count = __get_cpu_var(xed_nesting_count);
1091                 __get_cpu_var(xed_nesting_count) = 0;
1092         } while (count != 1 || vcpu_info->evtchn_upcall_pending);
1093
1094 out:
1095
1096         put_cpu();
1097 }
1098
1099 void xen_evtchn_do_upcall(struct pt_regs *regs)
1100 {
1101         struct pt_regs *old_regs = set_irq_regs(regs);
1102
1103         exit_idle();
1104         irq_enter();
1105
1106         __xen_evtchn_do_upcall();
1107
1108         irq_exit();
1109         set_irq_regs(old_regs);
1110 }
1111
1112 void xen_hvm_evtchn_do_upcall(void)
1113 {
1114         __xen_evtchn_do_upcall();
1115 }
1116 EXPORT_SYMBOL_GPL(xen_hvm_evtchn_do_upcall);
1117
1118 /* Rebind a new event channel to an existing irq. */
1119 void rebind_evtchn_irq(int evtchn, int irq)
1120 {
1121         struct irq_info *info = info_for_irq(irq);
1122
1123         /* Make sure the irq is masked, since the new event channel
1124            will also be masked. */
1125         disable_irq(irq);
1126
1127         spin_lock(&irq_mapping_update_lock);
1128
1129         /* After resume the irq<->evtchn mappings are all cleared out */
1130         BUG_ON(evtchn_to_irq[evtchn] != -1);
1131         /* Expect irq to have been bound before,
1132            so there should be a proper type */
1133         BUG_ON(info->type == IRQT_UNBOUND);
1134
1135         evtchn_to_irq[evtchn] = irq;
1136         irq_info[irq] = mk_evtchn_info(evtchn);
1137
1138         spin_unlock(&irq_mapping_update_lock);
1139
1140         /* new event channels are always bound to cpu 0 */
1141         irq_set_affinity(irq, cpumask_of(0));
1142
1143         /* Unmask the event channel. */
1144         enable_irq(irq);
1145 }
1146
1147 /* Rebind an evtchn so that it gets delivered to a specific cpu */
1148 static int rebind_irq_to_cpu(unsigned irq, unsigned tcpu)
1149 {
1150         struct evtchn_bind_vcpu bind_vcpu;
1151         int evtchn = evtchn_from_irq(irq);
1152
1153         /* events delivered via platform PCI interrupts are always
1154          * routed to vcpu 0 */
1155         if (!VALID_EVTCHN(evtchn) ||
1156                 (xen_hvm_domain() && !xen_have_vector_callback))
1157                 return -1;
1158
1159         /* Send future instances of this interrupt to other vcpu. */
1160         bind_vcpu.port = evtchn;
1161         bind_vcpu.vcpu = tcpu;
1162
1163         /*
1164          * If this fails, it usually just indicates that we're dealing with a
1165          * virq or IPI channel, which don't actually need to be rebound. Ignore
1166          * it, but don't do the xenlinux-level rebind in that case.
1167          */
1168         if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_vcpu, &bind_vcpu) >= 0)
1169                 bind_evtchn_to_cpu(evtchn, tcpu);
1170
1171         return 0;
1172 }
1173
1174 static int set_affinity_irq(unsigned irq, const struct cpumask *dest)
1175 {
1176         unsigned tcpu = cpumask_first(dest);
1177
1178         return rebind_irq_to_cpu(irq, tcpu);
1179 }
1180
1181 int resend_irq_on_evtchn(unsigned int irq)
1182 {
1183         int masked, evtchn = evtchn_from_irq(irq);
1184         struct shared_info *s = HYPERVISOR_shared_info;
1185
1186         if (!VALID_EVTCHN(evtchn))
1187                 return 1;
1188
1189         masked = sync_test_and_set_bit(evtchn, s->evtchn_mask);
1190         sync_set_bit(evtchn, s->evtchn_pending);
1191         if (!masked)
1192                 unmask_evtchn(evtchn);
1193
1194         return 1;
1195 }
1196
1197 static void enable_dynirq(unsigned int irq)
1198 {
1199         int evtchn = evtchn_from_irq(irq);
1200
1201         if (VALID_EVTCHN(evtchn))
1202                 unmask_evtchn(evtchn);
1203 }
1204
1205 static void disable_dynirq(unsigned int irq)
1206 {
1207         int evtchn = evtchn_from_irq(irq);
1208
1209         if (VALID_EVTCHN(evtchn))
1210                 mask_evtchn(evtchn);
1211 }
1212
1213 static void ack_dynirq(unsigned int irq)
1214 {
1215         int evtchn = evtchn_from_irq(irq);
1216
1217         move_native_irq(irq);
1218
1219         if (VALID_EVTCHN(evtchn))
1220                 clear_evtchn(evtchn);
1221 }
1222
1223 static int retrigger_dynirq(unsigned int irq)
1224 {
1225         int evtchn = evtchn_from_irq(irq);
1226         struct shared_info *sh = HYPERVISOR_shared_info;
1227         int ret = 0;
1228
1229         if (VALID_EVTCHN(evtchn)) {
1230                 int masked;
1231
1232                 masked = sync_test_and_set_bit(evtchn, sh->evtchn_mask);
1233                 sync_set_bit(evtchn, sh->evtchn_pending);
1234                 if (!masked)
1235                         unmask_evtchn(evtchn);
1236                 ret = 1;
1237         }
1238
1239         return ret;
1240 }
1241
1242 static void restore_cpu_virqs(unsigned int cpu)
1243 {
1244         struct evtchn_bind_virq bind_virq;
1245         int virq, irq, evtchn;
1246
1247         for (virq = 0; virq < NR_VIRQS; virq++) {
1248                 if ((irq = per_cpu(virq_to_irq, cpu)[virq]) == -1)
1249                         continue;
1250
1251                 BUG_ON(virq_from_irq(irq) != virq);
1252
1253                 /* Get a new binding from Xen. */
1254                 bind_virq.virq = virq;
1255                 bind_virq.vcpu = cpu;
1256                 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_virq,
1257                                                 &bind_virq) != 0)
1258                         BUG();
1259                 evtchn = bind_virq.port;
1260
1261                 /* Record the new mapping. */
1262                 evtchn_to_irq[evtchn] = irq;
1263                 irq_info[irq] = mk_virq_info(evtchn, virq);
1264                 bind_evtchn_to_cpu(evtchn, cpu);
1265
1266                 /* Ready for use. */
1267                 unmask_evtchn(evtchn);
1268         }
1269 }
1270
1271 static void restore_cpu_ipis(unsigned int cpu)
1272 {
1273         struct evtchn_bind_ipi bind_ipi;
1274         int ipi, irq, evtchn;
1275
1276         for (ipi = 0; ipi < XEN_NR_IPIS; ipi++) {
1277                 if ((irq = per_cpu(ipi_to_irq, cpu)[ipi]) == -1)
1278                         continue;
1279
1280                 BUG_ON(ipi_from_irq(irq) != ipi);
1281
1282                 /* Get a new binding from Xen. */
1283                 bind_ipi.vcpu = cpu;
1284                 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_ipi,
1285                                                 &bind_ipi) != 0)
1286                         BUG();
1287                 evtchn = bind_ipi.port;
1288
1289                 /* Record the new mapping. */
1290                 evtchn_to_irq[evtchn] = irq;
1291                 irq_info[irq] = mk_ipi_info(evtchn, ipi);
1292                 bind_evtchn_to_cpu(evtchn, cpu);
1293
1294                 /* Ready for use. */
1295                 unmask_evtchn(evtchn);
1296
1297         }
1298 }
1299
1300 /* Clear an irq's pending state, in preparation for polling on it */
1301 void xen_clear_irq_pending(int irq)
1302 {
1303         int evtchn = evtchn_from_irq(irq);
1304
1305         if (VALID_EVTCHN(evtchn))
1306                 clear_evtchn(evtchn);
1307 }
1308 EXPORT_SYMBOL(xen_clear_irq_pending);
1309 void xen_set_irq_pending(int irq)
1310 {
1311         int evtchn = evtchn_from_irq(irq);
1312
1313         if (VALID_EVTCHN(evtchn))
1314                 set_evtchn(evtchn);
1315 }
1316
1317 bool xen_test_irq_pending(int irq)
1318 {
1319         int evtchn = evtchn_from_irq(irq);
1320         bool ret = false;
1321
1322         if (VALID_EVTCHN(evtchn))
1323                 ret = test_evtchn(evtchn);
1324
1325         return ret;
1326 }
1327
1328 /* Poll waiting for an irq to become pending with timeout.  In the usual case,
1329  * the irq will be disabled so it won't deliver an interrupt. */
1330 void xen_poll_irq_timeout(int irq, u64 timeout)
1331 {
1332         evtchn_port_t evtchn = evtchn_from_irq(irq);
1333
1334         if (VALID_EVTCHN(evtchn)) {
1335                 struct sched_poll poll;
1336
1337                 poll.nr_ports = 1;
1338                 poll.timeout = timeout;
1339                 set_xen_guest_handle(poll.ports, &evtchn);
1340
1341                 if (HYPERVISOR_sched_op(SCHEDOP_poll, &poll) != 0)
1342                         BUG();
1343         }
1344 }
1345 EXPORT_SYMBOL(xen_poll_irq_timeout);
1346 /* Poll waiting for an irq to become pending.  In the usual case, the
1347  * irq will be disabled so it won't deliver an interrupt. */
1348 void xen_poll_irq(int irq)
1349 {
1350         xen_poll_irq_timeout(irq, 0 /* no timeout */);
1351 }
1352
1353 void xen_irq_resume(void)
1354 {
1355         unsigned int cpu, irq, evtchn;
1356
1357         init_evtchn_cpu_bindings();
1358
1359         /* New event-channel space is not 'live' yet. */
1360         for (evtchn = 0; evtchn < NR_EVENT_CHANNELS; evtchn++)
1361                 mask_evtchn(evtchn);
1362
1363         /* No IRQ <-> event-channel mappings. */
1364         for (irq = 0; irq < nr_irqs; irq++)
1365                 irq_info[irq].evtchn = 0; /* zap event-channel binding */
1366
1367         for (evtchn = 0; evtchn < NR_EVENT_CHANNELS; evtchn++)
1368                 evtchn_to_irq[evtchn] = -1;
1369
1370         for_each_possible_cpu(cpu) {
1371                 restore_cpu_virqs(cpu);
1372                 restore_cpu_ipis(cpu);
1373         }
1374 }
1375
1376 static struct irq_chip xen_dynamic_chip __read_mostly = {
1377         .name           = "xen-dyn",
1378
1379         .disable        = disable_dynirq,
1380         .mask           = disable_dynirq,
1381         .unmask         = enable_dynirq,
1382
1383         .ack            = ack_dynirq,
1384         .set_affinity   = set_affinity_irq,
1385         .retrigger      = retrigger_dynirq,
1386 };
1387
1388 static struct irq_chip xen_pirq_chip __read_mostly = {
1389         .name           = "xen-pirq",
1390
1391         .startup        = startup_pirq,
1392         .shutdown       = shutdown_pirq,
1393
1394         .enable         = enable_pirq,
1395         .unmask         = enable_pirq,
1396
1397         .disable        = disable_pirq,
1398         .mask           = disable_pirq,
1399
1400         .ack            = ack_pirq,
1401         .end            = end_pirq,
1402
1403         .set_affinity   = set_affinity_irq,
1404
1405         .retrigger      = retrigger_dynirq,
1406 };
1407
1408 static struct irq_chip xen_percpu_chip __read_mostly = {
1409         .name           = "xen-percpu",
1410
1411         .disable        = disable_dynirq,
1412         .mask           = disable_dynirq,
1413         .unmask         = enable_dynirq,
1414
1415         .ack            = ack_dynirq,
1416 };
1417
1418 int xen_set_callback_via(uint64_t via)
1419 {
1420         struct xen_hvm_param a;
1421         a.domid = DOMID_SELF;
1422         a.index = HVM_PARAM_CALLBACK_IRQ;
1423         a.value = via;
1424         return HYPERVISOR_hvm_op(HVMOP_set_param, &a);
1425 }
1426 EXPORT_SYMBOL_GPL(xen_set_callback_via);
1427
1428 #ifdef CONFIG_XEN_PVHVM
1429 /* Vector callbacks are better than PCI interrupts to receive event
1430  * channel notifications because we can receive vector callbacks on any
1431  * vcpu and we don't need PCI support or APIC interactions. */
1432 void xen_callback_vector(void)
1433 {
1434         int rc;
1435         uint64_t callback_via;
1436         if (xen_have_vector_callback) {
1437                 callback_via = HVM_CALLBACK_VECTOR(XEN_HVM_EVTCHN_CALLBACK);
1438                 rc = xen_set_callback_via(callback_via);
1439                 if (rc) {
1440                         printk(KERN_ERR "Request for Xen HVM callback vector"
1441                                         " failed.\n");
1442                         xen_have_vector_callback = 0;
1443                         return;
1444                 }
1445                 printk(KERN_INFO "Xen HVM callback vector for event delivery is "
1446                                 "enabled\n");
1447                 /* in the restore case the vector has already been allocated */
1448                 if (!test_bit(XEN_HVM_EVTCHN_CALLBACK, used_vectors))
1449                         alloc_intr_gate(XEN_HVM_EVTCHN_CALLBACK, xen_hvm_callback_vector);
1450         }
1451 }
1452 #else
1453 void xen_callback_vector(void) {}
1454 #endif
1455
1456 void __init xen_init_IRQ(void)
1457 {
1458         int i, rc;
1459         struct physdev_nr_pirqs op_nr_pirqs;
1460
1461         cpu_evtchn_mask_p = kcalloc(nr_cpu_ids, sizeof(struct cpu_evtchn_s),
1462                                     GFP_KERNEL);
1463         irq_info = kcalloc(nr_irqs, sizeof(*irq_info), GFP_KERNEL);
1464
1465         rc = HYPERVISOR_physdev_op(PHYSDEVOP_get_nr_pirqs, &op_nr_pirqs);
1466         if (rc < 0) {
1467                 nr_pirqs = nr_irqs;
1468                 if (rc != -ENOSYS)
1469                         printk(KERN_WARNING "PHYSDEVOP_get_nr_pirqs returned rc=%d\n", rc);
1470         } else {
1471                 if (xen_pv_domain() && !xen_initial_domain())
1472                         nr_pirqs = max((int)op_nr_pirqs.nr_pirqs, nr_irqs);
1473                 else
1474                         nr_pirqs = op_nr_pirqs.nr_pirqs;
1475         }
1476         pirq_to_irq = kcalloc(nr_pirqs, sizeof(*pirq_to_irq), GFP_KERNEL);
1477         for (i = 0; i < nr_pirqs; i++)
1478                 pirq_to_irq[i] = -1;
1479
1480         evtchn_to_irq = kcalloc(NR_EVENT_CHANNELS, sizeof(*evtchn_to_irq),
1481                                     GFP_KERNEL);
1482         for (i = 0; i < NR_EVENT_CHANNELS; i++)
1483                 evtchn_to_irq[i] = -1;
1484
1485         init_evtchn_cpu_bindings();
1486
1487         /* No event channels are 'live' right now. */
1488         for (i = 0; i < NR_EVENT_CHANNELS; i++)
1489                 mask_evtchn(i);
1490
1491         if (xen_hvm_domain()) {
1492                 xen_callback_vector();
1493                 native_init_IRQ();
1494                 /* pci_xen_hvm_init must be called after native_init_IRQ so that
1495                  * __acpi_register_gsi can point at the right function */
1496                 pci_xen_hvm_init();
1497         } else {
1498                 irq_ctx_init(smp_processor_id());
1499                 if (xen_initial_domain())
1500                         xen_setup_pirqs();
1501         }
1502 }