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
 109 static int *evtchn_to_irq;
 110 struct cpu_evtchn_s {
 111         unsigned long bits[NR_EVENT_CHANNELS/BITS_PER_LONG];
 112 };
 113
 114 static __initdata struct cpu_evtchn_s init_evtchn_mask = {
 115         .bits[0 ... (NR_EVENT_CHANNELS/BITS_PER_LONG)-1] = ~0ul,
 116 };
 117 static struct cpu_evtchn_s *cpu_evtchn_mask_p = &init_evtchn_mask;
 118
 119 static inline unsigned long *cpu_evtchn_mask(int cpu)
 120 {
 121         return cpu_evtchn_mask_p[cpu].bits;
 122 }
 123
 124 /* Xen will never allocate port zero for any purpose. */
 125 #define VALID_EVTCHN(chn)       ((chn) != 0)
 126
 127 static struct irq_chip xen_dynamic_chip;
 128 static struct irq_chip xen_percpu_chip;
 129 static struct irq_chip xen_pirq_chip;
 130
 131 /* Constructor for packed IRQ information. */
 132 static struct irq_info mk_unbound_info(void)
 133 {
 134         return (struct irq_info) { .type = IRQT_UNBOUND };
 135 }
 136
 137 static struct irq_info mk_evtchn_info(unsigned short evtchn)
 138 {
 139         return (struct irq_info) { .type = IRQT_EVTCHN, .evtchn = evtchn,
 140                         .cpu = 0 };
 141 }
 142
 143 static struct irq_info mk_ipi_info(unsigned short evtchn, enum ipi_vector ipi)
 144 {
 145         return (struct irq_info) { .type = IRQT_IPI, .evtchn = evtchn,
 146                         .cpu = 0, .u.ipi = ipi };
 147 }
 148
 149 static struct irq_info mk_virq_info(unsigned short evtchn, unsigned short virq)
 150 {
 151         return (struct irq_info) { .type = IRQT_VIRQ, .evtchn = evtchn,
 152                         .cpu = 0, .u.virq = virq };
 153 }
 154
 155 static struct irq_info mk_pirq_info(unsigned short evtchn, unsigned short pirq,
 156                                     unsigned short gsi, unsigned short vector)
 157 {
 158         return (struct irq_info) { .type = IRQT_PIRQ, .evtchn = evtchn,
 159                         .cpu = 0,
 160                         .u.pirq = { .pirq = pirq, .gsi = gsi, .vector = vector } };
 161 }
 162
 163 /*
 164  * Accessors for packed IRQ information.
 165  */
 166 static struct irq_info *info_for_irq(unsigned irq)
 167 {
 168         return &irq_info[irq];
 169 }
 170
 171 static unsigned int evtchn_from_irq(unsigned irq)
 172 {
 173         if (unlikely(WARN(irq < 0 || irq >= nr_irqs, "Invalid irq %d!\n", irq)))
 174                 return 0;
 175
 176         return info_for_irq(irq)->evtchn;
 177 }
 178
 179 unsigned irq_from_evtchn(unsigned int evtchn)
 180 {
 181         return evtchn_to_irq[evtchn];
 182 }
 183 EXPORT_SYMBOL_GPL(irq_from_evtchn);
 184
 185 static enum ipi_vector ipi_from_irq(unsigned irq)
 186 {
 187         struct irq_info *info = info_for_irq(irq);
 188
 189         BUG_ON(info == NULL);
 190         BUG_ON(info->type != IRQT_IPI);
 191
 192         return info->u.ipi;
 193 }
 194
 195 static unsigned virq_from_irq(unsigned irq)
 196 {
 197         struct irq_info *info = info_for_irq(irq);
 198
 199         BUG_ON(info == NULL);
 200         BUG_ON(info->type != IRQT_VIRQ);
 201
 202         return info->u.virq;
 203 }
 204
 205 static unsigned pirq_from_irq(unsigned irq)
 206 {
 207         struct irq_info *info = info_for_irq(irq);
 208
 209         BUG_ON(info == NULL);
 210         BUG_ON(info->type != IRQT_PIRQ);
 211
 212         return info->u.pirq.pirq;
 213 }
 214
 215 static unsigned gsi_from_irq(unsigned irq)
 216 {
 217         struct irq_info *info = info_for_irq(irq);
 218
 219         BUG_ON(info == NULL);
 220         BUG_ON(info->type != IRQT_PIRQ);
 221
 222         return info->u.pirq.gsi;
 223 }
 224
 225 static unsigned vector_from_irq(unsigned irq)
 226 {
 227         struct irq_info *info = info_for_irq(irq);
 228
 229         BUG_ON(info == NULL);
 230         BUG_ON(info->type != IRQT_PIRQ);
 231
 232         return info->u.pirq.vector;
 233 }
 234
 235 static enum xen_irq_type type_from_irq(unsigned irq)
 236 {
 237         return info_for_irq(irq)->type;
 238 }
 239
 240 static unsigned cpu_from_irq(unsigned irq)
 241 {
 242         return info_for_irq(irq)->cpu;
 243 }
 244
 245 static unsigned int cpu_from_evtchn(unsigned int evtchn)
 246 {
 247         int irq = evtchn_to_irq[evtchn];
 248         unsigned ret = 0;
 249
 250         if (irq != -1)
 251                 ret = cpu_from_irq(irq);
 252
 253         return ret;
 254 }
 255
 256 static bool pirq_needs_eoi(unsigned irq)
 257 {
 258         struct irq_info *info = info_for_irq(irq);
 259
 260         BUG_ON(info->type != IRQT_PIRQ);
 261
 262         return info->u.pirq.flags & PIRQ_NEEDS_EOI;
 263 }
 264
 265 static inline unsigned long active_evtchns(unsigned int cpu,
 266                                            struct shared_info *sh,
 267                                            unsigned int idx)
 268 {
 269         return (sh->evtchn_pending[idx] &
 270                 cpu_evtchn_mask(cpu)[idx] &
 271                 ~sh->evtchn_mask[idx]);
 272 }
 273
 274 static void bind_evtchn_to_cpu(unsigned int chn, unsigned int cpu)
 275 {
 276         int irq = evtchn_to_irq[chn];
 277
 278         BUG_ON(irq == -1);
 279 #ifdef CONFIG_SMP
 280         cpumask_copy(irq_to_desc(irq)->irq_data.affinity, cpumask_of(cpu));
 281 #endif
 282
 283         clear_bit(chn, cpu_evtchn_mask(cpu_from_irq(irq)));
 284         set_bit(chn, cpu_evtchn_mask(cpu));
 285
 286         irq_info[irq].cpu = cpu;
 287 }
 288
 289 static void init_evtchn_cpu_bindings(void)
 290 {
 291         int i;
 292 #ifdef CONFIG_SMP
 293         struct irq_desc *desc;
 294
 295         /* By default all event channels notify CPU#0. */
 296         for_each_irq_desc(i, desc) {
 297                 cpumask_copy(desc->irq_data.affinity, cpumask_of(0));
 298         }
 299 #endif
 300
 301         for_each_possible_cpu(i)
 302                 memset(cpu_evtchn_mask(i),
 303                        (i == 0) ? ~0 : 0, sizeof(struct cpu_evtchn_s));
 304
 305 }
 306
 307 static inline void clear_evtchn(int port)
 308 {
 309         struct shared_info *s = HYPERVISOR_shared_info;
 310         sync_clear_bit(port, &s->evtchn_pending[0]);
 311 }
 312
 313 static inline void set_evtchn(int port)
 314 {
 315         struct shared_info *s = HYPERVISOR_shared_info;
 316         sync_set_bit(port, &s->evtchn_pending[0]);
 317 }
 318
 319 static inline int test_evtchn(int port)
 320 {
 321         struct shared_info *s = HYPERVISOR_shared_info;
 322         return sync_test_bit(port, &s->evtchn_pending[0]);
 323 }
 324
 325
 326 /**
 327  * notify_remote_via_irq - send event to remote end of event channel via irq
 328  * @irq: irq of event channel to send event to
 329  *
 330  * Unlike notify_remote_via_evtchn(), this is safe to use across
 331  * save/restore. Notifications on a broken connection are silently
 332  * dropped.
 333  */
 334 void notify_remote_via_irq(int irq)
 335 {
 336         int evtchn = evtchn_from_irq(irq);
 337
 338         if (VALID_EVTCHN(evtchn))
 339                 notify_remote_via_evtchn(evtchn);
 340 }
 341 EXPORT_SYMBOL_GPL(notify_remote_via_irq);
 342
 343 static void mask_evtchn(int port)
 344 {
 345         struct shared_info *s = HYPERVISOR_shared_info;
 346         sync_set_bit(port, &s->evtchn_mask[0]);
 347 }
 348
 349 static void unmask_evtchn(int port)
 350 {
 351         struct shared_info *s = HYPERVISOR_shared_info;
 352         unsigned int cpu = get_cpu();
 353
 354         BUG_ON(!irqs_disabled());
 355
 356         /* Slow path (hypercall) if this is a non-local port. */
 357         if (unlikely(cpu != cpu_from_evtchn(port))) {
 358                 struct evtchn_unmask unmask = { .port = port };
 359                 (void)HYPERVISOR_event_channel_op(EVTCHNOP_unmask, &unmask);
 360         } else {
 361                 struct vcpu_info *vcpu_info = __this_cpu_read(xen_vcpu);
 362
 363                 sync_clear_bit(port, &s->evtchn_mask[0]);
 364
 365                 /*
 366                  * The following is basically the equivalent of
 367                  * 'hw_resend_irq'. Just like a real IO-APIC we 'lose
 368                  * the interrupt edge' if the channel is masked.
 369                  */
 370                 if (sync_test_bit(port, &s->evtchn_pending[0]) &&
 371                     !sync_test_and_set_bit(port / BITS_PER_LONG,
 372                                            &vcpu_info->evtchn_pending_sel))
 373                         vcpu_info->evtchn_upcall_pending = 1;
 374         }
 375
 376         put_cpu();
 377 }
 378
 379 static int xen_allocate_irq_dynamic(void)
 380 {
 381         int first = 0;
 382         int irq;
 383
 384 #ifdef CONFIG_X86_IO_APIC
 385         /*
 386          * For an HVM guest or domain 0 which see "real" (emulated or
 387          * actual repectively) GSIs we allocate dynamic IRQs
 388          * e.g. those corresponding to event channels or MSIs
 389          * etc. from the range above those "real" GSIs to avoid
 390          * collisions.
 391          */
 392         if (xen_initial_domain() || xen_hvm_domain())
 393                 first = get_nr_irqs_gsi();
 394 #endif
 395
 396 retry:
 397         irq = irq_alloc_desc_from(first, -1);
 398
 399         if (irq == -ENOMEM && first > NR_IRQS_LEGACY) {
 400                 printk(KERN_ERR "Out of dynamic IRQ space and eating into GSI space. You should increase nr_irqs\n");
 401                 first = max(NR_IRQS_LEGACY, first - NR_IRQS_LEGACY);
 402                 goto retry;
 403         }
 404
 405         if (irq < 0)
 406                 panic("No available IRQ to bind to: increase nr_irqs!\n");
 407
 408         return irq;
 409 }
 410
 411 static int xen_allocate_irq_gsi(unsigned gsi)
 412 {
 413         int irq;
 414
 415         /*
 416          * A PV guest has no concept of a GSI (since it has no ACPI
 417          * nor access to/knowledge of the physical APICs). Therefore
 418          * all IRQs are dynamically allocated from the entire IRQ
 419          * space.
 420          */
 421         if (xen_pv_domain() && !xen_initial_domain())
 422                 return xen_allocate_irq_dynamic();
 423
 424         /* Legacy IRQ descriptors are already allocated by the arch. */
 425         if (gsi < NR_IRQS_LEGACY)
 426                 return gsi;
 427
 428         irq = irq_alloc_desc_at(gsi, -1);
 429         if (irq < 0)
 430                 panic("Unable to allocate to IRQ%d (%d)\n", gsi, irq);
 431
 432         return irq;
 433 }
 434
 435 static void xen_free_irq(unsigned irq)
 436 {
 437         /* Legacy IRQ descriptors are managed by the arch. */
 438         if (irq < NR_IRQS_LEGACY)
 439                 return;
 440
 441         irq_free_desc(irq);
 442 }
 443
 444 static void pirq_unmask_notify(int irq)
 445 {
 446         struct physdev_eoi eoi = { .irq = pirq_from_irq(irq) };
 447
 448         if (unlikely(pirq_needs_eoi(irq))) {
 449                 int rc = HYPERVISOR_physdev_op(PHYSDEVOP_eoi, &eoi);
 450                 WARN_ON(rc);
 451         }
 452 }
 453
 454 static void pirq_query_unmask(int irq)
 455 {
 456         struct physdev_irq_status_query irq_status;
 457         struct irq_info *info = info_for_irq(irq);
 458
 459         BUG_ON(info->type != IRQT_PIRQ);
 460
 461         irq_status.irq = pirq_from_irq(irq);
 462         if (HYPERVISOR_physdev_op(PHYSDEVOP_irq_status_query, &irq_status))
 463                 irq_status.flags = 0;
 464
 465         info->u.pirq.flags &= ~PIRQ_NEEDS_EOI;
 466         if (irq_status.flags & XENIRQSTAT_needs_eoi)
 467                 info->u.pirq.flags |= PIRQ_NEEDS_EOI;
 468 }
 469
 470 static bool probing_irq(int irq)
 471 {
 472         struct irq_desc *desc = irq_to_desc(irq);
 473
 474         return desc && desc->action == NULL;
 475 }
 476
 477 static unsigned int __startup_pirq(unsigned int irq)
 478 {
 479         struct evtchn_bind_pirq bind_pirq;
 480         struct irq_info *info = info_for_irq(irq);
 481         int evtchn = evtchn_from_irq(irq);
 482         int rc;
 483
 484         BUG_ON(info->type != IRQT_PIRQ);
 485
 486         if (VALID_EVTCHN(evtchn))
 487                 goto out;
 488
 489         bind_pirq.pirq = pirq_from_irq(irq);
 490         /* NB. We are happy to share unless we are probing. */
 491         bind_pirq.flags = info->u.pirq.flags & PIRQ_SHAREABLE ?
 492                                         BIND_PIRQ__WILL_SHARE : 0;
 493         rc = HYPERVISOR_event_channel_op(EVTCHNOP_bind_pirq, &bind_pirq);
 494         if (rc != 0) {
 495                 if (!probing_irq(irq))
 496                         printk(KERN_INFO "Failed to obtain physical IRQ %d\n",
 497                                irq);
 498                 return 0;
 499         }
 500         evtchn = bind_pirq.port;
 501
 502         pirq_query_unmask(irq);
 503
 504         evtchn_to_irq[evtchn] = irq;
 505         bind_evtchn_to_cpu(evtchn, 0);
 506         info->evtchn = evtchn;
 507
 508 out:
 509         unmask_evtchn(evtchn);
 510         pirq_unmask_notify(irq);
 511
 512         return 0;
 513 }
 514
 515 static unsigned int startup_pirq(struct irq_data *data)
 516 {
 517         return __startup_pirq(data->irq);
 518 }
 519
 520 static void shutdown_pirq(struct irq_data *data)
 521 {
 522         struct evtchn_close close;
 523         unsigned int irq = data->irq;
 524         struct irq_info *info = info_for_irq(irq);
 525         int evtchn = evtchn_from_irq(irq);
 526
 527         BUG_ON(info->type != IRQT_PIRQ);
 528
 529         if (!VALID_EVTCHN(evtchn))
 530                 return;
 531
 532         mask_evtchn(evtchn);
 533
 534         close.port = evtchn;
 535         if (HYPERVISOR_event_channel_op(EVTCHNOP_close, &close) != 0)
 536                 BUG();
 537
 538         bind_evtchn_to_cpu(evtchn, 0);
 539         evtchn_to_irq[evtchn] = -1;
 540         info->evtchn = 0;
 541 }
 542
 543 static void enable_pirq(struct irq_data *data)
 544 {
 545         startup_pirq(data);
 546 }
 547
 548 static void disable_pirq(struct irq_data *data)
 549 {
 550 }
 551
 552 static void ack_pirq(struct irq_data *data)
 553 {
 554         int evtchn = evtchn_from_irq(data->irq);
 555
 556         move_native_irq(data->irq);
 557
 558         if (VALID_EVTCHN(evtchn)) {
 559                 mask_evtchn(evtchn);
 560                 clear_evtchn(evtchn);
 561         }
 562 }
 563
 564 static int find_irq_by_gsi(unsigned gsi)
 565 {
 566         int irq;
 567
 568         for (irq = 0; irq < nr_irqs; irq++) {
 569                 struct irq_info *info = info_for_irq(irq);
 570
 571                 if (info == NULL || info->type != IRQT_PIRQ)
 572                         continue;
 573
 574                 if (gsi_from_irq(irq) == gsi)
 575                         return irq;
 576         }
 577
 578         return -1;
 579 }
 580
 581 int xen_allocate_pirq(unsigned gsi, int shareable, char *name)
 582 {
 583         return xen_map_pirq_gsi(gsi, gsi, shareable, name);
 584 }
 585
 586 /* xen_map_pirq_gsi might allocate irqs from the top down, as a
 587  * consequence don't assume that the irq number returned has a low value
 588  * or can be used as a pirq number unless you know otherwise.
 589  *
 590  * One notable exception is when xen_map_pirq_gsi is called passing an
 591  * hardware gsi as argument, in that case the irq number returned
 592  * matches the gsi number passed as second argument.
 593  *
 594  * Note: We don't assign an event channel until the irq actually started
 595  * up.  Return an existing irq if we've already got one for the gsi.
 596  */
 597 int xen_map_pirq_gsi(unsigned pirq, unsigned gsi, int shareable, char *name)
 598 {
 599         int irq = 0;
 600         struct physdev_irq irq_op;
 601
 602         spin_lock(&irq_mapping_update_lock);
 603
 604         if ((pirq > nr_irqs) || (gsi > nr_irqs)) {
 605                 printk(KERN_WARNING "xen_map_pirq_gsi: %s %s is incorrect!\n",
 606                         pirq > nr_irqs ? "pirq" :"",
 607                         gsi > nr_irqs ? "gsi" : "");
 608                 goto out;
 609         }
 610
 611         irq = find_irq_by_gsi(gsi);
 612         if (irq != -1) {
 613                 printk(KERN_INFO "xen_map_pirq_gsi: returning irq %d for gsi %u\n",
 614                        irq, gsi);
 615                 goto out;       /* XXX need refcount? */
 616         }
 617
 618         irq = xen_allocate_irq_gsi(gsi);
 619
 620         set_irq_chip_and_handler_name(irq, &xen_pirq_chip,
 621                                       handle_level_irq, name);
 622
 623         irq_op.irq = irq;
 624         irq_op.vector = 0;
 625
 626         /* Only the privileged domain can do this. For non-priv, the pcifront
 627          * driver provides a PCI bus that does the call to do exactly
 628          * this in the priv domain. */
 629         if (xen_initial_domain() &&
 630             HYPERVISOR_physdev_op(PHYSDEVOP_alloc_irq_vector, &irq_op)) {
 631                 xen_free_irq(irq);
 632                 irq = -ENOSPC;
 633                 goto out;
 634         }
 635
 636         irq_info[irq] = mk_pirq_info(0, pirq, gsi, irq_op.vector);
 637         irq_info[irq].u.pirq.flags |= shareable ? PIRQ_SHAREABLE : 0;
 638         pirq_to_irq[pirq] = irq;
 639
 640 out:
 641         spin_unlock(&irq_mapping_update_lock);
 642
 643         return irq;
 644 }
 645
 646 #ifdef CONFIG_PCI_MSI
 647 #include <linux/msi.h>
 648 #include "../pci/msi.h"
 649
 650 int xen_allocate_pirq_msi(struct pci_dev *dev, struct msi_desc *msidesc)
 651 {
 652         int rc;
 653         struct physdev_get_free_pirq op_get_free_pirq;
 654
 655         op_get_free_pirq.type = MAP_PIRQ_TYPE_MSI;
 656         rc = HYPERVISOR_physdev_op(PHYSDEVOP_get_free_pirq, &op_get_free_pirq);
 657
 658         WARN_ONCE(rc == -ENOSYS,
 659                   "hypervisor does not support the PHYSDEVOP_get_free_pirq interface\n");
 660
 661         return rc ? -1 : op_get_free_pirq.pirq;
 662 }
 663
 664 int xen_bind_pirq_msi_to_irq(struct pci_dev *dev, struct msi_desc *msidesc,
 665                              int pirq, const char *name)
 666 {
 667         int irq, ret;
 668
 669         spin_lock(&irq_mapping_update_lock);
 670
 671         irq = xen_allocate_irq_dynamic();
 672         if (irq == -1)
 673                 goto out;
 674
 675         set_irq_chip_and_handler_name(irq, &xen_pirq_chip,
 676                                       handle_level_irq, name);
 677
 678         irq_info[irq] = mk_pirq_info(0, pirq, 0, 0);
 679         pirq_to_irq[pirq] = irq;
 680         ret = set_irq_msi(irq, msidesc);
 681         if (ret < 0)
 682                 goto error_irq;
 683 out:
 684         spin_unlock(&irq_mapping_update_lock);
 685         return irq;
 686 error_irq:
 687         spin_unlock(&irq_mapping_update_lock);
 688         xen_free_irq(irq);
 689         return -1;
 690 }
 691
 692 int xen_create_msi_irq(struct pci_dev *dev, struct msi_desc *msidesc, int type)
 693 {
 694         int ret, irq = -1;
 695         struct physdev_map_pirq map_irq;
 696         int rc;
 697         int pos;
 698         u32 table_offset, bir;
 699
 700         memset(&map_irq, 0, sizeof(map_irq));
 701         map_irq.domid = DOMID_SELF;
 702         map_irq.type = MAP_PIRQ_TYPE_MSI;
 703         map_irq.index = -1;
 704         map_irq.pirq = -1;
 705         map_irq.bus = dev->bus->number;
 706         map_irq.devfn = dev->devfn;
 707
 708         if (type == PCI_CAP_ID_MSIX) {
 709                 pos = pci_find_capability(dev, PCI_CAP_ID_MSIX);
 710
 711                 pci_read_config_dword(dev, msix_table_offset_reg(pos),
 712                                         &table_offset);
 713                 bir = (u8)(table_offset & PCI_MSIX_FLAGS_BIRMASK);
 714
 715                 map_irq.table_base = pci_resource_start(dev, bir);
 716                 map_irq.entry_nr = msidesc->msi_attrib.entry_nr;
 717         }
 718
 719         rc = HYPERVISOR_physdev_op(PHYSDEVOP_map_pirq, &map_irq);
 720         if (rc) {
 721                 dev_warn(&dev->dev, "xen map irq failed %d\n", rc);
 722                 goto out;
 723         }
 724
 725         spin_lock(&irq_mapping_update_lock);
 726
 727         irq = xen_allocate_irq_dynamic();
 728
 729         if (irq == -1)
 730                 goto out;
 731
 732         irq_info[irq] = mk_pirq_info(0, map_irq.pirq, 0, map_irq.index);
 733         pirq_to_irq[map_irq.pirq] = irq;
 734
 735         set_irq_chip_and_handler_name(irq, &xen_pirq_chip,
 736                         handle_level_irq,
 737                         (type == PCI_CAP_ID_MSIX) ? "msi-x":"msi");
 738
 739         ret = set_irq_msi(irq, msidesc);
 740         if (ret)
 741                 goto out_irq;
 742
 743 out:
 744         spin_unlock(&irq_mapping_update_lock);
 745         return irq;
 746 out_irq:
 747         spin_unlock(&irq_mapping_update_lock);
 748         xen_free_irq(irq);
 749         return -1;
 750 }
 751 #endif
 752
 753 int xen_destroy_irq(int irq)
 754 {
 755         struct irq_desc *desc;
 756         struct physdev_unmap_pirq unmap_irq;
 757         struct irq_info *info = info_for_irq(irq);
 758         int rc = -ENOENT;
 759
 760         spin_lock(&irq_mapping_update_lock);
 761
 762         desc = irq_to_desc(irq);
 763         if (!desc)
 764                 goto out;
 765
 766         if (xen_initial_domain()) {
 767                 unmap_irq.pirq = info->u.pirq.pirq;
 768                 unmap_irq.domid = DOMID_SELF;
 769                 rc = HYPERVISOR_physdev_op(PHYSDEVOP_unmap_pirq, &unmap_irq);
 770                 if (rc) {
 771                         printk(KERN_WARNING "unmap irq failed %d\n", rc);
 772                         goto out;
 773                 }
 774         }
 775         pirq_to_irq[info->u.pirq.pirq] = -1;
 776
 777         irq_info[irq] = mk_unbound_info();
 778
 779         xen_free_irq(irq);
 780
 781 out:
 782         spin_unlock(&irq_mapping_update_lock);
 783         return rc;
 784 }
 785
 786 int xen_vector_from_irq(unsigned irq)
 787 {
 788         return vector_from_irq(irq);
 789 }
 790
 791 int xen_gsi_from_irq(unsigned irq)
 792 {
 793         return gsi_from_irq(irq);
 794 }
 795
 796 int xen_irq_from_pirq(unsigned pirq)
 797 {
 798         return pirq_to_irq[pirq];
 799 }
 800
 801 int bind_evtchn_to_irq(unsigned int evtchn)
 802 {
 803         int irq;
 804
 805         spin_lock(&irq_mapping_update_lock);
 806
 807         irq = evtchn_to_irq[evtchn];
 808
 809         if (irq == -1) {
 810                 irq = xen_allocate_irq_dynamic();
 811
 812                 set_irq_chip_and_handler_name(irq, &xen_dynamic_chip,
 813                                               handle_fasteoi_irq, "event");
 814
 815                 evtchn_to_irq[evtchn] = irq;
 816                 irq_info[irq] = mk_evtchn_info(evtchn);
 817         }
 818
 819         spin_unlock(&irq_mapping_update_lock);
 820
 821         return irq;
 822 }
 823 EXPORT_SYMBOL_GPL(bind_evtchn_to_irq);
 824
 825 static int bind_ipi_to_irq(unsigned int ipi, unsigned int cpu)
 826 {
 827         struct evtchn_bind_ipi bind_ipi;
 828         int evtchn, irq;
 829
 830         spin_lock(&irq_mapping_update_lock);
 831
 832         irq = per_cpu(ipi_to_irq, cpu)[ipi];
 833
 834         if (irq == -1) {
 835                 irq = xen_allocate_irq_dynamic();
 836                 if (irq < 0)
 837                         goto out;
 838
 839                 set_irq_chip_and_handler_name(irq, &xen_percpu_chip,
 840                                               handle_percpu_irq, "ipi");
 841
 842                 bind_ipi.vcpu = cpu;
 843                 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_ipi,
 844                                                 &bind_ipi) != 0)
 845                         BUG();
 846                 evtchn = bind_ipi.port;
 847
 848                 evtchn_to_irq[evtchn] = irq;
 849                 irq_info[irq] = mk_ipi_info(evtchn, ipi);
 850                 per_cpu(ipi_to_irq, cpu)[ipi] = irq;
 851
 852                 bind_evtchn_to_cpu(evtchn, cpu);
 853         }
 854
 855  out:
 856         spin_unlock(&irq_mapping_update_lock);
 857         return irq;
 858 }
 859
 860
 861 int bind_virq_to_irq(unsigned int virq, unsigned int cpu)
 862 {
 863         struct evtchn_bind_virq bind_virq;
 864         int evtchn, irq;
 865
 866         spin_lock(&irq_mapping_update_lock);
 867
 868         irq = per_cpu(virq_to_irq, cpu)[virq];
 869
 870         if (irq == -1) {
 871                 irq = xen_allocate_irq_dynamic();
 872
 873                 set_irq_chip_and_handler_name(irq, &xen_percpu_chip,
 874                                               handle_percpu_irq, "virq");
 875
 876                 bind_virq.virq = virq;
 877                 bind_virq.vcpu = cpu;
 878                 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_virq,
 879                                                 &bind_virq) != 0)
 880                         BUG();
 881                 evtchn = bind_virq.port;
 882
 883                 evtchn_to_irq[evtchn] = irq;
 884                 irq_info[irq] = mk_virq_info(evtchn, virq);
 885
 886                 per_cpu(virq_to_irq, cpu)[virq] = irq;
 887
 888                 bind_evtchn_to_cpu(evtchn, cpu);
 889         }
 890
 891         spin_unlock(&irq_mapping_update_lock);
 892
 893         return irq;
 894 }
 895
 896 static void unbind_from_irq(unsigned int irq)
 897 {
 898         struct evtchn_close close;
 899         int evtchn = evtchn_from_irq(irq);
 900
 901         spin_lock(&irq_mapping_update_lock);
 902
 903         if (VALID_EVTCHN(evtchn)) {
 904                 close.port = evtchn;
 905                 if (HYPERVISOR_event_channel_op(EVTCHNOP_close, &close) != 0)
 906                         BUG();
 907
 908                 switch (type_from_irq(irq)) {
 909                 case IRQT_VIRQ:
 910                         per_cpu(virq_to_irq, cpu_from_evtchn(evtchn))
 911                                 [virq_from_irq(irq)] = -1;
 912                         break;
 913                 case IRQT_IPI:
 914                         per_cpu(ipi_to_irq, cpu_from_evtchn(evtchn))
 915                                 [ipi_from_irq(irq)] = -1;
 916                         break;
 917                 default:
 918                         break;
 919                 }
 920
 921                 /* Closed ports are implicitly re-bound to VCPU0. */
 922                 bind_evtchn_to_cpu(evtchn, 0);
 923
 924                 evtchn_to_irq[evtchn] = -1;
 925         }
 926
 927         if (irq_info[irq].type != IRQT_UNBOUND) {
 928                 irq_info[irq] = mk_unbound_info();
 929
 930                 xen_free_irq(irq);
 931         }
 932
 933         spin_unlock(&irq_mapping_update_lock);
 934 }
 935
 936 int bind_evtchn_to_irqhandler(unsigned int evtchn,
 937                               irq_handler_t handler,
 938                               unsigned long irqflags,
 939                               const char *devname, void *dev_id)
 940 {
 941         unsigned int irq;
 942         int retval;
 943
 944         irq = bind_evtchn_to_irq(evtchn);
 945         retval = request_irq(irq, handler, irqflags, devname, dev_id);
 946         if (retval != 0) {
 947                 unbind_from_irq(irq);
 948                 return retval;
 949         }
 950
 951         return irq;
 952 }
 953 EXPORT_SYMBOL_GPL(bind_evtchn_to_irqhandler);
 954
 955 int bind_virq_to_irqhandler(unsigned int virq, unsigned int cpu,
 956                             irq_handler_t handler,
 957                             unsigned long irqflags, const char *devname, void *dev_id)
 958 {
 959         unsigned int irq;
 960         int retval;
 961
 962         irq = bind_virq_to_irq(virq, cpu);
 963         retval = request_irq(irq, handler, irqflags, devname, dev_id);
 964         if (retval != 0) {
 965                 unbind_from_irq(irq);
 966                 return retval;
 967         }
 968
 969         return irq;
 970 }
 971 EXPORT_SYMBOL_GPL(bind_virq_to_irqhandler);
 972
 973 int bind_ipi_to_irqhandler(enum ipi_vector ipi,
 974                            unsigned int cpu,
 975                            irq_handler_t handler,
 976                            unsigned long irqflags,
 977                            const char *devname,
 978                            void *dev_id)
 979 {
 980         int irq, retval;
 981
 982         irq = bind_ipi_to_irq(ipi, cpu);
 983         if (irq < 0)
 984                 return irq;
 985
 986         irqflags |= IRQF_NO_SUSPEND | IRQF_FORCE_RESUME;
 987         retval = request_irq(irq, handler, irqflags, devname, dev_id);
 988         if (retval != 0) {
 989                 unbind_from_irq(irq);
 990                 return retval;
 991         }
 992
 993         return irq;
 994 }
 995
 996 void unbind_from_irqhandler(unsigned int irq, void *dev_id)
 997 {
 998         free_irq(irq, dev_id);
 999         unbind_from_irq(irq);
1000 }
1001 EXPORT_SYMBOL_GPL(unbind_from_irqhandler);
1002
1003 void xen_send_IPI_one(unsigned int cpu, enum ipi_vector vector)
1004 {
1005         int irq = per_cpu(ipi_to_irq, cpu)[vector];
1006         BUG_ON(irq < 0);
1007         notify_remote_via_irq(irq);
1008 }
1009
1010 irqreturn_t xen_debug_interrupt(int irq, void *dev_id)
1011 {
1012         struct shared_info *sh = HYPERVISOR_shared_info;
1013         int cpu = smp_processor_id();
1014         unsigned long *cpu_evtchn = cpu_evtchn_mask(cpu);
1015         int i;
1016         unsigned long flags;
1017         static DEFINE_SPINLOCK(debug_lock);
1018         struct vcpu_info *v;
1019
1020         spin_lock_irqsave(&debug_lock, flags);
1021
1022         printk("\nvcpu %d\n  ", cpu);
1023
1024         for_each_online_cpu(i) {
1025                 int pending;
1026                 v = per_cpu(xen_vcpu, i);
1027                 pending = (get_irq_regs() && i == cpu)
1028                         ? xen_irqs_disabled(get_irq_regs())
1029                         : v->evtchn_upcall_mask;
1030                 printk("%d: masked=%d pending=%d event_sel %0*lx\n  ", i,
1031                        pending, v->evtchn_upcall_pending,
1032                        (int)(sizeof(v->evtchn_pending_sel)*2),
1033                        v->evtchn_pending_sel);
1034         }
1035         v = per_cpu(xen_vcpu, cpu);
1036
1037         printk("\npending:\n   ");
1038         for (i = ARRAY_SIZE(sh->evtchn_pending)-1; i >= 0; i--)
1039                 printk("%0*lx%s", (int)sizeof(sh->evtchn_pending[0])*2,
1040                        sh->evtchn_pending[i],
1041                        i % 8 == 0 ? "\n   " : " ");
1042         printk("\nglobal mask:\n   ");
1043         for (i = ARRAY_SIZE(sh->evtchn_mask)-1; i >= 0; i--)
1044                 printk("%0*lx%s",
1045                        (int)(sizeof(sh->evtchn_mask[0])*2),
1046                        sh->evtchn_mask[i],
1047                        i % 8 == 0 ? "\n   " : " ");
1048
1049         printk("\nglobally unmasked:\n   ");
1050         for (i = ARRAY_SIZE(sh->evtchn_mask)-1; i >= 0; i--)
1051                 printk("%0*lx%s", (int)(sizeof(sh->evtchn_mask[0])*2),
1052                        sh->evtchn_pending[i] & ~sh->evtchn_mask[i],
1053                        i % 8 == 0 ? "\n   " : " ");
1054
1055         printk("\nlocal cpu%d mask:\n   ", cpu);
1056         for (i = (NR_EVENT_CHANNELS/BITS_PER_LONG)-1; i >= 0; i--)
1057                 printk("%0*lx%s", (int)(sizeof(cpu_evtchn[0])*2),
1058                        cpu_evtchn[i],
1059                        i % 8 == 0 ? "\n   " : " ");
1060
1061         printk("\nlocally unmasked:\n   ");
1062         for (i = ARRAY_SIZE(sh->evtchn_mask)-1; i >= 0; i--) {
1063                 unsigned long pending = sh->evtchn_pending[i]
1064                         & ~sh->evtchn_mask[i]
1065                         & cpu_evtchn[i];
1066                 printk("%0*lx%s", (int)(sizeof(sh->evtchn_mask[0])*2),
1067                        pending, i % 8 == 0 ? "\n   " : " ");
1068         }
1069
1070         printk("\npending list:\n");
1071         for (i = 0; i < NR_EVENT_CHANNELS; i++) {
1072                 if (sync_test_bit(i, sh->evtchn_pending)) {
1073                         int word_idx = i / BITS_PER_LONG;
1074                         printk("  %d: event %d -> irq %d%s%s%s\n",
1075                                cpu_from_evtchn(i), i,
1076                                evtchn_to_irq[i],
1077                                sync_test_bit(word_idx, &v->evtchn_pending_sel)
1078                                              ? "" : " l2-clear",
1079                                !sync_test_bit(i, sh->evtchn_mask)
1080                                              ? "" : " globally-masked",
1081                                sync_test_bit(i, cpu_evtchn)
1082                                              ? "" : " locally-masked");
1083                 }
1084         }
1085
1086         spin_unlock_irqrestore(&debug_lock, flags);
1087
1088         return IRQ_HANDLED;
1089 }
1090
1091 static DEFINE_PER_CPU(unsigned, xed_nesting_count);
1092
1093 /*
1094  * Search the CPUs pending events bitmasks.  For each one found, map
1095  * the event number to an irq, and feed it into do_IRQ() for
1096  * handling.
1097  *
1098  * Xen uses a two-level bitmap to speed searching.  The first level is
1099  * a bitset of words which contain pending event bits.  The second
1100  * level is a bitset of pending events themselves.
1101  */
1102 static void __xen_evtchn_do_upcall(void)
1103 {
1104         int cpu = get_cpu();
1105         struct shared_info *s = HYPERVISOR_shared_info;
1106         struct vcpu_info *vcpu_info = __this_cpu_read(xen_vcpu);
1107         unsigned count;
1108
1109         do {
1110                 unsigned long pending_words;
1111
1112                 vcpu_info->evtchn_upcall_pending = 0;
1113
1114                 if (__this_cpu_inc_return(xed_nesting_count) - 1)
1115                         goto out;
1116
1117 #ifndef CONFIG_X86 /* No need for a barrier -- XCHG is a barrier on x86. */
1118                 /* Clear master flag /before/ clearing selector flag. */
1119                 wmb();
1120 #endif
1121                 pending_words = xchg(&vcpu_info->evtchn_pending_sel, 0);
1122                 while (pending_words != 0) {
1123                         unsigned long pending_bits;
1124                         int word_idx = __ffs(pending_words);
1125                         pending_words &= ~(1UL << word_idx);
1126
1127                         while ((pending_bits = active_evtchns(cpu, s, word_idx)) != 0) {
1128                                 int bit_idx = __ffs(pending_bits);
1129                                 int port = (word_idx * BITS_PER_LONG) + bit_idx;
1130                                 int irq = evtchn_to_irq[port];
1131                                 struct irq_desc *desc;
1132
1133                                 mask_evtchn(port);
1134                                 clear_evtchn(port);
1135
1136                                 if (irq != -1) {
1137                                         desc = irq_to_desc(irq);
1138                                         if (desc)
1139                                                 generic_handle_irq_desc(irq, desc);
1140                                 }
1141                         }
1142                 }
1143
1144                 BUG_ON(!irqs_disabled());
1145
1146                 count = __this_cpu_read(xed_nesting_count);
1147                 __this_cpu_write(xed_nesting_count, 0);
1148         } while (count != 1 || vcpu_info->evtchn_upcall_pending);
1149
1150 out:
1151
1152         put_cpu();
1153 }
1154
1155 void xen_evtchn_do_upcall(struct pt_regs *regs)
1156 {
1157         struct pt_regs *old_regs = set_irq_regs(regs);
1158
1159         exit_idle();
1160         irq_enter();
1161
1162         __xen_evtchn_do_upcall();
1163
1164         irq_exit();
1165         set_irq_regs(old_regs);
1166 }
1167
1168 void xen_hvm_evtchn_do_upcall(void)
1169 {
1170         __xen_evtchn_do_upcall();
1171 }
1172 EXPORT_SYMBOL_GPL(xen_hvm_evtchn_do_upcall);
1173
1174 /* Rebind a new event channel to an existing irq. */
1175 void rebind_evtchn_irq(int evtchn, int irq)
1176 {
1177         struct irq_info *info = info_for_irq(irq);
1178
1179         /* Make sure the irq is masked, since the new event channel
1180            will also be masked. */
1181         disable_irq(irq);
1182
1183         spin_lock(&irq_mapping_update_lock);
1184
1185         /* After resume the irq<->evtchn mappings are all cleared out */
1186         BUG_ON(evtchn_to_irq[evtchn] != -1);
1187         /* Expect irq to have been bound before,
1188            so there should be a proper type */
1189         BUG_ON(info->type == IRQT_UNBOUND);
1190
1191         evtchn_to_irq[evtchn] = irq;
1192         irq_info[irq] = mk_evtchn_info(evtchn);
1193
1194         spin_unlock(&irq_mapping_update_lock);
1195
1196         /* new event channels are always bound to cpu 0 */
1197         irq_set_affinity(irq, cpumask_of(0));
1198
1199         /* Unmask the event channel. */
1200         enable_irq(irq);
1201 }
1202
1203 /* Rebind an evtchn so that it gets delivered to a specific cpu */
1204 static int rebind_irq_to_cpu(unsigned irq, unsigned tcpu)
1205 {
1206         struct evtchn_bind_vcpu bind_vcpu;
1207         int evtchn = evtchn_from_irq(irq);
1208
1209         /* events delivered via platform PCI interrupts are always
1210          * routed to vcpu 0 */
1211         if (!VALID_EVTCHN(evtchn) ||
1212                 (xen_hvm_domain() && !xen_have_vector_callback))
1213                 return -1;
1214
1215         /* Send future instances of this interrupt to other vcpu. */
1216         bind_vcpu.port = evtchn;
1217         bind_vcpu.vcpu = tcpu;
1218
1219         /*
1220          * If this fails, it usually just indicates that we're dealing with a
1221          * virq or IPI channel, which don't actually need to be rebound. Ignore
1222          * it, but don't do the xenlinux-level rebind in that case.
1223          */
1224         if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_vcpu, &bind_vcpu) >= 0)
1225                 bind_evtchn_to_cpu(evtchn, tcpu);
1226
1227         return 0;
1228 }
1229
1230 static int set_affinity_irq(struct irq_data *data, const struct cpumask *dest,
1231                             bool force)
1232 {
1233         unsigned tcpu = cpumask_first(dest);
1234
1235         return rebind_irq_to_cpu(data->irq, tcpu);
1236 }
1237
1238 int resend_irq_on_evtchn(unsigned int irq)
1239 {
1240         int masked, evtchn = evtchn_from_irq(irq);
1241         struct shared_info *s = HYPERVISOR_shared_info;
1242
1243         if (!VALID_EVTCHN(evtchn))
1244                 return 1;
1245
1246         masked = sync_test_and_set_bit(evtchn, s->evtchn_mask);
1247         sync_set_bit(evtchn, s->evtchn_pending);
1248         if (!masked)
1249                 unmask_evtchn(evtchn);
1250
1251         return 1;
1252 }
1253
1254 static void enable_dynirq(struct irq_data *data)
1255 {
1256         int evtchn = evtchn_from_irq(data->irq);
1257
1258         if (VALID_EVTCHN(evtchn))
1259                 unmask_evtchn(evtchn);
1260 }
1261
1262 static void disable_dynirq(struct irq_data *data)
1263 {
1264         int evtchn = evtchn_from_irq(data->irq);
1265
1266         if (VALID_EVTCHN(evtchn))
1267                 mask_evtchn(evtchn);
1268 }
1269
1270 static void ack_dynirq(struct irq_data *data)
1271 {
1272         int evtchn = evtchn_from_irq(data->irq);
1273
1274         move_masked_irq(data->irq);
1275
1276         if (VALID_EVTCHN(evtchn))
1277                 unmask_evtchn(evtchn);
1278 }
1279
1280 static int retrigger_dynirq(struct irq_data *data)
1281 {
1282         int evtchn = evtchn_from_irq(data->irq);
1283         struct shared_info *sh = HYPERVISOR_shared_info;
1284         int ret = 0;
1285
1286         if (VALID_EVTCHN(evtchn)) {
1287                 int masked;
1288
1289                 masked = sync_test_and_set_bit(evtchn, sh->evtchn_mask);
1290                 sync_set_bit(evtchn, sh->evtchn_pending);
1291                 if (!masked)
1292                         unmask_evtchn(evtchn);
1293                 ret = 1;
1294         }
1295
1296         return ret;
1297 }
1298
1299 static void restore_cpu_pirqs(void)
1300 {
1301         int pirq, rc, irq, gsi;
1302         struct physdev_map_pirq map_irq;
1303
1304         for (pirq = 0; pirq < nr_irqs; pirq++) {
1305                 irq = pirq_to_irq[pirq];
1306                 if (irq == -1)
1307                         continue;
1308
1309                 /* save/restore of PT devices doesn't work, so at this point the
1310                  * only devices present are GSI based emulated devices */
1311                 gsi = gsi_from_irq(irq);
1312                 if (!gsi)
1313                         continue;
1314
1315                 map_irq.domid = DOMID_SELF;
1316                 map_irq.type = MAP_PIRQ_TYPE_GSI;
1317                 map_irq.index = gsi;
1318                 map_irq.pirq = pirq;
1319
1320                 rc = HYPERVISOR_physdev_op(PHYSDEVOP_map_pirq, &map_irq);
1321                 if (rc) {
1322                         printk(KERN_WARNING "xen map irq failed gsi=%d irq=%d pirq=%d rc=%d\n",
1323                                         gsi, irq, pirq, rc);
1324                         irq_info[irq] = mk_unbound_info();
1325                         pirq_to_irq[pirq] = -1;
1326                         continue;
1327                 }
1328
1329                 printk(KERN_DEBUG "xen: --> irq=%d, pirq=%d\n", irq, map_irq.pirq);
1330
1331                 __startup_pirq(irq);
1332         }
1333 }
1334
1335 static void restore_cpu_virqs(unsigned int cpu)
1336 {
1337         struct evtchn_bind_virq bind_virq;
1338         int virq, irq, evtchn;
1339
1340         for (virq = 0; virq < NR_VIRQS; virq++) {
1341                 if ((irq = per_cpu(virq_to_irq, cpu)[virq]) == -1)
1342                         continue;
1343
1344                 BUG_ON(virq_from_irq(irq) != virq);
1345
1346                 /* Get a new binding from Xen. */
1347                 bind_virq.virq = virq;
1348                 bind_virq.vcpu = cpu;
1349                 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_virq,
1350                                                 &bind_virq) != 0)
1351                         BUG();
1352                 evtchn = bind_virq.port;
1353
1354                 /* Record the new mapping. */
1355                 evtchn_to_irq[evtchn] = irq;
1356                 irq_info[irq] = mk_virq_info(evtchn, virq);
1357                 bind_evtchn_to_cpu(evtchn, cpu);
1358         }
1359 }
1360
1361 static void restore_cpu_ipis(unsigned int cpu)
1362 {
1363         struct evtchn_bind_ipi bind_ipi;
1364         int ipi, irq, evtchn;
1365
1366         for (ipi = 0; ipi < XEN_NR_IPIS; ipi++) {
1367                 if ((irq = per_cpu(ipi_to_irq, cpu)[ipi]) == -1)
1368                         continue;
1369
1370                 BUG_ON(ipi_from_irq(irq) != ipi);
1371
1372                 /* Get a new binding from Xen. */
1373                 bind_ipi.vcpu = cpu;
1374                 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_ipi,
1375                                                 &bind_ipi) != 0)
1376                         BUG();
1377                 evtchn = bind_ipi.port;
1378
1379                 /* Record the new mapping. */
1380                 evtchn_to_irq[evtchn] = irq;
1381                 irq_info[irq] = mk_ipi_info(evtchn, ipi);
1382                 bind_evtchn_to_cpu(evtchn, cpu);
1383         }
1384 }
1385
1386 /* Clear an irq's pending state, in preparation for polling on it */
1387 void xen_clear_irq_pending(int irq)
1388 {
1389         int evtchn = evtchn_from_irq(irq);
1390
1391         if (VALID_EVTCHN(evtchn))
1392                 clear_evtchn(evtchn);
1393 }
1394 EXPORT_SYMBOL(xen_clear_irq_pending);
1395 void xen_set_irq_pending(int irq)
1396 {
1397         int evtchn = evtchn_from_irq(irq);
1398
1399         if (VALID_EVTCHN(evtchn))
1400                 set_evtchn(evtchn);
1401 }
1402
1403 bool xen_test_irq_pending(int irq)
1404 {
1405         int evtchn = evtchn_from_irq(irq);
1406         bool ret = false;
1407
1408         if (VALID_EVTCHN(evtchn))
1409                 ret = test_evtchn(evtchn);
1410
1411         return ret;
1412 }
1413
1414 /* Poll waiting for an irq to become pending with timeout.  In the usual case,
1415  * the irq will be disabled so it won't deliver an interrupt. */
1416 void xen_poll_irq_timeout(int irq, u64 timeout)
1417 {
1418         evtchn_port_t evtchn = evtchn_from_irq(irq);
1419
1420         if (VALID_EVTCHN(evtchn)) {
1421                 struct sched_poll poll;
1422
1423                 poll.nr_ports = 1;
1424                 poll.timeout = timeout;
1425                 set_xen_guest_handle(poll.ports, &evtchn);
1426
1427                 if (HYPERVISOR_sched_op(SCHEDOP_poll, &poll) != 0)
1428                         BUG();
1429         }
1430 }
1431 EXPORT_SYMBOL(xen_poll_irq_timeout);
1432 /* Poll waiting for an irq to become pending.  In the usual case, the
1433  * irq will be disabled so it won't deliver an interrupt. */
1434 void xen_poll_irq(int irq)
1435 {
1436         xen_poll_irq_timeout(irq, 0 /* no timeout */);
1437 }
1438
1439 void xen_irq_resume(void)
1440 {
1441         unsigned int cpu, irq, evtchn;
1442
1443         init_evtchn_cpu_bindings();
1444
1445         /* New event-channel space is not 'live' yet. */
1446         for (evtchn = 0; evtchn < NR_EVENT_CHANNELS; evtchn++)
1447                 mask_evtchn(evtchn);
1448
1449         /* No IRQ <-> event-channel mappings. */
1450         for (irq = 0; irq < nr_irqs; irq++)
1451                 irq_info[irq].evtchn = 0; /* zap event-channel binding */
1452
1453         for (evtchn = 0; evtchn < NR_EVENT_CHANNELS; evtchn++)
1454                 evtchn_to_irq[evtchn] = -1;
1455
1456         for_each_possible_cpu(cpu) {
1457                 restore_cpu_virqs(cpu);
1458                 restore_cpu_ipis(cpu);
1459         }
1460
1461         restore_cpu_pirqs();
1462 }
1463
1464 static struct irq_chip xen_dynamic_chip __read_mostly = {
1465         .name                   = "xen-dyn",
1466
1467         .irq_disable            = disable_dynirq,
1468         .irq_mask               = disable_dynirq,
1469         .irq_unmask             = enable_dynirq,
1470
1471         .irq_eoi                = ack_dynirq,
1472         .irq_set_affinity       = set_affinity_irq,
1473         .irq_retrigger          = retrigger_dynirq,
1474 };
1475
1476 static struct irq_chip xen_pirq_chip __read_mostly = {
1477         .name                   = "xen-pirq",
1478
1479         .irq_startup            = startup_pirq,
1480         .irq_shutdown           = shutdown_pirq,
1481
1482         .irq_enable             = enable_pirq,
1483         .irq_unmask             = enable_pirq,
1484
1485         .irq_disable            = disable_pirq,
1486         .irq_mask               = disable_pirq,
1487
1488         .irq_ack                = ack_pirq,
1489
1490         .irq_set_affinity       = set_affinity_irq,
1491
1492         .irq_retrigger          = retrigger_dynirq,
1493 };
1494
1495 static struct irq_chip xen_percpu_chip __read_mostly = {
1496         .name                   = "xen-percpu",
1497
1498         .irq_disable            = disable_dynirq,
1499         .irq_mask               = disable_dynirq,
1500         .irq_unmask             = enable_dynirq,
1501
1502         .irq_ack                = ack_dynirq,
1503 };
1504
1505 int xen_set_callback_via(uint64_t via)
1506 {
1507         struct xen_hvm_param a;
1508         a.domid = DOMID_SELF;
1509         a.index = HVM_PARAM_CALLBACK_IRQ;
1510         a.value = via;
1511         return HYPERVISOR_hvm_op(HVMOP_set_param, &a);
1512 }
1513 EXPORT_SYMBOL_GPL(xen_set_callback_via);
1514
1515 #ifdef CONFIG_XEN_PVHVM
1516 /* Vector callbacks are better than PCI interrupts to receive event
1517  * channel notifications because we can receive vector callbacks on any
1518  * vcpu and we don't need PCI support or APIC interactions. */
1519 void xen_callback_vector(void)
1520 {
1521         int rc;
1522         uint64_t callback_via;
1523         if (xen_have_vector_callback) {
1524                 callback_via = HVM_CALLBACK_VECTOR(XEN_HVM_EVTCHN_CALLBACK);
1525                 rc = xen_set_callback_via(callback_via);
1526                 if (rc) {
1527                         printk(KERN_ERR "Request for Xen HVM callback vector"
1528                                         " failed.\n");
1529                         xen_have_vector_callback = 0;
1530                         return;
1531                 }
1532                 printk(KERN_INFO "Xen HVM callback vector for event delivery is "
1533                                 "enabled\n");
1534                 /* in the restore case the vector has already been allocated */
1535                 if (!test_bit(XEN_HVM_EVTCHN_CALLBACK, used_vectors))
1536                         alloc_intr_gate(XEN_HVM_EVTCHN_CALLBACK, xen_hvm_callback_vector);
1537         }
1538 }
1539 #else
1540 void xen_callback_vector(void) {}
1541 #endif
1542
1543 void __init xen_init_IRQ(void)
1544 {
1545         int i;
1546
1547         cpu_evtchn_mask_p = kcalloc(nr_cpu_ids, sizeof(struct cpu_evtchn_s),
1548                                     GFP_KERNEL);
1549         irq_info = kcalloc(nr_irqs, sizeof(*irq_info), GFP_KERNEL);
1550
1551         /* We are using nr_irqs as the maximum number of pirq available but
1552          * that number is actually chosen by Xen and we don't know exactly
1553          * what it is. Be careful choosing high pirq numbers. */
1554         pirq_to_irq = kcalloc(nr_irqs, sizeof(*pirq_to_irq), GFP_KERNEL);
1555         for (i = 0; i < nr_irqs; i++)
1556                 pirq_to_irq[i] = -1;
1557
1558         evtchn_to_irq = kcalloc(NR_EVENT_CHANNELS, sizeof(*evtchn_to_irq),
1559                                     GFP_KERNEL);
1560         for (i = 0; i < NR_EVENT_CHANNELS; i++)
1561                 evtchn_to_irq[i] = -1;
1562
1563         init_evtchn_cpu_bindings();
1564
1565         /* No event channels are 'live' right now. */
1566         for (i = 0; i < NR_EVENT_CHANNELS; i++)
1567                 mask_evtchn(i);
1568
1569         if (xen_hvm_domain()) {
1570                 xen_callback_vector();
1571                 native_init_IRQ();
1572                 /* pci_xen_hvm_init must be called after native_init_IRQ so that
1573                  * __acpi_register_gsi can point at the right function */
1574                 pci_xen_hvm_init();
1575         } else {
1576                 irq_ctx_init(smp_processor_id());
1577                 if (xen_initial_domain())
1578                         xen_setup_pirqs();
1579         }
1580 }