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 static int find_unbound_pirq(int type)
 651 {
 652         int rc, i;
 653         struct physdev_get_free_pirq op_get_free_pirq;
 654         op_get_free_pirq.type = type;
 655
 656         rc = HYPERVISOR_physdev_op(PHYSDEVOP_get_free_pirq, &op_get_free_pirq);
 657         if (!rc)
 658                 return op_get_free_pirq.pirq;
 659
 660         for (i = 0; i < nr_irqs; i++) {
 661                 if (pirq_to_irq[i] < 0)
 662                         return i;
 663         }
 664         return -1;
 665 }
 666
 667 void xen_allocate_pirq_msi(char *name, int *irq, int *pirq, int alloc_pirq)
 668 {
 669         spin_lock(&irq_mapping_update_lock);
 670
 671         *irq = xen_allocate_irq_dynamic();
 672         if (*irq == -1)
 673                 goto out;
 674
 675         if (alloc_pirq) {
 676                 *pirq = find_unbound_pirq(MAP_PIRQ_TYPE_MSI);
 677                 if (*pirq == -1) {
 678                         xen_free_irq(*irq);
 679                         *irq = -1;
 680                         goto out;
 681                 }
 682         }
 683
 684         set_irq_chip_and_handler_name(*irq, &xen_pirq_chip,
 685                                       handle_level_irq, name);
 686
 687         irq_info[*irq] = mk_pirq_info(0, *pirq, 0, 0);
 688         pirq_to_irq[*pirq] = *irq;
 689
 690 out:
 691         spin_unlock(&irq_mapping_update_lock);
 692 }
 693
 694 int xen_create_msi_irq(struct pci_dev *dev, struct msi_desc *msidesc, int type)
 695 {
 696         int irq = -1;
 697         struct physdev_map_pirq map_irq;
 698         int rc;
 699         int pos;
 700         u32 table_offset, bir;
 701
 702         memset(&map_irq, 0, sizeof(map_irq));
 703         map_irq.domid = DOMID_SELF;
 704         map_irq.type = MAP_PIRQ_TYPE_MSI;
 705         map_irq.index = -1;
 706         map_irq.pirq = -1;
 707         map_irq.bus = dev->bus->number;
 708         map_irq.devfn = dev->devfn;
 709
 710         if (type == PCI_CAP_ID_MSIX) {
 711                 pos = pci_find_capability(dev, PCI_CAP_ID_MSIX);
 712
 713                 pci_read_config_dword(dev, msix_table_offset_reg(pos),
 714                                         &table_offset);
 715                 bir = (u8)(table_offset & PCI_MSIX_FLAGS_BIRMASK);
 716
 717                 map_irq.table_base = pci_resource_start(dev, bir);
 718                 map_irq.entry_nr = msidesc->msi_attrib.entry_nr;
 719         }
 720
 721         spin_lock(&irq_mapping_update_lock);
 722
 723         irq = xen_allocate_irq_dynamic();
 724
 725         if (irq == -1)
 726                 goto out;
 727
 728         rc = HYPERVISOR_physdev_op(PHYSDEVOP_map_pirq, &map_irq);
 729         if (rc) {
 730                 printk(KERN_WARNING "xen map irq failed %d\n", rc);
 731
 732                 xen_free_irq(irq);
 733
 734                 irq = -1;
 735                 goto out;
 736         }
 737         irq_info[irq] = mk_pirq_info(0, map_irq.pirq, 0, map_irq.index);
 738
 739         set_irq_chip_and_handler_name(irq, &xen_pirq_chip,
 740                         handle_level_irq,
 741                         (type == PCI_CAP_ID_MSIX) ? "msi-x":"msi");
 742
 743 out:
 744         spin_unlock(&irq_mapping_update_lock);
 745         return irq;
 746 }
 747 #endif
 748
 749 int xen_destroy_irq(int irq)
 750 {
 751         struct irq_desc *desc;
 752         struct physdev_unmap_pirq unmap_irq;
 753         struct irq_info *info = info_for_irq(irq);
 754         int rc = -ENOENT;
 755
 756         spin_lock(&irq_mapping_update_lock);
 757
 758         desc = irq_to_desc(irq);
 759         if (!desc)
 760                 goto out;
 761
 762         if (xen_initial_domain()) {
 763                 unmap_irq.pirq = info->u.pirq.pirq;
 764                 unmap_irq.domid = DOMID_SELF;
 765                 rc = HYPERVISOR_physdev_op(PHYSDEVOP_unmap_pirq, &unmap_irq);
 766                 if (rc) {
 767                         printk(KERN_WARNING "unmap irq failed %d\n", rc);
 768                         goto out;
 769                 }
 770         }
 771         pirq_to_irq[info->u.pirq.pirq] = -1;
 772
 773         irq_info[irq] = mk_unbound_info();
 774
 775         xen_free_irq(irq);
 776
 777 out:
 778         spin_unlock(&irq_mapping_update_lock);
 779         return rc;
 780 }
 781
 782 int xen_vector_from_irq(unsigned irq)
 783 {
 784         return vector_from_irq(irq);
 785 }
 786
 787 int xen_gsi_from_irq(unsigned irq)
 788 {
 789         return gsi_from_irq(irq);
 790 }
 791
 792 int xen_irq_from_pirq(unsigned pirq)
 793 {
 794         return pirq_to_irq[pirq];
 795 }
 796
 797 int bind_evtchn_to_irq(unsigned int evtchn)
 798 {
 799         int irq;
 800
 801         spin_lock(&irq_mapping_update_lock);
 802
 803         irq = evtchn_to_irq[evtchn];
 804
 805         if (irq == -1) {
 806                 irq = xen_allocate_irq_dynamic();
 807
 808                 set_irq_chip_and_handler_name(irq, &xen_dynamic_chip,
 809                                               handle_fasteoi_irq, "event");
 810
 811                 evtchn_to_irq[evtchn] = irq;
 812                 irq_info[irq] = mk_evtchn_info(evtchn);
 813         }
 814
 815         spin_unlock(&irq_mapping_update_lock);
 816
 817         return irq;
 818 }
 819 EXPORT_SYMBOL_GPL(bind_evtchn_to_irq);
 820
 821 static int bind_ipi_to_irq(unsigned int ipi, unsigned int cpu)
 822 {
 823         struct evtchn_bind_ipi bind_ipi;
 824         int evtchn, irq;
 825
 826         spin_lock(&irq_mapping_update_lock);
 827
 828         irq = per_cpu(ipi_to_irq, cpu)[ipi];
 829
 830         if (irq == -1) {
 831                 irq = xen_allocate_irq_dynamic();
 832                 if (irq < 0)
 833                         goto out;
 834
 835                 set_irq_chip_and_handler_name(irq, &xen_percpu_chip,
 836                                               handle_percpu_irq, "ipi");
 837
 838                 bind_ipi.vcpu = cpu;
 839                 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_ipi,
 840                                                 &bind_ipi) != 0)
 841                         BUG();
 842                 evtchn = bind_ipi.port;
 843
 844                 evtchn_to_irq[evtchn] = irq;
 845                 irq_info[irq] = mk_ipi_info(evtchn, ipi);
 846                 per_cpu(ipi_to_irq, cpu)[ipi] = irq;
 847
 848                 bind_evtchn_to_cpu(evtchn, cpu);
 849         }
 850
 851  out:
 852         spin_unlock(&irq_mapping_update_lock);
 853         return irq;
 854 }
 855
 856
 857 int bind_virq_to_irq(unsigned int virq, unsigned int cpu)
 858 {
 859         struct evtchn_bind_virq bind_virq;
 860         int evtchn, irq;
 861
 862         spin_lock(&irq_mapping_update_lock);
 863
 864         irq = per_cpu(virq_to_irq, cpu)[virq];
 865
 866         if (irq == -1) {
 867                 irq = xen_allocate_irq_dynamic();
 868
 869                 set_irq_chip_and_handler_name(irq, &xen_percpu_chip,
 870                                               handle_percpu_irq, "virq");
 871
 872                 bind_virq.virq = virq;
 873                 bind_virq.vcpu = cpu;
 874                 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_virq,
 875                                                 &bind_virq) != 0)
 876                         BUG();
 877                 evtchn = bind_virq.port;
 878
 879                 evtchn_to_irq[evtchn] = irq;
 880                 irq_info[irq] = mk_virq_info(evtchn, virq);
 881
 882                 per_cpu(virq_to_irq, cpu)[virq] = irq;
 883
 884                 bind_evtchn_to_cpu(evtchn, cpu);
 885         }
 886
 887         spin_unlock(&irq_mapping_update_lock);
 888
 889         return irq;
 890 }
 891
 892 static void unbind_from_irq(unsigned int irq)
 893 {
 894         struct evtchn_close close;
 895         int evtchn = evtchn_from_irq(irq);
 896
 897         spin_lock(&irq_mapping_update_lock);
 898
 899         if (VALID_EVTCHN(evtchn)) {
 900                 close.port = evtchn;
 901                 if (HYPERVISOR_event_channel_op(EVTCHNOP_close, &close) != 0)
 902                         BUG();
 903
 904                 switch (type_from_irq(irq)) {
 905                 case IRQT_VIRQ:
 906                         per_cpu(virq_to_irq, cpu_from_evtchn(evtchn))
 907                                 [virq_from_irq(irq)] = -1;
 908                         break;
 909                 case IRQT_IPI:
 910                         per_cpu(ipi_to_irq, cpu_from_evtchn(evtchn))
 911                                 [ipi_from_irq(irq)] = -1;
 912                         break;
 913                 default:
 914                         break;
 915                 }
 916
 917                 /* Closed ports are implicitly re-bound to VCPU0. */
 918                 bind_evtchn_to_cpu(evtchn, 0);
 919
 920                 evtchn_to_irq[evtchn] = -1;
 921         }
 922
 923         if (irq_info[irq].type != IRQT_UNBOUND) {
 924                 irq_info[irq] = mk_unbound_info();
 925
 926                 xen_free_irq(irq);
 927         }
 928
 929         spin_unlock(&irq_mapping_update_lock);
 930 }
 931
 932 int bind_evtchn_to_irqhandler(unsigned int evtchn,
 933                               irq_handler_t handler,
 934                               unsigned long irqflags,
 935                               const char *devname, void *dev_id)
 936 {
 937         unsigned int irq;
 938         int retval;
 939
 940         irq = bind_evtchn_to_irq(evtchn);
 941         retval = request_irq(irq, handler, irqflags, devname, dev_id);
 942         if (retval != 0) {
 943                 unbind_from_irq(irq);
 944                 return retval;
 945         }
 946
 947         return irq;
 948 }
 949 EXPORT_SYMBOL_GPL(bind_evtchn_to_irqhandler);
 950
 951 int bind_virq_to_irqhandler(unsigned int virq, unsigned int cpu,
 952                             irq_handler_t handler,
 953                             unsigned long irqflags, const char *devname, void *dev_id)
 954 {
 955         unsigned int irq;
 956         int retval;
 957
 958         irq = bind_virq_to_irq(virq, cpu);
 959         retval = request_irq(irq, handler, irqflags, devname, dev_id);
 960         if (retval != 0) {
 961                 unbind_from_irq(irq);
 962                 return retval;
 963         }
 964
 965         return irq;
 966 }
 967 EXPORT_SYMBOL_GPL(bind_virq_to_irqhandler);
 968
 969 int bind_ipi_to_irqhandler(enum ipi_vector ipi,
 970                            unsigned int cpu,
 971                            irq_handler_t handler,
 972                            unsigned long irqflags,
 973                            const char *devname,
 974                            void *dev_id)
 975 {
 976         int irq, retval;
 977
 978         irq = bind_ipi_to_irq(ipi, cpu);
 979         if (irq < 0)
 980                 return irq;
 981
 982         irqflags |= IRQF_NO_SUSPEND | IRQF_FORCE_RESUME;
 983         retval = request_irq(irq, handler, irqflags, devname, dev_id);
 984         if (retval != 0) {
 985                 unbind_from_irq(irq);
 986                 return retval;
 987         }
 988
 989         return irq;
 990 }
 991
 992 void unbind_from_irqhandler(unsigned int irq, void *dev_id)
 993 {
 994         free_irq(irq, dev_id);
 995         unbind_from_irq(irq);
 996 }
 997 EXPORT_SYMBOL_GPL(unbind_from_irqhandler);
 998
 999 void xen_send_IPI_one(unsigned int cpu, enum ipi_vector vector)
1000 {
1001         int irq = per_cpu(ipi_to_irq, cpu)[vector];
1002         BUG_ON(irq < 0);
1003         notify_remote_via_irq(irq);
1004 }
1005
1006 irqreturn_t xen_debug_interrupt(int irq, void *dev_id)
1007 {
1008         struct shared_info *sh = HYPERVISOR_shared_info;
1009         int cpu = smp_processor_id();
1010         unsigned long *cpu_evtchn = cpu_evtchn_mask(cpu);
1011         int i;
1012         unsigned long flags;
1013         static DEFINE_SPINLOCK(debug_lock);
1014         struct vcpu_info *v;
1015
1016         spin_lock_irqsave(&debug_lock, flags);
1017
1018         printk("\nvcpu %d\n  ", cpu);
1019
1020         for_each_online_cpu(i) {
1021                 int pending;
1022                 v = per_cpu(xen_vcpu, i);
1023                 pending = (get_irq_regs() && i == cpu)
1024                         ? xen_irqs_disabled(get_irq_regs())
1025                         : v->evtchn_upcall_mask;
1026                 printk("%d: masked=%d pending=%d event_sel %0*lx\n  ", i,
1027                        pending, v->evtchn_upcall_pending,
1028                        (int)(sizeof(v->evtchn_pending_sel)*2),
1029                        v->evtchn_pending_sel);
1030         }
1031         v = per_cpu(xen_vcpu, cpu);
1032
1033         printk("\npending:\n   ");
1034         for (i = ARRAY_SIZE(sh->evtchn_pending)-1; i >= 0; i--)
1035                 printk("%0*lx%s", (int)sizeof(sh->evtchn_pending[0])*2,
1036                        sh->evtchn_pending[i],
1037                        i % 8 == 0 ? "\n   " : " ");
1038         printk("\nglobal mask:\n   ");
1039         for (i = ARRAY_SIZE(sh->evtchn_mask)-1; i >= 0; i--)
1040                 printk("%0*lx%s",
1041                        (int)(sizeof(sh->evtchn_mask[0])*2),
1042                        sh->evtchn_mask[i],
1043                        i % 8 == 0 ? "\n   " : " ");
1044
1045         printk("\nglobally unmasked:\n   ");
1046         for (i = ARRAY_SIZE(sh->evtchn_mask)-1; i >= 0; i--)
1047                 printk("%0*lx%s", (int)(sizeof(sh->evtchn_mask[0])*2),
1048                        sh->evtchn_pending[i] & ~sh->evtchn_mask[i],
1049                        i % 8 == 0 ? "\n   " : " ");
1050
1051         printk("\nlocal cpu%d mask:\n   ", cpu);
1052         for (i = (NR_EVENT_CHANNELS/BITS_PER_LONG)-1; i >= 0; i--)
1053                 printk("%0*lx%s", (int)(sizeof(cpu_evtchn[0])*2),
1054                        cpu_evtchn[i],
1055                        i % 8 == 0 ? "\n   " : " ");
1056
1057         printk("\nlocally unmasked:\n   ");
1058         for (i = ARRAY_SIZE(sh->evtchn_mask)-1; i >= 0; i--) {
1059                 unsigned long pending = sh->evtchn_pending[i]
1060                         & ~sh->evtchn_mask[i]
1061                         & cpu_evtchn[i];
1062                 printk("%0*lx%s", (int)(sizeof(sh->evtchn_mask[0])*2),
1063                        pending, i % 8 == 0 ? "\n   " : " ");
1064         }
1065
1066         printk("\npending list:\n");
1067         for (i = 0; i < NR_EVENT_CHANNELS; i++) {
1068                 if (sync_test_bit(i, sh->evtchn_pending)) {
1069                         int word_idx = i / BITS_PER_LONG;
1070                         printk("  %d: event %d -> irq %d%s%s%s\n",
1071                                cpu_from_evtchn(i), i,
1072                                evtchn_to_irq[i],
1073                                sync_test_bit(word_idx, &v->evtchn_pending_sel)
1074                                              ? "" : " l2-clear",
1075                                !sync_test_bit(i, sh->evtchn_mask)
1076                                              ? "" : " globally-masked",
1077                                sync_test_bit(i, cpu_evtchn)
1078                                              ? "" : " locally-masked");
1079                 }
1080         }
1081
1082         spin_unlock_irqrestore(&debug_lock, flags);
1083
1084         return IRQ_HANDLED;
1085 }
1086
1087 static DEFINE_PER_CPU(unsigned, xed_nesting_count);
1088
1089 /*
1090  * Search the CPUs pending events bitmasks.  For each one found, map
1091  * the event number to an irq, and feed it into do_IRQ() for
1092  * handling.
1093  *
1094  * Xen uses a two-level bitmap to speed searching.  The first level is
1095  * a bitset of words which contain pending event bits.  The second
1096  * level is a bitset of pending events themselves.
1097  */
1098 static void __xen_evtchn_do_upcall(void)
1099 {
1100         int cpu = get_cpu();
1101         struct shared_info *s = HYPERVISOR_shared_info;
1102         struct vcpu_info *vcpu_info = __this_cpu_read(xen_vcpu);
1103         unsigned count;
1104
1105         do {
1106                 unsigned long pending_words;
1107
1108                 vcpu_info->evtchn_upcall_pending = 0;
1109
1110                 if (__this_cpu_inc_return(xed_nesting_count) - 1)
1111                         goto out;
1112
1113 #ifndef CONFIG_X86 /* No need for a barrier -- XCHG is a barrier on x86. */
1114                 /* Clear master flag /before/ clearing selector flag. */
1115                 wmb();
1116 #endif
1117                 pending_words = xchg(&vcpu_info->evtchn_pending_sel, 0);
1118                 while (pending_words != 0) {
1119                         unsigned long pending_bits;
1120                         int word_idx = __ffs(pending_words);
1121                         pending_words &= ~(1UL << word_idx);
1122
1123                         while ((pending_bits = active_evtchns(cpu, s, word_idx)) != 0) {
1124                                 int bit_idx = __ffs(pending_bits);
1125                                 int port = (word_idx * BITS_PER_LONG) + bit_idx;
1126                                 int irq = evtchn_to_irq[port];
1127                                 struct irq_desc *desc;
1128
1129                                 mask_evtchn(port);
1130                                 clear_evtchn(port);
1131
1132                                 if (irq != -1) {
1133                                         desc = irq_to_desc(irq);
1134                                         if (desc)
1135                                                 generic_handle_irq_desc(irq, desc);
1136                                 }
1137                         }
1138                 }
1139
1140                 BUG_ON(!irqs_disabled());
1141
1142                 count = __this_cpu_read(xed_nesting_count);
1143                 __this_cpu_write(xed_nesting_count, 0);
1144         } while (count != 1 || vcpu_info->evtchn_upcall_pending);
1145
1146 out:
1147
1148         put_cpu();
1149 }
1150
1151 void xen_evtchn_do_upcall(struct pt_regs *regs)
1152 {
1153         struct pt_regs *old_regs = set_irq_regs(regs);
1154
1155         exit_idle();
1156         irq_enter();
1157
1158         __xen_evtchn_do_upcall();
1159
1160         irq_exit();
1161         set_irq_regs(old_regs);
1162 }
1163
1164 void xen_hvm_evtchn_do_upcall(void)
1165 {
1166         __xen_evtchn_do_upcall();
1167 }
1168 EXPORT_SYMBOL_GPL(xen_hvm_evtchn_do_upcall);
1169
1170 /* Rebind a new event channel to an existing irq. */
1171 void rebind_evtchn_irq(int evtchn, int irq)
1172 {
1173         struct irq_info *info = info_for_irq(irq);
1174
1175         /* Make sure the irq is masked, since the new event channel
1176            will also be masked. */
1177         disable_irq(irq);
1178
1179         spin_lock(&irq_mapping_update_lock);
1180
1181         /* After resume the irq<->evtchn mappings are all cleared out */
1182         BUG_ON(evtchn_to_irq[evtchn] != -1);
1183         /* Expect irq to have been bound before,
1184            so there should be a proper type */
1185         BUG_ON(info->type == IRQT_UNBOUND);
1186
1187         evtchn_to_irq[evtchn] = irq;
1188         irq_info[irq] = mk_evtchn_info(evtchn);
1189
1190         spin_unlock(&irq_mapping_update_lock);
1191
1192         /* new event channels are always bound to cpu 0 */
1193         irq_set_affinity(irq, cpumask_of(0));
1194
1195         /* Unmask the event channel. */
1196         enable_irq(irq);
1197 }
1198
1199 /* Rebind an evtchn so that it gets delivered to a specific cpu */
1200 static int rebind_irq_to_cpu(unsigned irq, unsigned tcpu)
1201 {
1202         struct evtchn_bind_vcpu bind_vcpu;
1203         int evtchn = evtchn_from_irq(irq);
1204
1205         /* events delivered via platform PCI interrupts are always
1206          * routed to vcpu 0 */
1207         if (!VALID_EVTCHN(evtchn) ||
1208                 (xen_hvm_domain() && !xen_have_vector_callback))
1209                 return -1;
1210
1211         /* Send future instances of this interrupt to other vcpu. */
1212         bind_vcpu.port = evtchn;
1213         bind_vcpu.vcpu = tcpu;
1214
1215         /*
1216          * If this fails, it usually just indicates that we're dealing with a
1217          * virq or IPI channel, which don't actually need to be rebound. Ignore
1218          * it, but don't do the xenlinux-level rebind in that case.
1219          */
1220         if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_vcpu, &bind_vcpu) >= 0)
1221                 bind_evtchn_to_cpu(evtchn, tcpu);
1222
1223         return 0;
1224 }
1225
1226 static int set_affinity_irq(struct irq_data *data, const struct cpumask *dest,
1227                             bool force)
1228 {
1229         unsigned tcpu = cpumask_first(dest);
1230
1231         return rebind_irq_to_cpu(data->irq, tcpu);
1232 }
1233
1234 int resend_irq_on_evtchn(unsigned int irq)
1235 {
1236         int masked, evtchn = evtchn_from_irq(irq);
1237         struct shared_info *s = HYPERVISOR_shared_info;
1238
1239         if (!VALID_EVTCHN(evtchn))
1240                 return 1;
1241
1242         masked = sync_test_and_set_bit(evtchn, s->evtchn_mask);
1243         sync_set_bit(evtchn, s->evtchn_pending);
1244         if (!masked)
1245                 unmask_evtchn(evtchn);
1246
1247         return 1;
1248 }
1249
1250 static void enable_dynirq(struct irq_data *data)
1251 {
1252         int evtchn = evtchn_from_irq(data->irq);
1253
1254         if (VALID_EVTCHN(evtchn))
1255                 unmask_evtchn(evtchn);
1256 }
1257
1258 static void disable_dynirq(struct irq_data *data)
1259 {
1260         int evtchn = evtchn_from_irq(data->irq);
1261
1262         if (VALID_EVTCHN(evtchn))
1263                 mask_evtchn(evtchn);
1264 }
1265
1266 static void ack_dynirq(struct irq_data *data)
1267 {
1268         int evtchn = evtchn_from_irq(data->irq);
1269
1270         move_masked_irq(data->irq);
1271
1272         if (VALID_EVTCHN(evtchn))
1273                 unmask_evtchn(evtchn);
1274 }
1275
1276 static int retrigger_dynirq(struct irq_data *data)
1277 {
1278         int evtchn = evtchn_from_irq(data->irq);
1279         struct shared_info *sh = HYPERVISOR_shared_info;
1280         int ret = 0;
1281
1282         if (VALID_EVTCHN(evtchn)) {
1283                 int masked;
1284
1285                 masked = sync_test_and_set_bit(evtchn, sh->evtchn_mask);
1286                 sync_set_bit(evtchn, sh->evtchn_pending);
1287                 if (!masked)
1288                         unmask_evtchn(evtchn);
1289                 ret = 1;
1290         }
1291
1292         return ret;
1293 }
1294
1295 static void restore_cpu_pirqs(void)
1296 {
1297         int pirq, rc, irq, gsi;
1298         struct physdev_map_pirq map_irq;
1299
1300         for (pirq = 0; pirq < nr_irqs; pirq++) {
1301                 irq = pirq_to_irq[pirq];
1302                 if (irq == -1)
1303                         continue;
1304
1305                 /* save/restore of PT devices doesn't work, so at this point the
1306                  * only devices present are GSI based emulated devices */
1307                 gsi = gsi_from_irq(irq);
1308                 if (!gsi)
1309                         continue;
1310
1311                 map_irq.domid = DOMID_SELF;
1312                 map_irq.type = MAP_PIRQ_TYPE_GSI;
1313                 map_irq.index = gsi;
1314                 map_irq.pirq = pirq;
1315
1316                 rc = HYPERVISOR_physdev_op(PHYSDEVOP_map_pirq, &map_irq);
1317                 if (rc) {
1318                         printk(KERN_WARNING "xen map irq failed gsi=%d irq=%d pirq=%d rc=%d\n",
1319                                         gsi, irq, pirq, rc);
1320                         irq_info[irq] = mk_unbound_info();
1321                         pirq_to_irq[pirq] = -1;
1322                         continue;
1323                 }
1324
1325                 printk(KERN_DEBUG "xen: --> irq=%d, pirq=%d\n", irq, map_irq.pirq);
1326
1327                 __startup_pirq(irq);
1328         }
1329 }
1330
1331 static void restore_cpu_virqs(unsigned int cpu)
1332 {
1333         struct evtchn_bind_virq bind_virq;
1334         int virq, irq, evtchn;
1335
1336         for (virq = 0; virq < NR_VIRQS; virq++) {
1337                 if ((irq = per_cpu(virq_to_irq, cpu)[virq]) == -1)
1338                         continue;
1339
1340                 BUG_ON(virq_from_irq(irq) != virq);
1341
1342                 /* Get a new binding from Xen. */
1343                 bind_virq.virq = virq;
1344                 bind_virq.vcpu = cpu;
1345                 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_virq,
1346                                                 &bind_virq) != 0)
1347                         BUG();
1348                 evtchn = bind_virq.port;
1349
1350                 /* Record the new mapping. */
1351                 evtchn_to_irq[evtchn] = irq;
1352                 irq_info[irq] = mk_virq_info(evtchn, virq);
1353                 bind_evtchn_to_cpu(evtchn, cpu);
1354         }
1355 }
1356
1357 static void restore_cpu_ipis(unsigned int cpu)
1358 {
1359         struct evtchn_bind_ipi bind_ipi;
1360         int ipi, irq, evtchn;
1361
1362         for (ipi = 0; ipi < XEN_NR_IPIS; ipi++) {
1363                 if ((irq = per_cpu(ipi_to_irq, cpu)[ipi]) == -1)
1364                         continue;
1365
1366                 BUG_ON(ipi_from_irq(irq) != ipi);
1367
1368                 /* Get a new binding from Xen. */
1369                 bind_ipi.vcpu = cpu;
1370                 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_ipi,
1371                                                 &bind_ipi) != 0)
1372                         BUG();
1373                 evtchn = bind_ipi.port;
1374
1375                 /* Record the new mapping. */
1376                 evtchn_to_irq[evtchn] = irq;
1377                 irq_info[irq] = mk_ipi_info(evtchn, ipi);
1378                 bind_evtchn_to_cpu(evtchn, cpu);
1379         }
1380 }
1381
1382 /* Clear an irq's pending state, in preparation for polling on it */
1383 void xen_clear_irq_pending(int irq)
1384 {
1385         int evtchn = evtchn_from_irq(irq);
1386
1387         if (VALID_EVTCHN(evtchn))
1388                 clear_evtchn(evtchn);
1389 }
1390 EXPORT_SYMBOL(xen_clear_irq_pending);
1391 void xen_set_irq_pending(int irq)
1392 {
1393         int evtchn = evtchn_from_irq(irq);
1394
1395         if (VALID_EVTCHN(evtchn))
1396                 set_evtchn(evtchn);
1397 }
1398
1399 bool xen_test_irq_pending(int irq)
1400 {
1401         int evtchn = evtchn_from_irq(irq);
1402         bool ret = false;
1403
1404         if (VALID_EVTCHN(evtchn))
1405                 ret = test_evtchn(evtchn);
1406
1407         return ret;
1408 }
1409
1410 /* Poll waiting for an irq to become pending with timeout.  In the usual case,
1411  * the irq will be disabled so it won't deliver an interrupt. */
1412 void xen_poll_irq_timeout(int irq, u64 timeout)
1413 {
1414         evtchn_port_t evtchn = evtchn_from_irq(irq);
1415
1416         if (VALID_EVTCHN(evtchn)) {
1417                 struct sched_poll poll;
1418
1419                 poll.nr_ports = 1;
1420                 poll.timeout = timeout;
1421                 set_xen_guest_handle(poll.ports, &evtchn);
1422
1423                 if (HYPERVISOR_sched_op(SCHEDOP_poll, &poll) != 0)
1424                         BUG();
1425         }
1426 }
1427 EXPORT_SYMBOL(xen_poll_irq_timeout);
1428 /* Poll waiting for an irq to become pending.  In the usual case, the
1429  * irq will be disabled so it won't deliver an interrupt. */
1430 void xen_poll_irq(int irq)
1431 {
1432         xen_poll_irq_timeout(irq, 0 /* no timeout */);
1433 }
1434
1435 void xen_irq_resume(void)
1436 {
1437         unsigned int cpu, irq, evtchn;
1438
1439         init_evtchn_cpu_bindings();
1440
1441         /* New event-channel space is not 'live' yet. */
1442         for (evtchn = 0; evtchn < NR_EVENT_CHANNELS; evtchn++)
1443                 mask_evtchn(evtchn);
1444
1445         /* No IRQ <-> event-channel mappings. */
1446         for (irq = 0; irq < nr_irqs; irq++)
1447                 irq_info[irq].evtchn = 0; /* zap event-channel binding */
1448
1449         for (evtchn = 0; evtchn < NR_EVENT_CHANNELS; evtchn++)
1450                 evtchn_to_irq[evtchn] = -1;
1451
1452         for_each_possible_cpu(cpu) {
1453                 restore_cpu_virqs(cpu);
1454                 restore_cpu_ipis(cpu);
1455         }
1456
1457         restore_cpu_pirqs();
1458 }
1459
1460 static struct irq_chip xen_dynamic_chip __read_mostly = {
1461         .name                   = "xen-dyn",
1462
1463         .irq_disable            = disable_dynirq,
1464         .irq_mask               = disable_dynirq,
1465         .irq_unmask             = enable_dynirq,
1466
1467         .irq_eoi                = ack_dynirq,
1468         .irq_set_affinity       = set_affinity_irq,
1469         .irq_retrigger          = retrigger_dynirq,
1470 };
1471
1472 static struct irq_chip xen_pirq_chip __read_mostly = {
1473         .name                   = "xen-pirq",
1474
1475         .irq_startup            = startup_pirq,
1476         .irq_shutdown           = shutdown_pirq,
1477
1478         .irq_enable             = enable_pirq,
1479         .irq_unmask             = enable_pirq,
1480
1481         .irq_disable            = disable_pirq,
1482         .irq_mask               = disable_pirq,
1483
1484         .irq_ack                = ack_pirq,
1485
1486         .irq_set_affinity       = set_affinity_irq,
1487
1488         .irq_retrigger          = retrigger_dynirq,
1489 };
1490
1491 static struct irq_chip xen_percpu_chip __read_mostly = {
1492         .name                   = "xen-percpu",
1493
1494         .irq_disable            = disable_dynirq,
1495         .irq_mask               = disable_dynirq,
1496         .irq_unmask             = enable_dynirq,
1497
1498         .irq_ack                = ack_dynirq,
1499 };
1500
1501 int xen_set_callback_via(uint64_t via)
1502 {
1503         struct xen_hvm_param a;
1504         a.domid = DOMID_SELF;
1505         a.index = HVM_PARAM_CALLBACK_IRQ;
1506         a.value = via;
1507         return HYPERVISOR_hvm_op(HVMOP_set_param, &a);
1508 }
1509 EXPORT_SYMBOL_GPL(xen_set_callback_via);
1510
1511 #ifdef CONFIG_XEN_PVHVM
1512 /* Vector callbacks are better than PCI interrupts to receive event
1513  * channel notifications because we can receive vector callbacks on any
1514  * vcpu and we don't need PCI support or APIC interactions. */
1515 void xen_callback_vector(void)
1516 {
1517         int rc;
1518         uint64_t callback_via;
1519         if (xen_have_vector_callback) {
1520                 callback_via = HVM_CALLBACK_VECTOR(XEN_HVM_EVTCHN_CALLBACK);
1521                 rc = xen_set_callback_via(callback_via);
1522                 if (rc) {
1523                         printk(KERN_ERR "Request for Xen HVM callback vector"
1524                                         " failed.\n");
1525                         xen_have_vector_callback = 0;
1526                         return;
1527                 }
1528                 printk(KERN_INFO "Xen HVM callback vector for event delivery is "
1529                                 "enabled\n");
1530                 /* in the restore case the vector has already been allocated */
1531                 if (!test_bit(XEN_HVM_EVTCHN_CALLBACK, used_vectors))
1532                         alloc_intr_gate(XEN_HVM_EVTCHN_CALLBACK, xen_hvm_callback_vector);
1533         }
1534 }
1535 #else
1536 void xen_callback_vector(void) {}
1537 #endif
1538
1539 void __init xen_init_IRQ(void)
1540 {
1541         int i;
1542
1543         cpu_evtchn_mask_p = kcalloc(nr_cpu_ids, sizeof(struct cpu_evtchn_s),
1544                                     GFP_KERNEL);
1545         irq_info = kcalloc(nr_irqs, sizeof(*irq_info), GFP_KERNEL);
1546
1547         /* We are using nr_irqs as the maximum number of pirq available but
1548          * that number is actually chosen by Xen and we don't know exactly
1549          * what it is. Be careful choosing high pirq numbers. */
1550         pirq_to_irq = kcalloc(nr_irqs, sizeof(*pirq_to_irq), GFP_KERNEL);
1551         for (i = 0; i < nr_irqs; i++)
1552                 pirq_to_irq[i] = -1;
1553
1554         evtchn_to_irq = kcalloc(NR_EVENT_CHANNELS, sizeof(*evtchn_to_irq),
1555                                     GFP_KERNEL);
1556         for (i = 0; i < NR_EVENT_CHANNELS; i++)
1557                 evtchn_to_irq[i] = -1;
1558
1559         init_evtchn_cpu_bindings();
1560
1561         /* No event channels are 'live' right now. */
1562         for (i = 0; i < NR_EVENT_CHANNELS; i++)
1563                 mask_evtchn(i);
1564
1565         if (xen_hvm_domain()) {
1566                 xen_callback_vector();
1567                 native_init_IRQ();
1568                 /* pci_xen_hvm_init must be called after native_init_IRQ so that
1569                  * __acpi_register_gsi can point at the right function */
1570                 pci_xen_hvm_init();
1571         } else {
1572                 irq_ctx_init(smp_processor_id());
1573                 if (xen_initial_domain())
1574                         xen_setup_pirqs();
1575         }
1576 }