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         return info_for_irq(irq)->evtchn;
 174 }
 175
 176 unsigned irq_from_evtchn(unsigned int evtchn)
 177 {
 178         return evtchn_to_irq[evtchn];
 179 }
 180 EXPORT_SYMBOL_GPL(irq_from_evtchn);
 181
 182 static enum ipi_vector ipi_from_irq(unsigned irq)
 183 {
 184         struct irq_info *info = info_for_irq(irq);
 185
 186         BUG_ON(info == NULL);
 187         BUG_ON(info->type != IRQT_IPI);
 188
 189         return info->u.ipi;
 190 }
 191
 192 static unsigned virq_from_irq(unsigned irq)
 193 {
 194         struct irq_info *info = info_for_irq(irq);
 195
 196         BUG_ON(info == NULL);
 197         BUG_ON(info->type != IRQT_VIRQ);
 198
 199         return info->u.virq;
 200 }
 201
 202 static unsigned pirq_from_irq(unsigned irq)
 203 {
 204         struct irq_info *info = info_for_irq(irq);
 205
 206         BUG_ON(info == NULL);
 207         BUG_ON(info->type != IRQT_PIRQ);
 208
 209         return info->u.pirq.pirq;
 210 }
 211
 212 static unsigned gsi_from_irq(unsigned irq)
 213 {
 214         struct irq_info *info = info_for_irq(irq);
 215
 216         BUG_ON(info == NULL);
 217         BUG_ON(info->type != IRQT_PIRQ);
 218
 219         return info->u.pirq.gsi;
 220 }
 221
 222 static unsigned vector_from_irq(unsigned irq)
 223 {
 224         struct irq_info *info = info_for_irq(irq);
 225
 226         BUG_ON(info == NULL);
 227         BUG_ON(info->type != IRQT_PIRQ);
 228
 229         return info->u.pirq.vector;
 230 }
 231
 232 static enum xen_irq_type type_from_irq(unsigned irq)
 233 {
 234         return info_for_irq(irq)->type;
 235 }
 236
 237 static unsigned cpu_from_irq(unsigned irq)
 238 {
 239         return info_for_irq(irq)->cpu;
 240 }
 241
 242 static unsigned int cpu_from_evtchn(unsigned int evtchn)
 243 {
 244         int irq = evtchn_to_irq[evtchn];
 245         unsigned ret = 0;
 246
 247         if (irq != -1)
 248                 ret = cpu_from_irq(irq);
 249
 250         return ret;
 251 }
 252
 253 static bool pirq_needs_eoi(unsigned irq)
 254 {
 255         struct irq_info *info = info_for_irq(irq);
 256
 257         BUG_ON(info->type != IRQT_PIRQ);
 258
 259         return info->u.pirq.flags & PIRQ_NEEDS_EOI;
 260 }
 261
 262 static inline unsigned long active_evtchns(unsigned int cpu,
 263                                            struct shared_info *sh,
 264                                            unsigned int idx)
 265 {
 266         return (sh->evtchn_pending[idx] &
 267                 cpu_evtchn_mask(cpu)[idx] &
 268                 ~sh->evtchn_mask[idx]);
 269 }
 270
 271 static void bind_evtchn_to_cpu(unsigned int chn, unsigned int cpu)
 272 {
 273         int irq = evtchn_to_irq[chn];
 274
 275         BUG_ON(irq == -1);
 276 #ifdef CONFIG_SMP
 277         cpumask_copy(irq_to_desc(irq)->affinity, cpumask_of(cpu));
 278 #endif
 279
 280         clear_bit(chn, cpu_evtchn_mask(cpu_from_irq(irq)));
 281         set_bit(chn, cpu_evtchn_mask(cpu));
 282
 283         irq_info[irq].cpu = cpu;
 284 }
 285
 286 static void init_evtchn_cpu_bindings(void)
 287 {
 288         int i;
 289 #ifdef CONFIG_SMP
 290         struct irq_desc *desc;
 291
 292         /* By default all event channels notify CPU#0. */
 293         for_each_irq_desc(i, desc) {
 294                 cpumask_copy(desc->affinity, cpumask_of(0));
 295         }
 296 #endif
 297
 298         for_each_possible_cpu(i)
 299                 memset(cpu_evtchn_mask(i),
 300                        (i == 0) ? ~0 : 0, sizeof(struct cpu_evtchn_s));
 301
 302 }
 303
 304 static inline void clear_evtchn(int port)
 305 {
 306         struct shared_info *s = HYPERVISOR_shared_info;
 307         sync_clear_bit(port, &s->evtchn_pending[0]);
 308 }
 309
 310 static inline void set_evtchn(int port)
 311 {
 312         struct shared_info *s = HYPERVISOR_shared_info;
 313         sync_set_bit(port, &s->evtchn_pending[0]);
 314 }
 315
 316 static inline int test_evtchn(int port)
 317 {
 318         struct shared_info *s = HYPERVISOR_shared_info;
 319         return sync_test_bit(port, &s->evtchn_pending[0]);
 320 }
 321
 322
 323 /**
 324  * notify_remote_via_irq - send event to remote end of event channel via irq
 325  * @irq: irq of event channel to send event to
 326  *
 327  * Unlike notify_remote_via_evtchn(), this is safe to use across
 328  * save/restore. Notifications on a broken connection are silently
 329  * dropped.
 330  */
 331 void notify_remote_via_irq(int irq)
 332 {
 333         int evtchn = evtchn_from_irq(irq);
 334
 335         if (VALID_EVTCHN(evtchn))
 336                 notify_remote_via_evtchn(evtchn);
 337 }
 338 EXPORT_SYMBOL_GPL(notify_remote_via_irq);
 339
 340 static void mask_evtchn(int port)
 341 {
 342         struct shared_info *s = HYPERVISOR_shared_info;
 343         sync_set_bit(port, &s->evtchn_mask[0]);
 344 }
 345
 346 static void unmask_evtchn(int port)
 347 {
 348         struct shared_info *s = HYPERVISOR_shared_info;
 349         unsigned int cpu = get_cpu();
 350
 351         BUG_ON(!irqs_disabled());
 352
 353         /* Slow path (hypercall) if this is a non-local port. */
 354         if (unlikely(cpu != cpu_from_evtchn(port))) {
 355                 struct evtchn_unmask unmask = { .port = port };
 356                 (void)HYPERVISOR_event_channel_op(EVTCHNOP_unmask, &unmask);
 357         } else {
 358                 struct vcpu_info *vcpu_info = __this_cpu_read(xen_vcpu);
 359
 360                 sync_clear_bit(port, &s->evtchn_mask[0]);
 361
 362                 /*
 363                  * The following is basically the equivalent of
 364                  * 'hw_resend_irq'. Just like a real IO-APIC we 'lose
 365                  * the interrupt edge' if the channel is masked.
 366                  */
 367                 if (sync_test_bit(port, &s->evtchn_pending[0]) &&
 368                     !sync_test_and_set_bit(port / BITS_PER_LONG,
 369                                            &vcpu_info->evtchn_pending_sel))
 370                         vcpu_info->evtchn_upcall_pending = 1;
 371         }
 372
 373         put_cpu();
 374 }
 375
 376 static int get_nr_hw_irqs(void)
 377 {
 378         int ret = 1;
 379
 380 #ifdef CONFIG_X86_IO_APIC
 381         ret = get_nr_irqs_gsi();
 382 #endif
 383
 384         return ret;
 385 }
 386
 387 static int find_unbound_pirq(int type)
 388 {
 389         int rc, i;
 390         struct physdev_get_free_pirq op_get_free_pirq;
 391         op_get_free_pirq.type = type;
 392
 393         rc = HYPERVISOR_physdev_op(PHYSDEVOP_get_free_pirq, &op_get_free_pirq);
 394         if (!rc)
 395                 return op_get_free_pirq.pirq;
 396
 397         for (i = 0; i < nr_irqs; i++) {
 398                 if (pirq_to_irq[i] < 0)
 399                         return i;
 400         }
 401         return -1;
 402 }
 403
 404 static int find_unbound_irq(void)
 405 {
 406         struct irq_data *data;
 407         int irq, res;
 408         int start = get_nr_hw_irqs();
 409
 410         if (start == nr_irqs)
 411                 goto no_irqs;
 412
 413         /* nr_irqs is a magic value. Must not use it.*/
 414         for (irq = nr_irqs-1; irq > start; irq--) {
 415                 data = irq_get_irq_data(irq);
 416                 /* only 0->15 have init'd desc; handle irq > 16 */
 417                 if (!data)
 418                         break;
 419                 if (data->chip == &no_irq_chip)
 420                         break;
 421                 if (data->chip != &xen_dynamic_chip)
 422                         continue;
 423                 if (irq_info[irq].type == IRQT_UNBOUND)
 424                         return irq;
 425         }
 426
 427         if (irq == start)
 428                 goto no_irqs;
 429
 430         res = irq_alloc_desc_at(irq, -1);
 431
 432         if (WARN_ON(res != irq))
 433                 return -1;
 434
 435         return irq;
 436
 437 no_irqs:
 438         panic("No available IRQ to bind to: increase nr_irqs!\n");
 439 }
 440
 441 static bool identity_mapped_irq(unsigned irq)
 442 {
 443         /* identity map all the hardware irqs */
 444         return irq < get_nr_hw_irqs();
 445 }
 446
 447 static void pirq_unmask_notify(int irq)
 448 {
 449         struct physdev_eoi eoi = { .irq = pirq_from_irq(irq) };
 450
 451         if (unlikely(pirq_needs_eoi(irq))) {
 452                 int rc = HYPERVISOR_physdev_op(PHYSDEVOP_eoi, &eoi);
 453                 WARN_ON(rc);
 454         }
 455 }
 456
 457 static void pirq_query_unmask(int irq)
 458 {
 459         struct physdev_irq_status_query irq_status;
 460         struct irq_info *info = info_for_irq(irq);
 461
 462         BUG_ON(info->type != IRQT_PIRQ);
 463
 464         irq_status.irq = pirq_from_irq(irq);
 465         if (HYPERVISOR_physdev_op(PHYSDEVOP_irq_status_query, &irq_status))
 466                 irq_status.flags = 0;
 467
 468         info->u.pirq.flags &= ~PIRQ_NEEDS_EOI;
 469         if (irq_status.flags & XENIRQSTAT_needs_eoi)
 470                 info->u.pirq.flags |= PIRQ_NEEDS_EOI;
 471 }
 472
 473 static bool probing_irq(int irq)
 474 {
 475         struct irq_desc *desc = irq_to_desc(irq);
 476
 477         return desc && desc->action == NULL;
 478 }
 479
 480 static unsigned int startup_pirq(unsigned int irq)
 481 {
 482         struct evtchn_bind_pirq bind_pirq;
 483         struct irq_info *info = info_for_irq(irq);
 484         int evtchn = evtchn_from_irq(irq);
 485         int rc;
 486
 487         BUG_ON(info->type != IRQT_PIRQ);
 488
 489         if (VALID_EVTCHN(evtchn))
 490                 goto out;
 491
 492         bind_pirq.pirq = pirq_from_irq(irq);
 493         /* NB. We are happy to share unless we are probing. */
 494         bind_pirq.flags = info->u.pirq.flags & PIRQ_SHAREABLE ?
 495                                         BIND_PIRQ__WILL_SHARE : 0;
 496         rc = HYPERVISOR_event_channel_op(EVTCHNOP_bind_pirq, &bind_pirq);
 497         if (rc != 0) {
 498                 if (!probing_irq(irq))
 499                         printk(KERN_INFO "Failed to obtain physical IRQ %d\n",
 500                                irq);
 501                 return 0;
 502         }
 503         evtchn = bind_pirq.port;
 504
 505         pirq_query_unmask(irq);
 506
 507         evtchn_to_irq[evtchn] = irq;
 508         bind_evtchn_to_cpu(evtchn, 0);
 509         info->evtchn = evtchn;
 510
 511 out:
 512         unmask_evtchn(evtchn);
 513         pirq_unmask_notify(irq);
 514
 515         return 0;
 516 }
 517
 518 static void shutdown_pirq(unsigned int irq)
 519 {
 520         struct evtchn_close close;
 521         struct irq_info *info = info_for_irq(irq);
 522         int evtchn = evtchn_from_irq(irq);
 523
 524         BUG_ON(info->type != IRQT_PIRQ);
 525
 526         if (!VALID_EVTCHN(evtchn))
 527                 return;
 528
 529         mask_evtchn(evtchn);
 530
 531         close.port = evtchn;
 532         if (HYPERVISOR_event_channel_op(EVTCHNOP_close, &close) != 0)
 533                 BUG();
 534
 535         bind_evtchn_to_cpu(evtchn, 0);
 536         evtchn_to_irq[evtchn] = -1;
 537         info->evtchn = 0;
 538 }
 539
 540 static void enable_pirq(unsigned int irq)
 541 {
 542         startup_pirq(irq);
 543 }
 544
 545 static void disable_pirq(unsigned int irq)
 546 {
 547 }
 548
 549 static void ack_pirq(unsigned int irq)
 550 {
 551         int evtchn = evtchn_from_irq(irq);
 552
 553         move_native_irq(irq);
 554
 555         if (VALID_EVTCHN(evtchn)) {
 556                 mask_evtchn(evtchn);
 557                 clear_evtchn(evtchn);
 558         }
 559 }
 560
 561 static void end_pirq(unsigned int irq)
 562 {
 563         int evtchn = evtchn_from_irq(irq);
 564         struct irq_desc *desc = irq_to_desc(irq);
 565
 566         if (WARN_ON(!desc))
 567                 return;
 568
 569         if ((desc->status & (IRQ_DISABLED|IRQ_PENDING)) ==
 570             (IRQ_DISABLED|IRQ_PENDING)) {
 571                 shutdown_pirq(irq);
 572         } else if (VALID_EVTCHN(evtchn)) {
 573                 unmask_evtchn(evtchn);
 574                 pirq_unmask_notify(irq);
 575         }
 576 }
 577
 578 static int find_irq_by_gsi(unsigned gsi)
 579 {
 580         int irq;
 581
 582         for (irq = 0; irq < nr_irqs; irq++) {
 583                 struct irq_info *info = info_for_irq(irq);
 584
 585                 if (info == NULL || info->type != IRQT_PIRQ)
 586                         continue;
 587
 588                 if (gsi_from_irq(irq) == gsi)
 589                         return irq;
 590         }
 591
 592         return -1;
 593 }
 594
 595 int xen_allocate_pirq(unsigned gsi, int shareable, char *name)
 596 {
 597         return xen_map_pirq_gsi(gsi, gsi, shareable, name);
 598 }
 599
 600 /* xen_map_pirq_gsi might allocate irqs from the top down, as a
 601  * consequence don't assume that the irq number returned has a low value
 602  * or can be used as a pirq number unless you know otherwise.
 603  *
 604  * One notable exception is when xen_map_pirq_gsi is called passing an
 605  * hardware gsi as argument, in that case the irq number returned
 606  * matches the gsi number passed as second argument.
 607  *
 608  * Note: We don't assign an event channel until the irq actually started
 609  * up.  Return an existing irq if we've already got one for the gsi.
 610  */
 611 int xen_map_pirq_gsi(unsigned pirq, unsigned gsi, int shareable, char *name)
 612 {
 613         int irq = 0;
 614         struct physdev_irq irq_op;
 615
 616         spin_lock(&irq_mapping_update_lock);
 617
 618         if ((pirq > nr_irqs) || (gsi > nr_irqs)) {
 619                 printk(KERN_WARNING "xen_map_pirq_gsi: %s %s is incorrect!\n",
 620                         pirq > nr_irqs ? "pirq" :"",
 621                         gsi > nr_irqs ? "gsi" : "");
 622                 goto out;
 623         }
 624
 625         irq = find_irq_by_gsi(gsi);
 626         if (irq != -1) {
 627                 printk(KERN_INFO "xen_map_pirq_gsi: returning irq %d for gsi %u\n",
 628                        irq, gsi);
 629                 goto out;       /* XXX need refcount? */
 630         }
 631
 632         /* If we are a PV guest, we don't have GSIs (no ACPI passed). Therefore
 633          * we are using the !xen_initial_domain() to drop in the function.*/
 634         if (identity_mapped_irq(gsi) || (!xen_initial_domain() &&
 635                                 xen_pv_domain())) {
 636                 irq = gsi;
 637                 irq_alloc_desc_at(irq, -1);
 638         } else
 639                 irq = find_unbound_irq();
 640
 641         set_irq_chip_and_handler_name(irq, &xen_pirq_chip,
 642                                       handle_level_irq, name);
 643
 644         irq_op.irq = irq;
 645         irq_op.vector = 0;
 646
 647         /* Only the privileged domain can do this. For non-priv, the pcifront
 648          * driver provides a PCI bus that does the call to do exactly
 649          * this in the priv domain. */
 650         if (xen_initial_domain() &&
 651             HYPERVISOR_physdev_op(PHYSDEVOP_alloc_irq_vector, &irq_op)) {
 652                 irq_free_desc(irq);
 653                 irq = -ENOSPC;
 654                 goto out;
 655         }
 656
 657         irq_info[irq] = mk_pirq_info(0, pirq, gsi, irq_op.vector);
 658         irq_info[irq].u.pirq.flags |= shareable ? PIRQ_SHAREABLE : 0;
 659         pirq_to_irq[pirq] = irq;
 660
 661 out:
 662         spin_unlock(&irq_mapping_update_lock);
 663
 664         return irq;
 665 }
 666
 667 #ifdef CONFIG_PCI_MSI
 668 #include <linux/msi.h>
 669 #include "../pci/msi.h"
 670
 671 void xen_allocate_pirq_msi(char *name, int *irq, int *pirq, int alloc)
 672 {
 673         spin_lock(&irq_mapping_update_lock);
 674
 675         if (alloc & XEN_ALLOC_IRQ) {
 676                 *irq = find_unbound_irq();
 677                 if (*irq == -1)
 678                         goto out;
 679         }
 680
 681         if (alloc & XEN_ALLOC_PIRQ) {
 682                 *pirq = find_unbound_pirq(MAP_PIRQ_TYPE_MSI);
 683                 if (*pirq == -1)
 684                         goto out;
 685         }
 686
 687         set_irq_chip_and_handler_name(*irq, &xen_pirq_chip,
 688                                       handle_level_irq, name);
 689
 690         irq_info[*irq] = mk_pirq_info(0, *pirq, 0, 0);
 691         pirq_to_irq[*pirq] = *irq;
 692
 693 out:
 694         spin_unlock(&irq_mapping_update_lock);
 695 }
 696
 697 int xen_create_msi_irq(struct pci_dev *dev, struct msi_desc *msidesc, int type)
 698 {
 699         int irq = -1;
 700         struct physdev_map_pirq map_irq;
 701         int rc;
 702         int pos;
 703         u32 table_offset, bir;
 704
 705         memset(&map_irq, 0, sizeof(map_irq));
 706         map_irq.domid = DOMID_SELF;
 707         map_irq.type = MAP_PIRQ_TYPE_MSI;
 708         map_irq.index = -1;
 709         map_irq.pirq = -1;
 710         map_irq.bus = dev->bus->number;
 711         map_irq.devfn = dev->devfn;
 712
 713         if (type == PCI_CAP_ID_MSIX) {
 714                 pos = pci_find_capability(dev, PCI_CAP_ID_MSIX);
 715
 716                 pci_read_config_dword(dev, msix_table_offset_reg(pos),
 717                                         &table_offset);
 718                 bir = (u8)(table_offset & PCI_MSIX_FLAGS_BIRMASK);
 719
 720                 map_irq.table_base = pci_resource_start(dev, bir);
 721                 map_irq.entry_nr = msidesc->msi_attrib.entry_nr;
 722         }
 723
 724         spin_lock(&irq_mapping_update_lock);
 725
 726         irq = find_unbound_irq();
 727
 728         if (irq == -1)
 729                 goto out;
 730
 731         rc = HYPERVISOR_physdev_op(PHYSDEVOP_map_pirq, &map_irq);
 732         if (rc) {
 733                 printk(KERN_WARNING "xen map irq failed %d\n", rc);
 734
 735                 irq_free_desc(irq);
 736
 737                 irq = -1;
 738                 goto out;
 739         }
 740         irq_info[irq] = mk_pirq_info(0, map_irq.pirq, 0, map_irq.index);
 741
 742         set_irq_chip_and_handler_name(irq, &xen_pirq_chip,
 743                         handle_level_irq,
 744                         (type == PCI_CAP_ID_MSIX) ? "msi-x":"msi");
 745
 746 out:
 747         spin_unlock(&irq_mapping_update_lock);
 748         return irq;
 749 }
 750 #endif
 751
 752 int xen_destroy_irq(int irq)
 753 {
 754         struct irq_desc *desc;
 755         struct physdev_unmap_pirq unmap_irq;
 756         struct irq_info *info = info_for_irq(irq);
 757         int rc = -ENOENT;
 758
 759         spin_lock(&irq_mapping_update_lock);
 760
 761         desc = irq_to_desc(irq);
 762         if (!desc)
 763                 goto out;
 764
 765         if (xen_initial_domain()) {
 766                 unmap_irq.pirq = info->u.pirq.pirq;
 767                 unmap_irq.domid = DOMID_SELF;
 768                 rc = HYPERVISOR_physdev_op(PHYSDEVOP_unmap_pirq, &unmap_irq);
 769                 if (rc) {
 770                         printk(KERN_WARNING "unmap irq failed %d\n", rc);
 771                         goto out;
 772                 }
 773                 pirq_to_irq[info->u.pirq.pirq] = -1;
 774         }
 775         irq_info[irq] = mk_unbound_info();
 776
 777         irq_free_desc(irq);
 778
 779 out:
 780         spin_unlock(&irq_mapping_update_lock);
 781         return rc;
 782 }
 783
 784 int xen_vector_from_irq(unsigned irq)
 785 {
 786         return vector_from_irq(irq);
 787 }
 788
 789 int xen_gsi_from_irq(unsigned irq)
 790 {
 791         return gsi_from_irq(irq);
 792 }
 793
 794 int xen_irq_from_pirq(unsigned pirq)
 795 {
 796         return pirq_to_irq[pirq];
 797 }
 798
 799 int bind_evtchn_to_irq(unsigned int evtchn)
 800 {
 801         int irq;
 802
 803         spin_lock(&irq_mapping_update_lock);
 804
 805         irq = evtchn_to_irq[evtchn];
 806
 807         if (irq == -1) {
 808                 irq = find_unbound_irq();
 809
 810                 set_irq_chip_and_handler_name(irq, &xen_dynamic_chip,
 811                                               handle_fasteoi_irq, "event");
 812
 813                 evtchn_to_irq[evtchn] = irq;
 814                 irq_info[irq] = mk_evtchn_info(evtchn);
 815         }
 816
 817         spin_unlock(&irq_mapping_update_lock);
 818
 819         return irq;
 820 }
 821 EXPORT_SYMBOL_GPL(bind_evtchn_to_irq);
 822
 823 static int bind_ipi_to_irq(unsigned int ipi, unsigned int cpu)
 824 {
 825         struct evtchn_bind_ipi bind_ipi;
 826         int evtchn, irq;
 827
 828         spin_lock(&irq_mapping_update_lock);
 829
 830         irq = per_cpu(ipi_to_irq, cpu)[ipi];
 831
 832         if (irq == -1) {
 833                 irq = find_unbound_irq();
 834                 if (irq < 0)
 835                         goto out;
 836
 837                 set_irq_chip_and_handler_name(irq, &xen_percpu_chip,
 838                                               handle_percpu_irq, "ipi");
 839
 840                 bind_ipi.vcpu = cpu;
 841                 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_ipi,
 842                                                 &bind_ipi) != 0)
 843                         BUG();
 844                 evtchn = bind_ipi.port;
 845
 846                 evtchn_to_irq[evtchn] = irq;
 847                 irq_info[irq] = mk_ipi_info(evtchn, ipi);
 848                 per_cpu(ipi_to_irq, cpu)[ipi] = irq;
 849
 850                 bind_evtchn_to_cpu(evtchn, cpu);
 851         }
 852
 853  out:
 854         spin_unlock(&irq_mapping_update_lock);
 855         return irq;
 856 }
 857
 858
 859 int bind_virq_to_irq(unsigned int virq, unsigned int cpu)
 860 {
 861         struct evtchn_bind_virq bind_virq;
 862         int evtchn, irq;
 863
 864         spin_lock(&irq_mapping_update_lock);
 865
 866         irq = per_cpu(virq_to_irq, cpu)[virq];
 867
 868         if (irq == -1) {
 869                 irq = find_unbound_irq();
 870
 871                 set_irq_chip_and_handler_name(irq, &xen_percpu_chip,
 872                                               handle_percpu_irq, "virq");
 873
 874                 bind_virq.virq = virq;
 875                 bind_virq.vcpu = cpu;
 876                 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_virq,
 877                                                 &bind_virq) != 0)
 878                         BUG();
 879                 evtchn = bind_virq.port;
 880
 881                 evtchn_to_irq[evtchn] = irq;
 882                 irq_info[irq] = mk_virq_info(evtchn, virq);
 883
 884                 per_cpu(virq_to_irq, cpu)[virq] = irq;
 885
 886                 bind_evtchn_to_cpu(evtchn, cpu);
 887         }
 888
 889         spin_unlock(&irq_mapping_update_lock);
 890
 891         return irq;
 892 }
 893
 894 static void unbind_from_irq(unsigned int irq)
 895 {
 896         struct evtchn_close close;
 897         int evtchn = evtchn_from_irq(irq);
 898
 899         spin_lock(&irq_mapping_update_lock);
 900
 901         if (VALID_EVTCHN(evtchn)) {
 902                 close.port = evtchn;
 903                 if (HYPERVISOR_event_channel_op(EVTCHNOP_close, &close) != 0)
 904                         BUG();
 905
 906                 switch (type_from_irq(irq)) {
 907                 case IRQT_VIRQ:
 908                         per_cpu(virq_to_irq, cpu_from_evtchn(evtchn))
 909                                 [virq_from_irq(irq)] = -1;
 910                         break;
 911                 case IRQT_IPI:
 912                         per_cpu(ipi_to_irq, cpu_from_evtchn(evtchn))
 913                                 [ipi_from_irq(irq)] = -1;
 914                         break;
 915                 default:
 916                         break;
 917                 }
 918
 919                 /* Closed ports are implicitly re-bound to VCPU0. */
 920                 bind_evtchn_to_cpu(evtchn, 0);
 921
 922                 evtchn_to_irq[evtchn] = -1;
 923         }
 924
 925         if (irq_info[irq].type != IRQT_UNBOUND) {
 926                 irq_info[irq] = mk_unbound_info();
 927
 928                 irq_free_desc(irq);
 929         }
 930
 931         spin_unlock(&irq_mapping_update_lock);
 932 }
 933
 934 int bind_evtchn_to_irqhandler(unsigned int evtchn,
 935                               irq_handler_t handler,
 936                               unsigned long irqflags,
 937                               const char *devname, void *dev_id)
 938 {
 939         unsigned int irq;
 940         int retval;
 941
 942         irq = bind_evtchn_to_irq(evtchn);
 943         retval = request_irq(irq, handler, irqflags, devname, dev_id);
 944         if (retval != 0) {
 945                 unbind_from_irq(irq);
 946                 return retval;
 947         }
 948
 949         return irq;
 950 }
 951 EXPORT_SYMBOL_GPL(bind_evtchn_to_irqhandler);
 952
 953 int bind_virq_to_irqhandler(unsigned int virq, unsigned int cpu,
 954                             irq_handler_t handler,
 955                             unsigned long irqflags, const char *devname, void *dev_id)
 956 {
 957         unsigned int irq;
 958         int retval;
 959
 960         irq = bind_virq_to_irq(virq, cpu);
 961         retval = request_irq(irq, handler, irqflags, devname, dev_id);
 962         if (retval != 0) {
 963                 unbind_from_irq(irq);
 964                 return retval;
 965         }
 966
 967         return irq;
 968 }
 969 EXPORT_SYMBOL_GPL(bind_virq_to_irqhandler);
 970
 971 int bind_ipi_to_irqhandler(enum ipi_vector ipi,
 972                            unsigned int cpu,
 973                            irq_handler_t handler,
 974                            unsigned long irqflags,
 975                            const char *devname,
 976                            void *dev_id)
 977 {
 978         int irq, retval;
 979
 980         irq = bind_ipi_to_irq(ipi, cpu);
 981         if (irq < 0)
 982                 return irq;
 983
 984         irqflags |= IRQF_NO_SUSPEND;
 985         retval = request_irq(irq, handler, irqflags, devname, dev_id);
 986         if (retval != 0) {
 987                 unbind_from_irq(irq);
 988                 return retval;
 989         }
 990
 991         return irq;
 992 }
 993
 994 void unbind_from_irqhandler(unsigned int irq, void *dev_id)
 995 {
 996         free_irq(irq, dev_id);
 997         unbind_from_irq(irq);
 998 }
 999 EXPORT_SYMBOL_GPL(unbind_from_irqhandler);
1000
1001 void xen_send_IPI_one(unsigned int cpu, enum ipi_vector vector)
1002 {
1003         int irq = per_cpu(ipi_to_irq, cpu)[vector];
1004         BUG_ON(irq < 0);
1005         notify_remote_via_irq(irq);
1006 }
1007
1008 irqreturn_t xen_debug_interrupt(int irq, void *dev_id)
1009 {
1010         struct shared_info *sh = HYPERVISOR_shared_info;
1011         int cpu = smp_processor_id();
1012         unsigned long *cpu_evtchn = cpu_evtchn_mask(cpu);
1013         int i;
1014         unsigned long flags;
1015         static DEFINE_SPINLOCK(debug_lock);
1016         struct vcpu_info *v;
1017
1018         spin_lock_irqsave(&debug_lock, flags);
1019
1020         printk("\nvcpu %d\n  ", cpu);
1021
1022         for_each_online_cpu(i) {
1023                 int pending;
1024                 v = per_cpu(xen_vcpu, i);
1025                 pending = (get_irq_regs() && i == cpu)
1026                         ? xen_irqs_disabled(get_irq_regs())
1027                         : v->evtchn_upcall_mask;
1028                 printk("%d: masked=%d pending=%d event_sel %0*lx\n  ", i,
1029                        pending, v->evtchn_upcall_pending,
1030                        (int)(sizeof(v->evtchn_pending_sel)*2),
1031                        v->evtchn_pending_sel);
1032         }
1033         v = per_cpu(xen_vcpu, cpu);
1034
1035         printk("\npending:\n   ");
1036         for (i = ARRAY_SIZE(sh->evtchn_pending)-1; i >= 0; i--)
1037                 printk("%0*lx%s", (int)sizeof(sh->evtchn_pending[0])*2,
1038                        sh->evtchn_pending[i],
1039                        i % 8 == 0 ? "\n   " : " ");
1040         printk("\nglobal mask:\n   ");
1041         for (i = ARRAY_SIZE(sh->evtchn_mask)-1; i >= 0; i--)
1042                 printk("%0*lx%s",
1043                        (int)(sizeof(sh->evtchn_mask[0])*2),
1044                        sh->evtchn_mask[i],
1045                        i % 8 == 0 ? "\n   " : " ");
1046
1047         printk("\nglobally unmasked:\n   ");
1048         for (i = ARRAY_SIZE(sh->evtchn_mask)-1; i >= 0; i--)
1049                 printk("%0*lx%s", (int)(sizeof(sh->evtchn_mask[0])*2),
1050                        sh->evtchn_pending[i] & ~sh->evtchn_mask[i],
1051                        i % 8 == 0 ? "\n   " : " ");
1052
1053         printk("\nlocal cpu%d mask:\n   ", cpu);
1054         for (i = (NR_EVENT_CHANNELS/BITS_PER_LONG)-1; i >= 0; i--)
1055                 printk("%0*lx%s", (int)(sizeof(cpu_evtchn[0])*2),
1056                        cpu_evtchn[i],
1057                        i % 8 == 0 ? "\n   " : " ");
1058
1059         printk("\nlocally unmasked:\n   ");
1060         for (i = ARRAY_SIZE(sh->evtchn_mask)-1; i >= 0; i--) {
1061                 unsigned long pending = sh->evtchn_pending[i]
1062                         & ~sh->evtchn_mask[i]
1063                         & cpu_evtchn[i];
1064                 printk("%0*lx%s", (int)(sizeof(sh->evtchn_mask[0])*2),
1065                        pending, i % 8 == 0 ? "\n   " : " ");
1066         }
1067
1068         printk("\npending list:\n");
1069         for (i = 0; i < NR_EVENT_CHANNELS; i++) {
1070                 if (sync_test_bit(i, sh->evtchn_pending)) {
1071                         int word_idx = i / BITS_PER_LONG;
1072                         printk("  %d: event %d -> irq %d%s%s%s\n",
1073                                cpu_from_evtchn(i), i,
1074                                evtchn_to_irq[i],
1075                                sync_test_bit(word_idx, &v->evtchn_pending_sel)
1076                                              ? "" : " l2-clear",
1077                                !sync_test_bit(i, sh->evtchn_mask)
1078                                              ? "" : " globally-masked",
1079                                sync_test_bit(i, cpu_evtchn)
1080                                              ? "" : " locally-masked");
1081                 }
1082         }
1083
1084         spin_unlock_irqrestore(&debug_lock, flags);
1085
1086         return IRQ_HANDLED;
1087 }
1088
1089 static DEFINE_PER_CPU(unsigned, xed_nesting_count);
1090
1091 /*
1092  * Search the CPUs pending events bitmasks.  For each one found, map
1093  * the event number to an irq, and feed it into do_IRQ() for
1094  * handling.
1095  *
1096  * Xen uses a two-level bitmap to speed searching.  The first level is
1097  * a bitset of words which contain pending event bits.  The second
1098  * level is a bitset of pending events themselves.
1099  */
1100 static void __xen_evtchn_do_upcall(void)
1101 {
1102         int cpu = get_cpu();
1103         struct shared_info *s = HYPERVISOR_shared_info;
1104         struct vcpu_info *vcpu_info = __this_cpu_read(xen_vcpu);
1105         unsigned count;
1106
1107         do {
1108                 unsigned long pending_words;
1109
1110                 vcpu_info->evtchn_upcall_pending = 0;
1111
1112                 if (__this_cpu_inc_return(xed_nesting_count) - 1)
1113                         goto out;
1114
1115 #ifndef CONFIG_X86 /* No need for a barrier -- XCHG is a barrier on x86. */
1116                 /* Clear master flag /before/ clearing selector flag. */
1117                 wmb();
1118 #endif
1119                 pending_words = xchg(&vcpu_info->evtchn_pending_sel, 0);
1120                 while (pending_words != 0) {
1121                         unsigned long pending_bits;
1122                         int word_idx = __ffs(pending_words);
1123                         pending_words &= ~(1UL << word_idx);
1124
1125                         while ((pending_bits = active_evtchns(cpu, s, word_idx)) != 0) {
1126                                 int bit_idx = __ffs(pending_bits);
1127                                 int port = (word_idx * BITS_PER_LONG) + bit_idx;
1128                                 int irq = evtchn_to_irq[port];
1129                                 struct irq_desc *desc;
1130
1131                                 mask_evtchn(port);
1132                                 clear_evtchn(port);
1133
1134                                 if (irq != -1) {
1135                                         desc = irq_to_desc(irq);
1136                                         if (desc)
1137                                                 generic_handle_irq_desc(irq, desc);
1138                                 }
1139                         }
1140                 }
1141
1142                 BUG_ON(!irqs_disabled());
1143
1144                 count = __this_cpu_read(xed_nesting_count);
1145                 __this_cpu_write(xed_nesting_count, 0);
1146         } while (count != 1 || vcpu_info->evtchn_upcall_pending);
1147
1148 out:
1149
1150         put_cpu();
1151 }
1152
1153 void xen_evtchn_do_upcall(struct pt_regs *regs)
1154 {
1155         struct pt_regs *old_regs = set_irq_regs(regs);
1156
1157         exit_idle();
1158         irq_enter();
1159
1160         __xen_evtchn_do_upcall();
1161
1162         irq_exit();
1163         set_irq_regs(old_regs);
1164 }
1165
1166 void xen_hvm_evtchn_do_upcall(void)
1167 {
1168         __xen_evtchn_do_upcall();
1169 }
1170 EXPORT_SYMBOL_GPL(xen_hvm_evtchn_do_upcall);
1171
1172 /* Rebind a new event channel to an existing irq. */
1173 void rebind_evtchn_irq(int evtchn, int irq)
1174 {
1175         struct irq_info *info = info_for_irq(irq);
1176
1177         /* Make sure the irq is masked, since the new event channel
1178            will also be masked. */
1179         disable_irq(irq);
1180
1181         spin_lock(&irq_mapping_update_lock);
1182
1183         /* After resume the irq<->evtchn mappings are all cleared out */
1184         BUG_ON(evtchn_to_irq[evtchn] != -1);
1185         /* Expect irq to have been bound before,
1186            so there should be a proper type */
1187         BUG_ON(info->type == IRQT_UNBOUND);
1188
1189         evtchn_to_irq[evtchn] = irq;
1190         irq_info[irq] = mk_evtchn_info(evtchn);
1191
1192         spin_unlock(&irq_mapping_update_lock);
1193
1194         /* new event channels are always bound to cpu 0 */
1195         irq_set_affinity(irq, cpumask_of(0));
1196
1197         /* Unmask the event channel. */
1198         enable_irq(irq);
1199 }
1200
1201 /* Rebind an evtchn so that it gets delivered to a specific cpu */
1202 static int rebind_irq_to_cpu(unsigned irq, unsigned tcpu)
1203 {
1204         struct evtchn_bind_vcpu bind_vcpu;
1205         int evtchn = evtchn_from_irq(irq);
1206
1207         /* events delivered via platform PCI interrupts are always
1208          * routed to vcpu 0 */
1209         if (!VALID_EVTCHN(evtchn) ||
1210                 (xen_hvm_domain() && !xen_have_vector_callback))
1211                 return -1;
1212
1213         /* Send future instances of this interrupt to other vcpu. */
1214         bind_vcpu.port = evtchn;
1215         bind_vcpu.vcpu = tcpu;
1216
1217         /*
1218          * If this fails, it usually just indicates that we're dealing with a
1219          * virq or IPI channel, which don't actually need to be rebound. Ignore
1220          * it, but don't do the xenlinux-level rebind in that case.
1221          */
1222         if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_vcpu, &bind_vcpu) >= 0)
1223                 bind_evtchn_to_cpu(evtchn, tcpu);
1224
1225         return 0;
1226 }
1227
1228 static int set_affinity_irq(unsigned irq, const struct cpumask *dest)
1229 {
1230         unsigned tcpu = cpumask_first(dest);
1231
1232         return rebind_irq_to_cpu(irq, tcpu);
1233 }
1234
1235 int resend_irq_on_evtchn(unsigned int irq)
1236 {
1237         int masked, evtchn = evtchn_from_irq(irq);
1238         struct shared_info *s = HYPERVISOR_shared_info;
1239
1240         if (!VALID_EVTCHN(evtchn))
1241                 return 1;
1242
1243         masked = sync_test_and_set_bit(evtchn, s->evtchn_mask);
1244         sync_set_bit(evtchn, s->evtchn_pending);
1245         if (!masked)
1246                 unmask_evtchn(evtchn);
1247
1248         return 1;
1249 }
1250
1251 static void enable_dynirq(unsigned int irq)
1252 {
1253         int evtchn = evtchn_from_irq(irq);
1254
1255         if (VALID_EVTCHN(evtchn))
1256                 unmask_evtchn(evtchn);
1257 }
1258
1259 static void disable_dynirq(unsigned int irq)
1260 {
1261         int evtchn = evtchn_from_irq(irq);
1262
1263         if (VALID_EVTCHN(evtchn))
1264                 mask_evtchn(evtchn);
1265 }
1266
1267 static void ack_dynirq(unsigned int irq)
1268 {
1269         int evtchn = evtchn_from_irq(irq);
1270
1271         move_masked_irq(irq);
1272
1273         if (VALID_EVTCHN(evtchn))
1274                 unmask_evtchn(evtchn);
1275 }
1276
1277 static int retrigger_dynirq(unsigned int irq)
1278 {
1279         int evtchn = evtchn_from_irq(irq);
1280         struct shared_info *sh = HYPERVISOR_shared_info;
1281         int ret = 0;
1282
1283         if (VALID_EVTCHN(evtchn)) {
1284                 int masked;
1285
1286                 masked = sync_test_and_set_bit(evtchn, sh->evtchn_mask);
1287                 sync_set_bit(evtchn, sh->evtchn_pending);
1288                 if (!masked)
1289                         unmask_evtchn(evtchn);
1290                 ret = 1;
1291         }
1292
1293         return ret;
1294 }
1295
1296 static void restore_cpu_pirqs(void)
1297 {
1298         int pirq, rc, irq, gsi;
1299         struct physdev_map_pirq map_irq;
1300
1301         for (pirq = 0; pirq < nr_irqs; pirq++) {
1302                 irq = pirq_to_irq[pirq];
1303                 if (irq == -1)
1304                         continue;
1305
1306                 /* save/restore of PT devices doesn't work, so at this point the
1307                  * only devices present are GSI based emulated devices */
1308                 gsi = gsi_from_irq(irq);
1309                 if (!gsi)
1310                         continue;
1311
1312                 map_irq.domid = DOMID_SELF;
1313                 map_irq.type = MAP_PIRQ_TYPE_GSI;
1314                 map_irq.index = gsi;
1315                 map_irq.pirq = pirq;
1316
1317                 rc = HYPERVISOR_physdev_op(PHYSDEVOP_map_pirq, &map_irq);
1318                 if (rc) {
1319                         printk(KERN_WARNING "xen map irq failed gsi=%d irq=%d pirq=%d rc=%d\n",
1320                                         gsi, irq, pirq, rc);
1321                         irq_info[irq] = mk_unbound_info();
1322                         pirq_to_irq[pirq] = -1;
1323                         continue;
1324                 }
1325
1326                 printk(KERN_DEBUG "xen: --> irq=%d, pirq=%d\n", irq, map_irq.pirq);
1327
1328                 startup_pirq(irq);
1329         }
1330 }
1331
1332 static void restore_cpu_virqs(unsigned int cpu)
1333 {
1334         struct evtchn_bind_virq bind_virq;
1335         int virq, irq, evtchn;
1336
1337         for (virq = 0; virq < NR_VIRQS; virq++) {
1338                 if ((irq = per_cpu(virq_to_irq, cpu)[virq]) == -1)
1339                         continue;
1340
1341                 BUG_ON(virq_from_irq(irq) != virq);
1342
1343                 /* Get a new binding from Xen. */
1344                 bind_virq.virq = virq;
1345                 bind_virq.vcpu = cpu;
1346                 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_virq,
1347                                                 &bind_virq) != 0)
1348                         BUG();
1349                 evtchn = bind_virq.port;
1350
1351                 /* Record the new mapping. */
1352                 evtchn_to_irq[evtchn] = irq;
1353                 irq_info[irq] = mk_virq_info(evtchn, virq);
1354                 bind_evtchn_to_cpu(evtchn, cpu);
1355         }
1356 }
1357
1358 static void restore_cpu_ipis(unsigned int cpu)
1359 {
1360         struct evtchn_bind_ipi bind_ipi;
1361         int ipi, irq, evtchn;
1362
1363         for (ipi = 0; ipi < XEN_NR_IPIS; ipi++) {
1364                 if ((irq = per_cpu(ipi_to_irq, cpu)[ipi]) == -1)
1365                         continue;
1366
1367                 BUG_ON(ipi_from_irq(irq) != ipi);
1368
1369                 /* Get a new binding from Xen. */
1370                 bind_ipi.vcpu = cpu;
1371                 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_ipi,
1372                                                 &bind_ipi) != 0)
1373                         BUG();
1374                 evtchn = bind_ipi.port;
1375
1376                 /* Record the new mapping. */
1377                 evtchn_to_irq[evtchn] = irq;
1378                 irq_info[irq] = mk_ipi_info(evtchn, ipi);
1379                 bind_evtchn_to_cpu(evtchn, cpu);
1380         }
1381 }
1382
1383 /* Clear an irq's pending state, in preparation for polling on it */
1384 void xen_clear_irq_pending(int irq)
1385 {
1386         int evtchn = evtchn_from_irq(irq);
1387
1388         if (VALID_EVTCHN(evtchn))
1389                 clear_evtchn(evtchn);
1390 }
1391 EXPORT_SYMBOL(xen_clear_irq_pending);
1392 void xen_set_irq_pending(int irq)
1393 {
1394         int evtchn = evtchn_from_irq(irq);
1395
1396         if (VALID_EVTCHN(evtchn))
1397                 set_evtchn(evtchn);
1398 }
1399
1400 bool xen_test_irq_pending(int irq)
1401 {
1402         int evtchn = evtchn_from_irq(irq);
1403         bool ret = false;
1404
1405         if (VALID_EVTCHN(evtchn))
1406                 ret = test_evtchn(evtchn);
1407
1408         return ret;
1409 }
1410
1411 /* Poll waiting for an irq to become pending with timeout.  In the usual case,
1412  * the irq will be disabled so it won't deliver an interrupt. */
1413 void xen_poll_irq_timeout(int irq, u64 timeout)
1414 {
1415         evtchn_port_t evtchn = evtchn_from_irq(irq);
1416
1417         if (VALID_EVTCHN(evtchn)) {
1418                 struct sched_poll poll;
1419
1420                 poll.nr_ports = 1;
1421                 poll.timeout = timeout;
1422                 set_xen_guest_handle(poll.ports, &evtchn);
1423
1424                 if (HYPERVISOR_sched_op(SCHEDOP_poll, &poll) != 0)
1425                         BUG();
1426         }
1427 }
1428 EXPORT_SYMBOL(xen_poll_irq_timeout);
1429 /* Poll waiting for an irq to become pending.  In the usual case, the
1430  * irq will be disabled so it won't deliver an interrupt. */
1431 void xen_poll_irq(int irq)
1432 {
1433         xen_poll_irq_timeout(irq, 0 /* no timeout */);
1434 }
1435
1436 void xen_irq_resume(void)
1437 {
1438         unsigned int cpu, irq, evtchn;
1439         struct irq_desc *desc;
1440
1441         init_evtchn_cpu_bindings();
1442
1443         /* New event-channel space is not 'live' yet. */
1444         for (evtchn = 0; evtchn < NR_EVENT_CHANNELS; evtchn++)
1445                 mask_evtchn(evtchn);
1446
1447         /* No IRQ <-> event-channel mappings. */
1448         for (irq = 0; irq < nr_irqs; irq++)
1449                 irq_info[irq].evtchn = 0; /* zap event-channel binding */
1450
1451         for (evtchn = 0; evtchn < NR_EVENT_CHANNELS; evtchn++)
1452                 evtchn_to_irq[evtchn] = -1;
1453
1454         for_each_possible_cpu(cpu) {
1455                 restore_cpu_virqs(cpu);
1456                 restore_cpu_ipis(cpu);
1457         }
1458
1459         /*
1460          * Unmask any IRQF_NO_SUSPEND IRQs which are enabled. These
1461          * are not handled by the IRQ core.
1462          */
1463         for_each_irq_desc(irq, desc) {
1464                 if (!desc->action || !(desc->action->flags & IRQF_NO_SUSPEND))
1465                         continue;
1466                 if (desc->status & IRQ_DISABLED)
1467                         continue;
1468
1469                 evtchn = evtchn_from_irq(irq);
1470                 if (evtchn == -1)
1471                         continue;
1472
1473                 unmask_evtchn(evtchn);
1474         }
1475
1476         restore_cpu_pirqs();
1477 }
1478
1479 static struct irq_chip xen_dynamic_chip __read_mostly = {
1480         .name           = "xen-dyn",
1481
1482         .disable        = disable_dynirq,
1483         .mask           = disable_dynirq,
1484         .unmask         = enable_dynirq,
1485
1486         .eoi            = ack_dynirq,
1487         .set_affinity   = set_affinity_irq,
1488         .retrigger      = retrigger_dynirq,
1489 };
1490
1491 static struct irq_chip xen_pirq_chip __read_mostly = {
1492         .name           = "xen-pirq",
1493
1494         .startup        = startup_pirq,
1495         .shutdown       = shutdown_pirq,
1496
1497         .enable         = enable_pirq,
1498         .unmask         = enable_pirq,
1499
1500         .disable        = disable_pirq,
1501         .mask           = disable_pirq,
1502
1503         .ack            = ack_pirq,
1504         .end            = end_pirq,
1505
1506         .set_affinity   = set_affinity_irq,
1507
1508         .retrigger      = retrigger_dynirq,
1509 };
1510
1511 static struct irq_chip xen_percpu_chip __read_mostly = {
1512         .name           = "xen-percpu",
1513
1514         .disable        = disable_dynirq,
1515         .mask           = disable_dynirq,
1516         .unmask         = enable_dynirq,
1517
1518         .ack            = ack_dynirq,
1519 };
1520
1521 int xen_set_callback_via(uint64_t via)
1522 {
1523         struct xen_hvm_param a;
1524         a.domid = DOMID_SELF;
1525         a.index = HVM_PARAM_CALLBACK_IRQ;
1526         a.value = via;
1527         return HYPERVISOR_hvm_op(HVMOP_set_param, &a);
1528 }
1529 EXPORT_SYMBOL_GPL(xen_set_callback_via);
1530
1531 #ifdef CONFIG_XEN_PVHVM
1532 /* Vector callbacks are better than PCI interrupts to receive event
1533  * channel notifications because we can receive vector callbacks on any
1534  * vcpu and we don't need PCI support or APIC interactions. */
1535 void xen_callback_vector(void)
1536 {
1537         int rc;
1538         uint64_t callback_via;
1539         if (xen_have_vector_callback) {
1540                 callback_via = HVM_CALLBACK_VECTOR(XEN_HVM_EVTCHN_CALLBACK);
1541                 rc = xen_set_callback_via(callback_via);
1542                 if (rc) {
1543                         printk(KERN_ERR "Request for Xen HVM callback vector"
1544                                         " failed.\n");
1545                         xen_have_vector_callback = 0;
1546                         return;
1547                 }
1548                 printk(KERN_INFO "Xen HVM callback vector for event delivery is "
1549                                 "enabled\n");
1550                 /* in the restore case the vector has already been allocated */
1551                 if (!test_bit(XEN_HVM_EVTCHN_CALLBACK, used_vectors))
1552                         alloc_intr_gate(XEN_HVM_EVTCHN_CALLBACK, xen_hvm_callback_vector);
1553         }
1554 }
1555 #else
1556 void xen_callback_vector(void) {}
1557 #endif
1558
1559 void __init xen_init_IRQ(void)
1560 {
1561         int i;
1562
1563         cpu_evtchn_mask_p = kcalloc(nr_cpu_ids, sizeof(struct cpu_evtchn_s),
1564                                     GFP_KERNEL);
1565         irq_info = kcalloc(nr_irqs, sizeof(*irq_info), GFP_KERNEL);
1566
1567         /* We are using nr_irqs as the maximum number of pirq available but
1568          * that number is actually chosen by Xen and we don't know exactly
1569          * what it is. Be careful choosing high pirq numbers. */
1570         pirq_to_irq = kcalloc(nr_irqs, sizeof(*pirq_to_irq), GFP_KERNEL);
1571         for (i = 0; i < nr_irqs; i++)
1572                 pirq_to_irq[i] = -1;
1573
1574         evtchn_to_irq = kcalloc(NR_EVENT_CHANNELS, sizeof(*evtchn_to_irq),
1575                                     GFP_KERNEL);
1576         for (i = 0; i < NR_EVENT_CHANNELS; i++)
1577                 evtchn_to_irq[i] = -1;
1578
1579         init_evtchn_cpu_bindings();
1580
1581         /* No event channels are 'live' right now. */
1582         for (i = 0; i < NR_EVENT_CHANNELS; i++)
1583                 mask_evtchn(i);
1584
1585         if (xen_hvm_domain()) {
1586                 xen_callback_vector();
1587                 native_init_IRQ();
1588                 /* pci_xen_hvm_init must be called after native_init_IRQ so that
1589                  * __acpi_register_gsi can point at the right function */
1590                 pci_xen_hvm_init();
1591         } else {
1592                 irq_ctx_init(smp_processor_id());
1593                 if (xen_initial_domain())
1594                         xen_setup_pirqs();
1595         }
1596 }