Pileus Git - ~andy/linux/blob - drivers/xen/events/events_base.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 received, 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 #define pr_fmt(fmt) "xen:" KBUILD_MODNAME ": " fmt
  25
  26 #include <linux/linkage.h>
  27 #include <linux/interrupt.h>
  28 #include <linux/irq.h>
  29 #include <linux/module.h>
  30 #include <linux/string.h>
  31 #include <linux/bootmem.h>
  32 #include <linux/slab.h>
  33 #include <linux/irqnr.h>
  34 #include <linux/pci.h>
  35
  36 #ifdef CONFIG_X86
  37 #include <asm/desc.h>
  38 #include <asm/ptrace.h>
  39 #include <asm/irq.h>
  40 #include <asm/idle.h>
  41 #include <asm/io_apic.h>
  42 #include <asm/xen/page.h>
  43 #include <asm/xen/pci.h>
  44 #endif
  45 #include <asm/sync_bitops.h>
  46 #include <asm/xen/hypercall.h>
  47 #include <asm/xen/hypervisor.h>
  48
  49 #include <xen/xen.h>
  50 #include <xen/hvm.h>
  51 #include <xen/xen-ops.h>
  52 #include <xen/events.h>
  53 #include <xen/interface/xen.h>
  54 #include <xen/interface/event_channel.h>
  55 #include <xen/interface/hvm/hvm_op.h>
  56 #include <xen/interface/hvm/params.h>
  57 #include <xen/interface/physdev.h>
  58 #include <xen/interface/sched.h>
  59 #include <xen/interface/vcpu.h>
  60 #include <asm/hw_irq.h>
  61
  62 #include "events_internal.h"
  63
  64 const struct evtchn_ops *evtchn_ops;
  65
  66 /*
  67  * This lock protects updates to the following mapping and reference-count
  68  * arrays. The lock does not need to be acquired to read the mapping tables.
  69  */
  70 static DEFINE_MUTEX(irq_mapping_update_lock);
  71
  72 static LIST_HEAD(xen_irq_list_head);
  73
  74 /* IRQ <-> VIRQ mapping. */
  75 static DEFINE_PER_CPU(int [NR_VIRQS], virq_to_irq) = {[0 ... NR_VIRQS-1] = -1};
  76
  77 /* IRQ <-> IPI mapping */
  78 static DEFINE_PER_CPU(int [XEN_NR_IPIS], ipi_to_irq) = {[0 ... XEN_NR_IPIS-1] = -1};
  79
  80 int *evtchn_to_irq;
  81 #ifdef CONFIG_X86
  82 static unsigned long *pirq_eoi_map;
  83 #endif
  84 static bool (*pirq_needs_eoi)(unsigned irq);
  85
  86 /* Xen will never allocate port zero for any purpose. */
  87 #define VALID_EVTCHN(chn)       ((chn) != 0)
  88
  89 static struct irq_chip xen_dynamic_chip;
  90 static struct irq_chip xen_percpu_chip;
  91 static struct irq_chip xen_pirq_chip;
  92 static void enable_dynirq(struct irq_data *data);
  93 static void disable_dynirq(struct irq_data *data);
  94
  95 /* Get info for IRQ */
  96 struct irq_info *info_for_irq(unsigned irq)
  97 {
  98         return irq_get_handler_data(irq);
  99 }
 100
 101 /* Constructors for packed IRQ information. */
 102 static int xen_irq_info_common_setup(struct irq_info *info,
 103                                      unsigned irq,
 104                                      enum xen_irq_type type,
 105                                      unsigned short evtchn,
 106                                      unsigned short cpu)
 107 {
 108
 109         BUG_ON(info->type != IRQT_UNBOUND && info->type != type);
 110
 111         info->type = type;
 112         info->irq = irq;
 113         info->evtchn = evtchn;
 114         info->cpu = cpu;
 115
 116         evtchn_to_irq[evtchn] = irq;
 117
 118         irq_clear_status_flags(irq, IRQ_NOREQUEST|IRQ_NOAUTOEN);
 119
 120         return xen_evtchn_port_setup(info);
 121 }
 122
 123 static int xen_irq_info_evtchn_setup(unsigned irq,
 124                                      unsigned short evtchn)
 125 {
 126         struct irq_info *info = info_for_irq(irq);
 127
 128         return xen_irq_info_common_setup(info, irq, IRQT_EVTCHN, evtchn, 0);
 129 }
 130
 131 static int xen_irq_info_ipi_setup(unsigned cpu,
 132                                   unsigned irq,
 133                                   unsigned short evtchn,
 134                                   enum ipi_vector ipi)
 135 {
 136         struct irq_info *info = info_for_irq(irq);
 137
 138         info->u.ipi = ipi;
 139
 140         per_cpu(ipi_to_irq, cpu)[ipi] = irq;
 141
 142         return xen_irq_info_common_setup(info, irq, IRQT_IPI, evtchn, 0);
 143 }
 144
 145 static int xen_irq_info_virq_setup(unsigned cpu,
 146                                    unsigned irq,
 147                                    unsigned short evtchn,
 148                                    unsigned short virq)
 149 {
 150         struct irq_info *info = info_for_irq(irq);
 151
 152         info->u.virq = virq;
 153
 154         per_cpu(virq_to_irq, cpu)[virq] = irq;
 155
 156         return xen_irq_info_common_setup(info, irq, IRQT_VIRQ, evtchn, 0);
 157 }
 158
 159 static int xen_irq_info_pirq_setup(unsigned irq,
 160                                    unsigned short evtchn,
 161                                    unsigned short pirq,
 162                                    unsigned short gsi,
 163                                    uint16_t domid,
 164                                    unsigned char flags)
 165 {
 166         struct irq_info *info = info_for_irq(irq);
 167
 168         info->u.pirq.pirq = pirq;
 169         info->u.pirq.gsi = gsi;
 170         info->u.pirq.domid = domid;
 171         info->u.pirq.flags = flags;
 172
 173         return xen_irq_info_common_setup(info, irq, IRQT_PIRQ, evtchn, 0);
 174 }
 175
 176 /*
 177  * Accessors for packed IRQ information.
 178  */
 179 unsigned int evtchn_from_irq(unsigned irq)
 180 {
 181         if (unlikely(WARN(irq < 0 || irq >= nr_irqs, "Invalid irq %d!\n", irq)))
 182                 return 0;
 183
 184         return info_for_irq(irq)->evtchn;
 185 }
 186
 187 unsigned irq_from_evtchn(unsigned int evtchn)
 188 {
 189         return evtchn_to_irq[evtchn];
 190 }
 191 EXPORT_SYMBOL_GPL(irq_from_evtchn);
 192
 193 int irq_from_virq(unsigned int cpu, unsigned int virq)
 194 {
 195         return per_cpu(virq_to_irq, cpu)[virq];
 196 }
 197
 198 static enum ipi_vector ipi_from_irq(unsigned irq)
 199 {
 200         struct irq_info *info = info_for_irq(irq);
 201
 202         BUG_ON(info == NULL);
 203         BUG_ON(info->type != IRQT_IPI);
 204
 205         return info->u.ipi;
 206 }
 207
 208 static unsigned virq_from_irq(unsigned irq)
 209 {
 210         struct irq_info *info = info_for_irq(irq);
 211
 212         BUG_ON(info == NULL);
 213         BUG_ON(info->type != IRQT_VIRQ);
 214
 215         return info->u.virq;
 216 }
 217
 218 static unsigned pirq_from_irq(unsigned irq)
 219 {
 220         struct irq_info *info = info_for_irq(irq);
 221
 222         BUG_ON(info == NULL);
 223         BUG_ON(info->type != IRQT_PIRQ);
 224
 225         return info->u.pirq.pirq;
 226 }
 227
 228 static enum xen_irq_type type_from_irq(unsigned irq)
 229 {
 230         return info_for_irq(irq)->type;
 231 }
 232
 233 unsigned cpu_from_irq(unsigned irq)
 234 {
 235         return info_for_irq(irq)->cpu;
 236 }
 237
 238 unsigned int cpu_from_evtchn(unsigned int evtchn)
 239 {
 240         int irq = evtchn_to_irq[evtchn];
 241         unsigned ret = 0;
 242
 243         if (irq != -1)
 244                 ret = cpu_from_irq(irq);
 245
 246         return ret;
 247 }
 248
 249 #ifdef CONFIG_X86
 250 static bool pirq_check_eoi_map(unsigned irq)
 251 {
 252         return test_bit(pirq_from_irq(irq), pirq_eoi_map);
 253 }
 254 #endif
 255
 256 static bool pirq_needs_eoi_flag(unsigned irq)
 257 {
 258         struct irq_info *info = info_for_irq(irq);
 259         BUG_ON(info->type != IRQT_PIRQ);
 260
 261         return info->u.pirq.flags & PIRQ_NEEDS_EOI;
 262 }
 263
 264 static void bind_evtchn_to_cpu(unsigned int chn, unsigned int cpu)
 265 {
 266         int irq = evtchn_to_irq[chn];
 267         struct irq_info *info = info_for_irq(irq);
 268
 269         BUG_ON(irq == -1);
 270 #ifdef CONFIG_SMP
 271         cpumask_copy(irq_to_desc(irq)->irq_data.affinity, cpumask_of(cpu));
 272 #endif
 273
 274         xen_evtchn_port_bind_to_cpu(info, cpu);
 275
 276         info->cpu = cpu;
 277 }
 278
 279 /**
 280  * notify_remote_via_irq - send event to remote end of event channel via irq
 281  * @irq: irq of event channel to send event to
 282  *
 283  * Unlike notify_remote_via_evtchn(), this is safe to use across
 284  * save/restore. Notifications on a broken connection are silently
 285  * dropped.
 286  */
 287 void notify_remote_via_irq(int irq)
 288 {
 289         int evtchn = evtchn_from_irq(irq);
 290
 291         if (VALID_EVTCHN(evtchn))
 292                 notify_remote_via_evtchn(evtchn);
 293 }
 294 EXPORT_SYMBOL_GPL(notify_remote_via_irq);
 295
 296 static void xen_irq_init(unsigned irq)
 297 {
 298         struct irq_info *info;
 299 #ifdef CONFIG_SMP
 300         struct irq_desc *desc = irq_to_desc(irq);
 301
 302         /* By default all event channels notify CPU#0. */
 303         cpumask_copy(desc->irq_data.affinity, cpumask_of(0));
 304 #endif
 305
 306         info = kzalloc(sizeof(*info), GFP_KERNEL);
 307         if (info == NULL)
 308                 panic("Unable to allocate metadata for IRQ%d\n", irq);
 309
 310         info->type = IRQT_UNBOUND;
 311         info->refcnt = -1;
 312
 313         irq_set_handler_data(irq, info);
 314
 315         list_add_tail(&info->list, &xen_irq_list_head);
 316 }
 317
 318 static int __must_check xen_allocate_irq_dynamic(void)
 319 {
 320         int first = 0;
 321         int irq;
 322
 323 #ifdef CONFIG_X86_IO_APIC
 324         /*
 325          * For an HVM guest or domain 0 which see "real" (emulated or
 326          * actual respectively) GSIs we allocate dynamic IRQs
 327          * e.g. those corresponding to event channels or MSIs
 328          * etc. from the range above those "real" GSIs to avoid
 329          * collisions.
 330          */
 331         if (xen_initial_domain() || xen_hvm_domain())
 332                 first = get_nr_irqs_gsi();
 333 #endif
 334
 335         irq = irq_alloc_desc_from(first, -1);
 336
 337         if (irq >= 0)
 338                 xen_irq_init(irq);
 339
 340         return irq;
 341 }
 342
 343 static int __must_check xen_allocate_irq_gsi(unsigned gsi)
 344 {
 345         int irq;
 346
 347         /*
 348          * A PV guest has no concept of a GSI (since it has no ACPI
 349          * nor access to/knowledge of the physical APICs). Therefore
 350          * all IRQs are dynamically allocated from the entire IRQ
 351          * space.
 352          */
 353         if (xen_pv_domain() && !xen_initial_domain())
 354                 return xen_allocate_irq_dynamic();
 355
 356         /* Legacy IRQ descriptors are already allocated by the arch. */
 357         if (gsi < NR_IRQS_LEGACY)
 358                 irq = gsi;
 359         else
 360                 irq = irq_alloc_desc_at(gsi, -1);
 361
 362         xen_irq_init(irq);
 363
 364         return irq;
 365 }
 366
 367 static void xen_free_irq(unsigned irq)
 368 {
 369         struct irq_info *info = irq_get_handler_data(irq);
 370
 371         if (WARN_ON(!info))
 372                 return;
 373
 374         list_del(&info->list);
 375
 376         irq_set_handler_data(irq, NULL);
 377
 378         WARN_ON(info->refcnt > 0);
 379
 380         kfree(info);
 381
 382         /* Legacy IRQ descriptors are managed by the arch. */
 383         if (irq < NR_IRQS_LEGACY)
 384                 return;
 385
 386         irq_free_desc(irq);
 387 }
 388
 389 static void pirq_query_unmask(int irq)
 390 {
 391         struct physdev_irq_status_query irq_status;
 392         struct irq_info *info = info_for_irq(irq);
 393
 394         BUG_ON(info->type != IRQT_PIRQ);
 395
 396         irq_status.irq = pirq_from_irq(irq);
 397         if (HYPERVISOR_physdev_op(PHYSDEVOP_irq_status_query, &irq_status))
 398                 irq_status.flags = 0;
 399
 400         info->u.pirq.flags &= ~PIRQ_NEEDS_EOI;
 401         if (irq_status.flags & XENIRQSTAT_needs_eoi)
 402                 info->u.pirq.flags |= PIRQ_NEEDS_EOI;
 403 }
 404
 405 static bool probing_irq(int irq)
 406 {
 407         struct irq_desc *desc = irq_to_desc(irq);
 408
 409         return desc && desc->action == NULL;
 410 }
 411
 412 static void eoi_pirq(struct irq_data *data)
 413 {
 414         int evtchn = evtchn_from_irq(data->irq);
 415         struct physdev_eoi eoi = { .irq = pirq_from_irq(data->irq) };
 416         int rc = 0;
 417
 418         irq_move_irq(data);
 419
 420         if (VALID_EVTCHN(evtchn))
 421                 clear_evtchn(evtchn);
 422
 423         if (pirq_needs_eoi(data->irq)) {
 424                 rc = HYPERVISOR_physdev_op(PHYSDEVOP_eoi, &eoi);
 425                 WARN_ON(rc);
 426         }
 427 }
 428
 429 static void mask_ack_pirq(struct irq_data *data)
 430 {
 431         disable_dynirq(data);
 432         eoi_pirq(data);
 433 }
 434
 435 static unsigned int __startup_pirq(unsigned int irq)
 436 {
 437         struct evtchn_bind_pirq bind_pirq;
 438         struct irq_info *info = info_for_irq(irq);
 439         int evtchn = evtchn_from_irq(irq);
 440         int rc;
 441
 442         BUG_ON(info->type != IRQT_PIRQ);
 443
 444         if (VALID_EVTCHN(evtchn))
 445                 goto out;
 446
 447         bind_pirq.pirq = pirq_from_irq(irq);
 448         /* NB. We are happy to share unless we are probing. */
 449         bind_pirq.flags = info->u.pirq.flags & PIRQ_SHAREABLE ?
 450                                         BIND_PIRQ__WILL_SHARE : 0;
 451         rc = HYPERVISOR_event_channel_op(EVTCHNOP_bind_pirq, &bind_pirq);
 452         if (rc != 0) {
 453                 if (!probing_irq(irq))
 454                         pr_info("Failed to obtain physical IRQ %d\n", irq);
 455                 return 0;
 456         }
 457         evtchn = bind_pirq.port;
 458
 459         pirq_query_unmask(irq);
 460
 461         evtchn_to_irq[evtchn] = irq;
 462         bind_evtchn_to_cpu(evtchn, 0);
 463         info->evtchn = evtchn;
 464
 465 out:
 466         unmask_evtchn(evtchn);
 467         eoi_pirq(irq_get_irq_data(irq));
 468
 469         return 0;
 470 }
 471
 472 static unsigned int startup_pirq(struct irq_data *data)
 473 {
 474         return __startup_pirq(data->irq);
 475 }
 476
 477 static void shutdown_pirq(struct irq_data *data)
 478 {
 479         struct evtchn_close close;
 480         unsigned int irq = data->irq;
 481         struct irq_info *info = info_for_irq(irq);
 482         int evtchn = evtchn_from_irq(irq);
 483
 484         BUG_ON(info->type != IRQT_PIRQ);
 485
 486         if (!VALID_EVTCHN(evtchn))
 487                 return;
 488
 489         mask_evtchn(evtchn);
 490
 491         close.port = evtchn;
 492         if (HYPERVISOR_event_channel_op(EVTCHNOP_close, &close) != 0)
 493                 BUG();
 494
 495         bind_evtchn_to_cpu(evtchn, 0);
 496         evtchn_to_irq[evtchn] = -1;
 497         info->evtchn = 0;
 498 }
 499
 500 static void enable_pirq(struct irq_data *data)
 501 {
 502         startup_pirq(data);
 503 }
 504
 505 static void disable_pirq(struct irq_data *data)
 506 {
 507         disable_dynirq(data);
 508 }
 509
 510 int xen_irq_from_gsi(unsigned gsi)
 511 {
 512         struct irq_info *info;
 513
 514         list_for_each_entry(info, &xen_irq_list_head, list) {
 515                 if (info->type != IRQT_PIRQ)
 516                         continue;
 517
 518                 if (info->u.pirq.gsi == gsi)
 519                         return info->irq;
 520         }
 521
 522         return -1;
 523 }
 524 EXPORT_SYMBOL_GPL(xen_irq_from_gsi);
 525
 526 static void __unbind_from_irq(unsigned int irq)
 527 {
 528         struct evtchn_close close;
 529         int evtchn = evtchn_from_irq(irq);
 530         struct irq_info *info = irq_get_handler_data(irq);
 531
 532         if (info->refcnt > 0) {
 533                 info->refcnt--;
 534                 if (info->refcnt != 0)
 535                         return;
 536         }
 537
 538         if (VALID_EVTCHN(evtchn)) {
 539                 close.port = evtchn;
 540                 if (HYPERVISOR_event_channel_op(EVTCHNOP_close, &close) != 0)
 541                         BUG();
 542
 543                 switch (type_from_irq(irq)) {
 544                 case IRQT_VIRQ:
 545                         per_cpu(virq_to_irq, cpu_from_evtchn(evtchn))
 546                                 [virq_from_irq(irq)] = -1;
 547                         break;
 548                 case IRQT_IPI:
 549                         per_cpu(ipi_to_irq, cpu_from_evtchn(evtchn))
 550                                 [ipi_from_irq(irq)] = -1;
 551                         break;
 552                 default:
 553                         break;
 554                 }
 555
 556                 /* Closed ports are implicitly re-bound to VCPU0. */
 557                 bind_evtchn_to_cpu(evtchn, 0);
 558
 559                 evtchn_to_irq[evtchn] = -1;
 560         }
 561
 562         BUG_ON(info_for_irq(irq)->type == IRQT_UNBOUND);
 563
 564         xen_free_irq(irq);
 565 }
 566
 567 /*
 568  * Do not make any assumptions regarding the relationship between the
 569  * IRQ number returned here and the Xen pirq argument.
 570  *
 571  * Note: We don't assign an event channel until the irq actually started
 572  * up.  Return an existing irq if we've already got one for the gsi.
 573  *
 574  * Shareable implies level triggered, not shareable implies edge
 575  * triggered here.
 576  */
 577 int xen_bind_pirq_gsi_to_irq(unsigned gsi,
 578                              unsigned pirq, int shareable, char *name)
 579 {
 580         int irq = -1;
 581         struct physdev_irq irq_op;
 582         int ret;
 583
 584         mutex_lock(&irq_mapping_update_lock);
 585
 586         irq = xen_irq_from_gsi(gsi);
 587         if (irq != -1) {
 588                 pr_info("%s: returning irq %d for gsi %u\n",
 589                         __func__, irq, gsi);
 590                 goto out;
 591         }
 592
 593         irq = xen_allocate_irq_gsi(gsi);
 594         if (irq < 0)
 595                 goto out;
 596
 597         irq_op.irq = irq;
 598         irq_op.vector = 0;
 599
 600         /* Only the privileged domain can do this. For non-priv, the pcifront
 601          * driver provides a PCI bus that does the call to do exactly
 602          * this in the priv domain. */
 603         if (xen_initial_domain() &&
 604             HYPERVISOR_physdev_op(PHYSDEVOP_alloc_irq_vector, &irq_op)) {
 605                 xen_free_irq(irq);
 606                 irq = -ENOSPC;
 607                 goto out;
 608         }
 609
 610         ret = xen_irq_info_pirq_setup(irq, 0, pirq, gsi, DOMID_SELF,
 611                                shareable ? PIRQ_SHAREABLE : 0);
 612         if (ret < 0) {
 613                 __unbind_from_irq(irq);
 614                 irq = ret;
 615                 goto out;
 616         }
 617
 618         pirq_query_unmask(irq);
 619         /* We try to use the handler with the appropriate semantic for the
 620          * type of interrupt: if the interrupt is an edge triggered
 621          * interrupt we use handle_edge_irq.
 622          *
 623          * On the other hand if the interrupt is level triggered we use
 624          * handle_fasteoi_irq like the native code does for this kind of
 625          * interrupts.
 626          *
 627          * Depending on the Xen version, pirq_needs_eoi might return true
 628          * not only for level triggered interrupts but for edge triggered
 629          * interrupts too. In any case Xen always honors the eoi mechanism,
 630          * not injecting any more pirqs of the same kind if the first one
 631          * hasn't received an eoi yet. Therefore using the fasteoi handler
 632          * is the right choice either way.
 633          */
 634         if (shareable)
 635                 irq_set_chip_and_handler_name(irq, &xen_pirq_chip,
 636                                 handle_fasteoi_irq, name);
 637         else
 638                 irq_set_chip_and_handler_name(irq, &xen_pirq_chip,
 639                                 handle_edge_irq, name);
 640
 641 out:
 642         mutex_unlock(&irq_mapping_update_lock);
 643
 644         return irq;
 645 }
 646
 647 #ifdef CONFIG_PCI_MSI
 648 int xen_allocate_pirq_msi(struct pci_dev *dev, struct msi_desc *msidesc)
 649 {
 650         int rc;
 651         struct physdev_get_free_pirq op_get_free_pirq;
 652
 653         op_get_free_pirq.type = MAP_PIRQ_TYPE_MSI;
 654         rc = HYPERVISOR_physdev_op(PHYSDEVOP_get_free_pirq, &op_get_free_pirq);
 655
 656         WARN_ONCE(rc == -ENOSYS,
 657                   "hypervisor does not support the PHYSDEVOP_get_free_pirq interface\n");
 658
 659         return rc ? -1 : op_get_free_pirq.pirq;
 660 }
 661
 662 int xen_bind_pirq_msi_to_irq(struct pci_dev *dev, struct msi_desc *msidesc,
 663                              int pirq, const char *name, domid_t domid)
 664 {
 665         int irq, ret;
 666
 667         mutex_lock(&irq_mapping_update_lock);
 668
 669         irq = xen_allocate_irq_dynamic();
 670         if (irq < 0)
 671                 goto out;
 672
 673         irq_set_chip_and_handler_name(irq, &xen_pirq_chip, handle_edge_irq,
 674                         name);
 675
 676         ret = xen_irq_info_pirq_setup(irq, 0, pirq, 0, domid, 0);
 677         if (ret < 0)
 678                 goto error_irq;
 679         ret = irq_set_msi_desc(irq, msidesc);
 680         if (ret < 0)
 681                 goto error_irq;
 682 out:
 683         mutex_unlock(&irq_mapping_update_lock);
 684         return irq;
 685 error_irq:
 686         __unbind_from_irq(irq);
 687         mutex_unlock(&irq_mapping_update_lock);
 688         return ret;
 689 }
 690 #endif
 691
 692 int xen_destroy_irq(int irq)
 693 {
 694         struct irq_desc *desc;
 695         struct physdev_unmap_pirq unmap_irq;
 696         struct irq_info *info = info_for_irq(irq);
 697         int rc = -ENOENT;
 698
 699         mutex_lock(&irq_mapping_update_lock);
 700
 701         desc = irq_to_desc(irq);
 702         if (!desc)
 703                 goto out;
 704
 705         if (xen_initial_domain()) {
 706                 unmap_irq.pirq = info->u.pirq.pirq;
 707                 unmap_irq.domid = info->u.pirq.domid;
 708                 rc = HYPERVISOR_physdev_op(PHYSDEVOP_unmap_pirq, &unmap_irq);
 709                 /* If another domain quits without making the pci_disable_msix
 710                  * call, the Xen hypervisor takes care of freeing the PIRQs
 711                  * (free_domain_pirqs).
 712                  */
 713                 if ((rc == -ESRCH && info->u.pirq.domid != DOMID_SELF))
 714                         pr_info("domain %d does not have %d anymore\n",
 715                                 info->u.pirq.domid, info->u.pirq.pirq);
 716                 else if (rc) {
 717                         pr_warn("unmap irq failed %d\n", rc);
 718                         goto out;
 719                 }
 720         }
 721
 722         xen_free_irq(irq);
 723
 724 out:
 725         mutex_unlock(&irq_mapping_update_lock);
 726         return rc;
 727 }
 728
 729 int xen_irq_from_pirq(unsigned pirq)
 730 {
 731         int irq;
 732
 733         struct irq_info *info;
 734
 735         mutex_lock(&irq_mapping_update_lock);
 736
 737         list_for_each_entry(info, &xen_irq_list_head, list) {
 738                 if (info->type != IRQT_PIRQ)
 739                         continue;
 740                 irq = info->irq;
 741                 if (info->u.pirq.pirq == pirq)
 742                         goto out;
 743         }
 744         irq = -1;
 745 out:
 746         mutex_unlock(&irq_mapping_update_lock);
 747
 748         return irq;
 749 }
 750
 751
 752 int xen_pirq_from_irq(unsigned irq)
 753 {
 754         return pirq_from_irq(irq);
 755 }
 756 EXPORT_SYMBOL_GPL(xen_pirq_from_irq);
 757
 758 int bind_evtchn_to_irq(unsigned int evtchn)
 759 {
 760         int irq;
 761         int ret;
 762
 763         mutex_lock(&irq_mapping_update_lock);
 764
 765         irq = evtchn_to_irq[evtchn];
 766
 767         if (irq == -1) {
 768                 irq = xen_allocate_irq_dynamic();
 769                 if (irq < 0)
 770                         goto out;
 771
 772                 irq_set_chip_and_handler_name(irq, &xen_dynamic_chip,
 773                                               handle_edge_irq, "event");
 774
 775                 ret = xen_irq_info_evtchn_setup(irq, evtchn);
 776                 if (ret < 0) {
 777                         __unbind_from_irq(irq);
 778                         irq = ret;
 779                         goto out;
 780                 }
 781         } else {
 782                 struct irq_info *info = info_for_irq(irq);
 783                 WARN_ON(info == NULL || info->type != IRQT_EVTCHN);
 784         }
 785
 786 out:
 787         mutex_unlock(&irq_mapping_update_lock);
 788
 789         return irq;
 790 }
 791 EXPORT_SYMBOL_GPL(bind_evtchn_to_irq);
 792
 793 static int bind_ipi_to_irq(unsigned int ipi, unsigned int cpu)
 794 {
 795         struct evtchn_bind_ipi bind_ipi;
 796         int evtchn, irq;
 797         int ret;
 798
 799         mutex_lock(&irq_mapping_update_lock);
 800
 801         irq = per_cpu(ipi_to_irq, cpu)[ipi];
 802
 803         if (irq == -1) {
 804                 irq = xen_allocate_irq_dynamic();
 805                 if (irq < 0)
 806                         goto out;
 807
 808                 irq_set_chip_and_handler_name(irq, &xen_percpu_chip,
 809                                               handle_percpu_irq, "ipi");
 810
 811                 bind_ipi.vcpu = cpu;
 812                 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_ipi,
 813                                                 &bind_ipi) != 0)
 814                         BUG();
 815                 evtchn = bind_ipi.port;
 816
 817                 ret = xen_irq_info_ipi_setup(cpu, irq, evtchn, ipi);
 818                 if (ret < 0) {
 819                         __unbind_from_irq(irq);
 820                         irq = ret;
 821                         goto out;
 822                 }
 823                 bind_evtchn_to_cpu(evtchn, cpu);
 824         } else {
 825                 struct irq_info *info = info_for_irq(irq);
 826                 WARN_ON(info == NULL || info->type != IRQT_IPI);
 827         }
 828
 829  out:
 830         mutex_unlock(&irq_mapping_update_lock);
 831         return irq;
 832 }
 833
 834 static int bind_interdomain_evtchn_to_irq(unsigned int remote_domain,
 835                                           unsigned int remote_port)
 836 {
 837         struct evtchn_bind_interdomain bind_interdomain;
 838         int err;
 839
 840         bind_interdomain.remote_dom  = remote_domain;
 841         bind_interdomain.remote_port = remote_port;
 842
 843         err = HYPERVISOR_event_channel_op(EVTCHNOP_bind_interdomain,
 844                                           &bind_interdomain);
 845
 846         return err ? : bind_evtchn_to_irq(bind_interdomain.local_port);
 847 }
 848
 849 static int find_virq(unsigned int virq, unsigned int cpu)
 850 {
 851         struct evtchn_status status;
 852         int port, rc = -ENOENT;
 853
 854         memset(&status, 0, sizeof(status));
 855         for (port = 0; port <= NR_EVENT_CHANNELS; port++) {
 856                 status.dom = DOMID_SELF;
 857                 status.port = port;
 858                 rc = HYPERVISOR_event_channel_op(EVTCHNOP_status, &status);
 859                 if (rc < 0)
 860                         continue;
 861                 if (status.status != EVTCHNSTAT_virq)
 862                         continue;
 863                 if (status.u.virq == virq && status.vcpu == cpu) {
 864                         rc = port;
 865                         break;
 866                 }
 867         }
 868         return rc;
 869 }
 870
 871 int bind_virq_to_irq(unsigned int virq, unsigned int cpu)
 872 {
 873         struct evtchn_bind_virq bind_virq;
 874         int evtchn, irq, ret;
 875
 876         mutex_lock(&irq_mapping_update_lock);
 877
 878         irq = per_cpu(virq_to_irq, cpu)[virq];
 879
 880         if (irq == -1) {
 881                 irq = xen_allocate_irq_dynamic();
 882                 if (irq < 0)
 883                         goto out;
 884
 885                 irq_set_chip_and_handler_name(irq, &xen_percpu_chip,
 886                                               handle_percpu_irq, "virq");
 887
 888                 bind_virq.virq = virq;
 889                 bind_virq.vcpu = cpu;
 890                 ret = HYPERVISOR_event_channel_op(EVTCHNOP_bind_virq,
 891                                                 &bind_virq);
 892                 if (ret == 0)
 893                         evtchn = bind_virq.port;
 894                 else {
 895                         if (ret == -EEXIST)
 896                                 ret = find_virq(virq, cpu);
 897                         BUG_ON(ret < 0);
 898                         evtchn = ret;
 899                 }
 900
 901                 ret = xen_irq_info_virq_setup(cpu, irq, evtchn, virq);
 902                 if (ret < 0) {
 903                         __unbind_from_irq(irq);
 904                         irq = ret;
 905                         goto out;
 906                 }
 907
 908                 bind_evtchn_to_cpu(evtchn, cpu);
 909         } else {
 910                 struct irq_info *info = info_for_irq(irq);
 911                 WARN_ON(info == NULL || info->type != IRQT_VIRQ);
 912         }
 913
 914 out:
 915         mutex_unlock(&irq_mapping_update_lock);
 916
 917         return irq;
 918 }
 919
 920 static void unbind_from_irq(unsigned int irq)
 921 {
 922         mutex_lock(&irq_mapping_update_lock);
 923         __unbind_from_irq(irq);
 924         mutex_unlock(&irq_mapping_update_lock);
 925 }
 926
 927 int bind_evtchn_to_irqhandler(unsigned int evtchn,
 928                               irq_handler_t handler,
 929                               unsigned long irqflags,
 930                               const char *devname, void *dev_id)
 931 {
 932         int irq, retval;
 933
 934         irq = bind_evtchn_to_irq(evtchn);
 935         if (irq < 0)
 936                 return irq;
 937         retval = request_irq(irq, handler, irqflags, devname, dev_id);
 938         if (retval != 0) {
 939                 unbind_from_irq(irq);
 940                 return retval;
 941         }
 942
 943         return irq;
 944 }
 945 EXPORT_SYMBOL_GPL(bind_evtchn_to_irqhandler);
 946
 947 int bind_interdomain_evtchn_to_irqhandler(unsigned int remote_domain,
 948                                           unsigned int remote_port,
 949                                           irq_handler_t handler,
 950                                           unsigned long irqflags,
 951                                           const char *devname,
 952                                           void *dev_id)
 953 {
 954         int irq, retval;
 955
 956         irq = bind_interdomain_evtchn_to_irq(remote_domain, remote_port);
 957         if (irq < 0)
 958                 return irq;
 959
 960         retval = request_irq(irq, handler, irqflags, devname, dev_id);
 961         if (retval != 0) {
 962                 unbind_from_irq(irq);
 963                 return retval;
 964         }
 965
 966         return irq;
 967 }
 968 EXPORT_SYMBOL_GPL(bind_interdomain_evtchn_to_irqhandler);
 969
 970 int bind_virq_to_irqhandler(unsigned int virq, unsigned int cpu,
 971                             irq_handler_t handler,
 972                             unsigned long irqflags, const char *devname, void *dev_id)
 973 {
 974         int irq, retval;
 975
 976         irq = bind_virq_to_irq(virq, cpu);
 977         if (irq < 0)
 978                 return irq;
 979         retval = request_irq(irq, handler, irqflags, devname, dev_id);
 980         if (retval != 0) {
 981                 unbind_from_irq(irq);
 982                 return retval;
 983         }
 984
 985         return irq;
 986 }
 987 EXPORT_SYMBOL_GPL(bind_virq_to_irqhandler);
 988
 989 int bind_ipi_to_irqhandler(enum ipi_vector ipi,
 990                            unsigned int cpu,
 991                            irq_handler_t handler,
 992                            unsigned long irqflags,
 993                            const char *devname,
 994                            void *dev_id)
 995 {
 996         int irq, retval;
 997
 998         irq = bind_ipi_to_irq(ipi, cpu);
 999         if (irq < 0)
1000                 return irq;
1001
1002         irqflags |= IRQF_NO_SUSPEND | IRQF_FORCE_RESUME | IRQF_EARLY_RESUME;
1003         retval = request_irq(irq, handler, irqflags, devname, dev_id);
1004         if (retval != 0) {
1005                 unbind_from_irq(irq);
1006                 return retval;
1007         }
1008
1009         return irq;
1010 }
1011
1012 void unbind_from_irqhandler(unsigned int irq, void *dev_id)
1013 {
1014         struct irq_info *info = irq_get_handler_data(irq);
1015
1016         if (WARN_ON(!info))
1017                 return;
1018         free_irq(irq, dev_id);
1019         unbind_from_irq(irq);
1020 }
1021 EXPORT_SYMBOL_GPL(unbind_from_irqhandler);
1022
1023 int evtchn_make_refcounted(unsigned int evtchn)
1024 {
1025         int irq = evtchn_to_irq[evtchn];
1026         struct irq_info *info;
1027
1028         if (irq == -1)
1029                 return -ENOENT;
1030
1031         info = irq_get_handler_data(irq);
1032
1033         if (!info)
1034                 return -ENOENT;
1035
1036         WARN_ON(info->refcnt != -1);
1037
1038         info->refcnt = 1;
1039
1040         return 0;
1041 }
1042 EXPORT_SYMBOL_GPL(evtchn_make_refcounted);
1043
1044 int evtchn_get(unsigned int evtchn)
1045 {
1046         int irq;
1047         struct irq_info *info;
1048         int err = -ENOENT;
1049
1050         if (evtchn >= NR_EVENT_CHANNELS)
1051                 return -EINVAL;
1052
1053         mutex_lock(&irq_mapping_update_lock);
1054
1055         irq = evtchn_to_irq[evtchn];
1056         if (irq == -1)
1057                 goto done;
1058
1059         info = irq_get_handler_data(irq);
1060
1061         if (!info)
1062                 goto done;
1063
1064         err = -EINVAL;
1065         if (info->refcnt <= 0)
1066                 goto done;
1067
1068         info->refcnt++;
1069         err = 0;
1070  done:
1071         mutex_unlock(&irq_mapping_update_lock);
1072
1073         return err;
1074 }
1075 EXPORT_SYMBOL_GPL(evtchn_get);
1076
1077 void evtchn_put(unsigned int evtchn)
1078 {
1079         int irq = evtchn_to_irq[evtchn];
1080         if (WARN_ON(irq == -1))
1081                 return;
1082         unbind_from_irq(irq);
1083 }
1084 EXPORT_SYMBOL_GPL(evtchn_put);
1085
1086 void xen_send_IPI_one(unsigned int cpu, enum ipi_vector vector)
1087 {
1088         int irq;
1089
1090 #ifdef CONFIG_X86
1091         if (unlikely(vector == XEN_NMI_VECTOR)) {
1092                 int rc =  HYPERVISOR_vcpu_op(VCPUOP_send_nmi, cpu, NULL);
1093                 if (rc < 0)
1094                         printk(KERN_WARNING "Sending nmi to CPU%d failed (rc:%d)\n", cpu, rc);
1095                 return;
1096         }
1097 #endif
1098         irq = per_cpu(ipi_to_irq, cpu)[vector];
1099         BUG_ON(irq < 0);
1100         notify_remote_via_irq(irq);
1101 }
1102
1103 static DEFINE_PER_CPU(unsigned, xed_nesting_count);
1104
1105 static void __xen_evtchn_do_upcall(void)
1106 {
1107         struct vcpu_info *vcpu_info = __this_cpu_read(xen_vcpu);
1108         int cpu = get_cpu();
1109         unsigned count;
1110
1111         do {
1112                 vcpu_info->evtchn_upcall_pending = 0;
1113
1114                 if (__this_cpu_inc_return(xed_nesting_count) - 1)
1115                         goto out;
1116
1117                 xen_evtchn_handle_events(cpu);
1118
1119                 BUG_ON(!irqs_disabled());
1120
1121                 count = __this_cpu_read(xed_nesting_count);
1122                 __this_cpu_write(xed_nesting_count, 0);
1123         } while (count != 1 || vcpu_info->evtchn_upcall_pending);
1124
1125 out:
1126
1127         put_cpu();
1128 }
1129
1130 void xen_evtchn_do_upcall(struct pt_regs *regs)
1131 {
1132         struct pt_regs *old_regs = set_irq_regs(regs);
1133
1134         irq_enter();
1135 #ifdef CONFIG_X86
1136         exit_idle();
1137 #endif
1138
1139         __xen_evtchn_do_upcall();
1140
1141         irq_exit();
1142         set_irq_regs(old_regs);
1143 }
1144
1145 void xen_hvm_evtchn_do_upcall(void)
1146 {
1147         __xen_evtchn_do_upcall();
1148 }
1149 EXPORT_SYMBOL_GPL(xen_hvm_evtchn_do_upcall);
1150
1151 /* Rebind a new event channel to an existing irq. */
1152 void rebind_evtchn_irq(int evtchn, int irq)
1153 {
1154         struct irq_info *info = info_for_irq(irq);
1155
1156         if (WARN_ON(!info))
1157                 return;
1158
1159         /* Make sure the irq is masked, since the new event channel
1160            will also be masked. */
1161         disable_irq(irq);
1162
1163         mutex_lock(&irq_mapping_update_lock);
1164
1165         /* After resume the irq<->evtchn mappings are all cleared out */
1166         BUG_ON(evtchn_to_irq[evtchn] != -1);
1167         /* Expect irq to have been bound before,
1168            so there should be a proper type */
1169         BUG_ON(info->type == IRQT_UNBOUND);
1170
1171         (void)xen_irq_info_evtchn_setup(irq, evtchn);
1172
1173         mutex_unlock(&irq_mapping_update_lock);
1174
1175         /* new event channels are always bound to cpu 0 */
1176         irq_set_affinity(irq, cpumask_of(0));
1177
1178         /* Unmask the event channel. */
1179         enable_irq(irq);
1180 }
1181
1182 /* Rebind an evtchn so that it gets delivered to a specific cpu */
1183 static int rebind_irq_to_cpu(unsigned irq, unsigned tcpu)
1184 {
1185         struct evtchn_bind_vcpu bind_vcpu;
1186         int evtchn = evtchn_from_irq(irq);
1187         int masked;
1188
1189         if (!VALID_EVTCHN(evtchn))
1190                 return -1;
1191
1192         /*
1193          * Events delivered via platform PCI interrupts are always
1194          * routed to vcpu 0 and hence cannot be rebound.
1195          */
1196         if (xen_hvm_domain() && !xen_have_vector_callback)
1197                 return -1;
1198
1199         /* Send future instances of this interrupt to other vcpu. */
1200         bind_vcpu.port = evtchn;
1201         bind_vcpu.vcpu = tcpu;
1202
1203         /*
1204          * Mask the event while changing the VCPU binding to prevent
1205          * it being delivered on an unexpected VCPU.
1206          */
1207         masked = test_and_set_mask(evtchn);
1208
1209         /*
1210          * If this fails, it usually just indicates that we're dealing with a
1211          * virq or IPI channel, which don't actually need to be rebound. Ignore
1212          * it, but don't do the xenlinux-level rebind in that case.
1213          */
1214         if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_vcpu, &bind_vcpu) >= 0)
1215                 bind_evtchn_to_cpu(evtchn, tcpu);
1216
1217         if (!masked)
1218                 unmask_evtchn(evtchn);
1219
1220         return 0;
1221 }
1222
1223 static int set_affinity_irq(struct irq_data *data, const struct cpumask *dest,
1224                             bool force)
1225 {
1226         unsigned tcpu = cpumask_first(dest);
1227
1228         return rebind_irq_to_cpu(data->irq, tcpu);
1229 }
1230
1231 static int retrigger_evtchn(int evtchn)
1232 {
1233         int masked;
1234
1235         if (!VALID_EVTCHN(evtchn))
1236                 return 0;
1237
1238         masked = test_and_set_mask(evtchn);
1239         set_evtchn(evtchn);
1240         if (!masked)
1241                 unmask_evtchn(evtchn);
1242
1243         return 1;
1244 }
1245
1246 int resend_irq_on_evtchn(unsigned int irq)
1247 {
1248         return retrigger_evtchn(evtchn_from_irq(irq));
1249 }
1250
1251 static void enable_dynirq(struct irq_data *data)
1252 {
1253         int evtchn = evtchn_from_irq(data->irq);
1254
1255         if (VALID_EVTCHN(evtchn))
1256                 unmask_evtchn(evtchn);
1257 }
1258
1259 static void disable_dynirq(struct irq_data *data)
1260 {
1261         int evtchn = evtchn_from_irq(data->irq);
1262
1263         if (VALID_EVTCHN(evtchn))
1264                 mask_evtchn(evtchn);
1265 }
1266
1267 static void ack_dynirq(struct irq_data *data)
1268 {
1269         int evtchn = evtchn_from_irq(data->irq);
1270
1271         irq_move_irq(data);
1272
1273         if (VALID_EVTCHN(evtchn))
1274                 clear_evtchn(evtchn);
1275 }
1276
1277 static void mask_ack_dynirq(struct irq_data *data)
1278 {
1279         disable_dynirq(data);
1280         ack_dynirq(data);
1281 }
1282
1283 static int retrigger_dynirq(struct irq_data *data)
1284 {
1285         return retrigger_evtchn(evtchn_from_irq(data->irq));
1286 }
1287
1288 static void restore_pirqs(void)
1289 {
1290         int pirq, rc, irq, gsi;
1291         struct physdev_map_pirq map_irq;
1292         struct irq_info *info;
1293
1294         list_for_each_entry(info, &xen_irq_list_head, list) {
1295                 if (info->type != IRQT_PIRQ)
1296                         continue;
1297
1298                 pirq = info->u.pirq.pirq;
1299                 gsi = info->u.pirq.gsi;
1300                 irq = info->irq;
1301
1302                 /* save/restore of PT devices doesn't work, so at this point the
1303                  * only devices present are GSI based emulated devices */
1304                 if (!gsi)
1305                         continue;
1306
1307                 map_irq.domid = DOMID_SELF;
1308                 map_irq.type = MAP_PIRQ_TYPE_GSI;
1309                 map_irq.index = gsi;
1310                 map_irq.pirq = pirq;
1311
1312                 rc = HYPERVISOR_physdev_op(PHYSDEVOP_map_pirq, &map_irq);
1313                 if (rc) {
1314                         pr_warn("xen map irq failed gsi=%d irq=%d pirq=%d rc=%d\n",
1315                                 gsi, irq, pirq, rc);
1316                         xen_free_irq(irq);
1317                         continue;
1318                 }
1319
1320                 printk(KERN_DEBUG "xen: --> irq=%d, pirq=%d\n", irq, map_irq.pirq);
1321
1322                 __startup_pirq(irq);
1323         }
1324 }
1325
1326 static void restore_cpu_virqs(unsigned int cpu)
1327 {
1328         struct evtchn_bind_virq bind_virq;
1329         int virq, irq, evtchn;
1330
1331         for (virq = 0; virq < NR_VIRQS; virq++) {
1332                 if ((irq = per_cpu(virq_to_irq, cpu)[virq]) == -1)
1333                         continue;
1334
1335                 BUG_ON(virq_from_irq(irq) != virq);
1336
1337                 /* Get a new binding from Xen. */
1338                 bind_virq.virq = virq;
1339                 bind_virq.vcpu = cpu;
1340                 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_virq,
1341                                                 &bind_virq) != 0)
1342                         BUG();
1343                 evtchn = bind_virq.port;
1344
1345                 /* Record the new mapping. */
1346                 (void)xen_irq_info_virq_setup(cpu, irq, evtchn, virq);
1347                 bind_evtchn_to_cpu(evtchn, cpu);
1348         }
1349 }
1350
1351 static void restore_cpu_ipis(unsigned int cpu)
1352 {
1353         struct evtchn_bind_ipi bind_ipi;
1354         int ipi, irq, evtchn;
1355
1356         for (ipi = 0; ipi < XEN_NR_IPIS; ipi++) {
1357                 if ((irq = per_cpu(ipi_to_irq, cpu)[ipi]) == -1)
1358                         continue;
1359
1360                 BUG_ON(ipi_from_irq(irq) != ipi);
1361
1362                 /* Get a new binding from Xen. */
1363                 bind_ipi.vcpu = cpu;
1364                 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_ipi,
1365                                                 &bind_ipi) != 0)
1366                         BUG();
1367                 evtchn = bind_ipi.port;
1368
1369                 /* Record the new mapping. */
1370                 (void)xen_irq_info_ipi_setup(cpu, irq, evtchn, ipi);
1371                 bind_evtchn_to_cpu(evtchn, cpu);
1372         }
1373 }
1374
1375 /* Clear an irq's pending state, in preparation for polling on it */
1376 void xen_clear_irq_pending(int irq)
1377 {
1378         int evtchn = evtchn_from_irq(irq);
1379
1380         if (VALID_EVTCHN(evtchn))
1381                 clear_evtchn(evtchn);
1382 }
1383 EXPORT_SYMBOL(xen_clear_irq_pending);
1384 void xen_set_irq_pending(int irq)
1385 {
1386         int evtchn = evtchn_from_irq(irq);
1387
1388         if (VALID_EVTCHN(evtchn))
1389                 set_evtchn(evtchn);
1390 }
1391
1392 bool xen_test_irq_pending(int irq)
1393 {
1394         int evtchn = evtchn_from_irq(irq);
1395         bool ret = false;
1396
1397         if (VALID_EVTCHN(evtchn))
1398                 ret = test_evtchn(evtchn);
1399
1400         return ret;
1401 }
1402
1403 /* Poll waiting for an irq to become pending with timeout.  In the usual case,
1404  * the irq will be disabled so it won't deliver an interrupt. */
1405 void xen_poll_irq_timeout(int irq, u64 timeout)
1406 {
1407         evtchn_port_t evtchn = evtchn_from_irq(irq);
1408
1409         if (VALID_EVTCHN(evtchn)) {
1410                 struct sched_poll poll;
1411
1412                 poll.nr_ports = 1;
1413                 poll.timeout = timeout;
1414                 set_xen_guest_handle(poll.ports, &evtchn);
1415
1416                 if (HYPERVISOR_sched_op(SCHEDOP_poll, &poll) != 0)
1417                         BUG();
1418         }
1419 }
1420 EXPORT_SYMBOL(xen_poll_irq_timeout);
1421 /* Poll waiting for an irq to become pending.  In the usual case, the
1422  * irq will be disabled so it won't deliver an interrupt. */
1423 void xen_poll_irq(int irq)
1424 {
1425         xen_poll_irq_timeout(irq, 0 /* no timeout */);
1426 }
1427
1428 /* Check whether the IRQ line is shared with other guests. */
1429 int xen_test_irq_shared(int irq)
1430 {
1431         struct irq_info *info = info_for_irq(irq);
1432         struct physdev_irq_status_query irq_status;
1433
1434         if (WARN_ON(!info))
1435                 return -ENOENT;
1436
1437         irq_status.irq = info->u.pirq.pirq;
1438
1439         if (HYPERVISOR_physdev_op(PHYSDEVOP_irq_status_query, &irq_status))
1440                 return 0;
1441         return !(irq_status.flags & XENIRQSTAT_shared);
1442 }
1443 EXPORT_SYMBOL_GPL(xen_test_irq_shared);
1444
1445 void xen_irq_resume(void)
1446 {
1447         unsigned int cpu, evtchn;
1448         struct irq_info *info;
1449
1450         /* New event-channel space is not 'live' yet. */
1451         for (evtchn = 0; evtchn < NR_EVENT_CHANNELS; evtchn++)
1452                 mask_evtchn(evtchn);
1453
1454         /* No IRQ <-> event-channel mappings. */
1455         list_for_each_entry(info, &xen_irq_list_head, list)
1456                 info->evtchn = 0; /* zap event-channel binding */
1457
1458         for (evtchn = 0; evtchn < NR_EVENT_CHANNELS; evtchn++)
1459                 evtchn_to_irq[evtchn] = -1;
1460
1461         for_each_possible_cpu(cpu) {
1462                 restore_cpu_virqs(cpu);
1463                 restore_cpu_ipis(cpu);
1464         }
1465
1466         restore_pirqs();
1467 }
1468
1469 static struct irq_chip xen_dynamic_chip __read_mostly = {
1470         .name                   = "xen-dyn",
1471
1472         .irq_disable            = disable_dynirq,
1473         .irq_mask               = disable_dynirq,
1474         .irq_unmask             = enable_dynirq,
1475
1476         .irq_ack                = ack_dynirq,
1477         .irq_mask_ack           = mask_ack_dynirq,
1478
1479         .irq_set_affinity       = set_affinity_irq,
1480         .irq_retrigger          = retrigger_dynirq,
1481 };
1482
1483 static struct irq_chip xen_pirq_chip __read_mostly = {
1484         .name                   = "xen-pirq",
1485
1486         .irq_startup            = startup_pirq,
1487         .irq_shutdown           = shutdown_pirq,
1488         .irq_enable             = enable_pirq,
1489         .irq_disable            = disable_pirq,
1490
1491         .irq_mask               = disable_dynirq,
1492         .irq_unmask             = enable_dynirq,
1493
1494         .irq_ack                = eoi_pirq,
1495         .irq_eoi                = eoi_pirq,
1496         .irq_mask_ack           = mask_ack_pirq,
1497
1498         .irq_set_affinity       = set_affinity_irq,
1499
1500         .irq_retrigger          = retrigger_dynirq,
1501 };
1502
1503 static struct irq_chip xen_percpu_chip __read_mostly = {
1504         .name                   = "xen-percpu",
1505
1506         .irq_disable            = disable_dynirq,
1507         .irq_mask               = disable_dynirq,
1508         .irq_unmask             = enable_dynirq,
1509
1510         .irq_ack                = ack_dynirq,
1511 };
1512
1513 int xen_set_callback_via(uint64_t via)
1514 {
1515         struct xen_hvm_param a;
1516         a.domid = DOMID_SELF;
1517         a.index = HVM_PARAM_CALLBACK_IRQ;
1518         a.value = via;
1519         return HYPERVISOR_hvm_op(HVMOP_set_param, &a);
1520 }
1521 EXPORT_SYMBOL_GPL(xen_set_callback_via);
1522
1523 #ifdef CONFIG_XEN_PVHVM
1524 /* Vector callbacks are better than PCI interrupts to receive event
1525  * channel notifications because we can receive vector callbacks on any
1526  * vcpu and we don't need PCI support or APIC interactions. */
1527 void xen_callback_vector(void)
1528 {
1529         int rc;
1530         uint64_t callback_via;
1531         if (xen_have_vector_callback) {
1532                 callback_via = HVM_CALLBACK_VECTOR(HYPERVISOR_CALLBACK_VECTOR);
1533                 rc = xen_set_callback_via(callback_via);
1534                 if (rc) {
1535                         pr_err("Request for Xen HVM callback vector failed\n");
1536                         xen_have_vector_callback = 0;
1537                         return;
1538                 }
1539                 pr_info("Xen HVM callback vector for event delivery is enabled\n");
1540                 /* in the restore case the vector has already been allocated */
1541                 if (!test_bit(HYPERVISOR_CALLBACK_VECTOR, used_vectors))
1542                         alloc_intr_gate(HYPERVISOR_CALLBACK_VECTOR,
1543                                         xen_hvm_callback_vector);
1544         }
1545 }
1546 #else
1547 void xen_callback_vector(void) {}
1548 #endif
1549
1550 void __init xen_init_IRQ(void)
1551 {
1552         int i;
1553
1554         xen_evtchn_2l_init();
1555
1556         evtchn_to_irq = kcalloc(NR_EVENT_CHANNELS, sizeof(*evtchn_to_irq),
1557                                     GFP_KERNEL);
1558         BUG_ON(!evtchn_to_irq);
1559         for (i = 0; i < NR_EVENT_CHANNELS; i++)
1560                 evtchn_to_irq[i] = -1;
1561
1562         /* No event channels are 'live' right now. */
1563         for (i = 0; i < NR_EVENT_CHANNELS; i++)
1564                 mask_evtchn(i);
1565
1566         pirq_needs_eoi = pirq_needs_eoi_flag;
1567
1568 #ifdef CONFIG_X86
1569         if (xen_hvm_domain()) {
1570                 xen_callback_vector();
1571                 native_init_IRQ();
1572                 /* pci_xen_hvm_init must be called after native_init_IRQ so that
1573                  * __acpi_register_gsi can point at the right function */
1574                 pci_xen_hvm_init();
1575         } else {
1576                 int rc;
1577                 struct physdev_pirq_eoi_gmfn eoi_gmfn;
1578
1579                 irq_ctx_init(smp_processor_id());
1580                 if (xen_initial_domain())
1581                         pci_xen_initial_domain();
1582
1583                 pirq_eoi_map = (void *)__get_free_page(GFP_KERNEL|__GFP_ZERO);
1584                 eoi_gmfn.gmfn = virt_to_mfn(pirq_eoi_map);
1585                 rc = HYPERVISOR_physdev_op(PHYSDEVOP_pirq_eoi_gmfn_v2, &eoi_gmfn);
1586                 if (rc != 0) {
1587                         free_page((unsigned long) pirq_eoi_map);
1588                         pirq_eoi_map = NULL;
1589                 } else
1590                         pirq_needs_eoi = pirq_check_eoi_map;
1591         }
1592 #endif
1593 }