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 #define EVTCHN_ROW(e)  (e / (PAGE_SIZE/sizeof(**evtchn_to_irq)))
  87 #define EVTCHN_COL(e)  (e % (PAGE_SIZE/sizeof(**evtchn_to_irq)))
  88 #define EVTCHN_PER_ROW (PAGE_SIZE / sizeof(**evtchn_to_irq))
  89
  90 /* Xen will never allocate port zero for any purpose. */
  91 #define VALID_EVTCHN(chn)       ((chn) != 0)
  92
  93 static struct irq_chip xen_dynamic_chip;
  94 static struct irq_chip xen_percpu_chip;
  95 static struct irq_chip xen_pirq_chip;
  96 static void enable_dynirq(struct irq_data *data);
  97 static void disable_dynirq(struct irq_data *data);
  98
  99 static void clear_evtchn_to_irq_row(unsigned row)
 100 {
 101         unsigned col;
 102
 103         for (col = 0; col < EVTCHN_PER_ROW; col++)
 104                 evtchn_to_irq[row][col] = -1;
 105 }
 106
 107 static void clear_evtchn_to_irq_all(void)
 108 {
 109         unsigned row;
 110
 111         for (row = 0; row < EVTCHN_ROW(xen_evtchn_max_channels()); row++) {
 112                 if (evtchn_to_irq[row] == NULL)
 113                         continue;
 114                 clear_evtchn_to_irq_row(row);
 115         }
 116 }
 117
 118 static int set_evtchn_to_irq(unsigned evtchn, unsigned irq)
 119 {
 120         unsigned row;
 121         unsigned col;
 122
 123         if (evtchn >= xen_evtchn_max_channels())
 124                 return -EINVAL;
 125
 126         row = EVTCHN_ROW(evtchn);
 127         col = EVTCHN_COL(evtchn);
 128
 129         if (evtchn_to_irq[row] == NULL) {
 130                 /* Unallocated irq entries return -1 anyway */
 131                 if (irq == -1)
 132                         return 0;
 133
 134                 evtchn_to_irq[row] = (int *)get_zeroed_page(GFP_KERNEL);
 135                 if (evtchn_to_irq[row] == NULL)
 136                         return -ENOMEM;
 137
 138                 clear_evtchn_to_irq_row(row);
 139         }
 140
 141         evtchn_to_irq[EVTCHN_ROW(evtchn)][EVTCHN_COL(evtchn)] = irq;
 142         return 0;
 143 }
 144
 145 int get_evtchn_to_irq(unsigned evtchn)
 146 {
 147         if (evtchn >= xen_evtchn_max_channels())
 148                 return -1;
 149         if (evtchn_to_irq[EVTCHN_ROW(evtchn)] == NULL)
 150                 return -1;
 151         return evtchn_to_irq[EVTCHN_ROW(evtchn)][EVTCHN_COL(evtchn)];
 152 }
 153
 154 /* Get info for IRQ */
 155 struct irq_info *info_for_irq(unsigned irq)
 156 {
 157         return irq_get_handler_data(irq);
 158 }
 159
 160 /* Constructors for packed IRQ information. */
 161 static int xen_irq_info_common_setup(struct irq_info *info,
 162                                      unsigned irq,
 163                                      enum xen_irq_type type,
 164                                      unsigned evtchn,
 165                                      unsigned short cpu)
 166 {
 167         int ret;
 168
 169         BUG_ON(info->type != IRQT_UNBOUND && info->type != type);
 170
 171         info->type = type;
 172         info->irq = irq;
 173         info->evtchn = evtchn;
 174         info->cpu = cpu;
 175
 176         ret = set_evtchn_to_irq(evtchn, irq);
 177         if (ret < 0)
 178                 return ret;
 179
 180         irq_clear_status_flags(irq, IRQ_NOREQUEST|IRQ_NOAUTOEN);
 181
 182         return xen_evtchn_port_setup(info);
 183 }
 184
 185 static int xen_irq_info_evtchn_setup(unsigned irq,
 186                                      unsigned evtchn)
 187 {
 188         struct irq_info *info = info_for_irq(irq);
 189
 190         return xen_irq_info_common_setup(info, irq, IRQT_EVTCHN, evtchn, 0);
 191 }
 192
 193 static int xen_irq_info_ipi_setup(unsigned cpu,
 194                                   unsigned irq,
 195                                   unsigned evtchn,
 196                                   enum ipi_vector ipi)
 197 {
 198         struct irq_info *info = info_for_irq(irq);
 199
 200         info->u.ipi = ipi;
 201
 202         per_cpu(ipi_to_irq, cpu)[ipi] = irq;
 203
 204         return xen_irq_info_common_setup(info, irq, IRQT_IPI, evtchn, 0);
 205 }
 206
 207 static int xen_irq_info_virq_setup(unsigned cpu,
 208                                    unsigned irq,
 209                                    unsigned evtchn,
 210                                    unsigned virq)
 211 {
 212         struct irq_info *info = info_for_irq(irq);
 213
 214         info->u.virq = virq;
 215
 216         per_cpu(virq_to_irq, cpu)[virq] = irq;
 217
 218         return xen_irq_info_common_setup(info, irq, IRQT_VIRQ, evtchn, 0);
 219 }
 220
 221 static int xen_irq_info_pirq_setup(unsigned irq,
 222                                    unsigned evtchn,
 223                                    unsigned pirq,
 224                                    unsigned gsi,
 225                                    uint16_t domid,
 226                                    unsigned char flags)
 227 {
 228         struct irq_info *info = info_for_irq(irq);
 229
 230         info->u.pirq.pirq = pirq;
 231         info->u.pirq.gsi = gsi;
 232         info->u.pirq.domid = domid;
 233         info->u.pirq.flags = flags;
 234
 235         return xen_irq_info_common_setup(info, irq, IRQT_PIRQ, evtchn, 0);
 236 }
 237
 238 static void xen_irq_info_cleanup(struct irq_info *info)
 239 {
 240         set_evtchn_to_irq(info->evtchn, -1);
 241         info->evtchn = 0;
 242 }
 243
 244 /*
 245  * Accessors for packed IRQ information.
 246  */
 247 unsigned int evtchn_from_irq(unsigned irq)
 248 {
 249         if (unlikely(WARN(irq < 0 || irq >= nr_irqs, "Invalid irq %d!\n", irq)))
 250                 return 0;
 251
 252         return info_for_irq(irq)->evtchn;
 253 }
 254
 255 unsigned irq_from_evtchn(unsigned int evtchn)
 256 {
 257         return get_evtchn_to_irq(evtchn);
 258 }
 259 EXPORT_SYMBOL_GPL(irq_from_evtchn);
 260
 261 int irq_from_virq(unsigned int cpu, unsigned int virq)
 262 {
 263         return per_cpu(virq_to_irq, cpu)[virq];
 264 }
 265
 266 static enum ipi_vector ipi_from_irq(unsigned irq)
 267 {
 268         struct irq_info *info = info_for_irq(irq);
 269
 270         BUG_ON(info == NULL);
 271         BUG_ON(info->type != IRQT_IPI);
 272
 273         return info->u.ipi;
 274 }
 275
 276 static unsigned virq_from_irq(unsigned irq)
 277 {
 278         struct irq_info *info = info_for_irq(irq);
 279
 280         BUG_ON(info == NULL);
 281         BUG_ON(info->type != IRQT_VIRQ);
 282
 283         return info->u.virq;
 284 }
 285
 286 static unsigned pirq_from_irq(unsigned irq)
 287 {
 288         struct irq_info *info = info_for_irq(irq);
 289
 290         BUG_ON(info == NULL);
 291         BUG_ON(info->type != IRQT_PIRQ);
 292
 293         return info->u.pirq.pirq;
 294 }
 295
 296 static enum xen_irq_type type_from_irq(unsigned irq)
 297 {
 298         return info_for_irq(irq)->type;
 299 }
 300
 301 unsigned cpu_from_irq(unsigned irq)
 302 {
 303         return info_for_irq(irq)->cpu;
 304 }
 305
 306 unsigned int cpu_from_evtchn(unsigned int evtchn)
 307 {
 308         int irq = get_evtchn_to_irq(evtchn);
 309         unsigned ret = 0;
 310
 311         if (irq != -1)
 312                 ret = cpu_from_irq(irq);
 313
 314         return ret;
 315 }
 316
 317 #ifdef CONFIG_X86
 318 static bool pirq_check_eoi_map(unsigned irq)
 319 {
 320         return test_bit(pirq_from_irq(irq), pirq_eoi_map);
 321 }
 322 #endif
 323
 324 static bool pirq_needs_eoi_flag(unsigned irq)
 325 {
 326         struct irq_info *info = info_for_irq(irq);
 327         BUG_ON(info->type != IRQT_PIRQ);
 328
 329         return info->u.pirq.flags & PIRQ_NEEDS_EOI;
 330 }
 331
 332 static void bind_evtchn_to_cpu(unsigned int chn, unsigned int cpu)
 333 {
 334         int irq = get_evtchn_to_irq(chn);
 335         struct irq_info *info = info_for_irq(irq);
 336
 337         BUG_ON(irq == -1);
 338 #ifdef CONFIG_SMP
 339         cpumask_copy(irq_to_desc(irq)->irq_data.affinity, cpumask_of(cpu));
 340 #endif
 341
 342         xen_evtchn_port_bind_to_cpu(info, cpu);
 343
 344         info->cpu = cpu;
 345 }
 346
 347 /**
 348  * notify_remote_via_irq - send event to remote end of event channel via irq
 349  * @irq: irq of event channel to send event to
 350  *
 351  * Unlike notify_remote_via_evtchn(), this is safe to use across
 352  * save/restore. Notifications on a broken connection are silently
 353  * dropped.
 354  */
 355 void notify_remote_via_irq(int irq)
 356 {
 357         int evtchn = evtchn_from_irq(irq);
 358
 359         if (VALID_EVTCHN(evtchn))
 360                 notify_remote_via_evtchn(evtchn);
 361 }
 362 EXPORT_SYMBOL_GPL(notify_remote_via_irq);
 363
 364 static void xen_irq_init(unsigned irq)
 365 {
 366         struct irq_info *info;
 367 #ifdef CONFIG_SMP
 368         struct irq_desc *desc = irq_to_desc(irq);
 369
 370         /* By default all event channels notify CPU#0. */
 371         cpumask_copy(desc->irq_data.affinity, cpumask_of(0));
 372 #endif
 373
 374         info = kzalloc(sizeof(*info), GFP_KERNEL);
 375         if (info == NULL)
 376                 panic("Unable to allocate metadata for IRQ%d\n", irq);
 377
 378         info->type = IRQT_UNBOUND;
 379         info->refcnt = -1;
 380
 381         irq_set_handler_data(irq, info);
 382
 383         list_add_tail(&info->list, &xen_irq_list_head);
 384 }
 385
 386 static int __must_check xen_allocate_irq_dynamic(void)
 387 {
 388         int first = 0;
 389         int irq;
 390
 391 #ifdef CONFIG_X86_IO_APIC
 392         /*
 393          * For an HVM guest or domain 0 which see "real" (emulated or
 394          * actual respectively) GSIs we allocate dynamic IRQs
 395          * e.g. those corresponding to event channels or MSIs
 396          * etc. from the range above those "real" GSIs to avoid
 397          * collisions.
 398          */
 399         if (xen_initial_domain() || xen_hvm_domain())
 400                 first = get_nr_irqs_gsi();
 401 #endif
 402
 403         irq = irq_alloc_desc_from(first, -1);
 404
 405         if (irq >= 0)
 406                 xen_irq_init(irq);
 407
 408         return irq;
 409 }
 410
 411 static int __must_check xen_allocate_irq_gsi(unsigned gsi)
 412 {
 413         int irq;
 414
 415         /*
 416          * A PV guest has no concept of a GSI (since it has no ACPI
 417          * nor access to/knowledge of the physical APICs). Therefore
 418          * all IRQs are dynamically allocated from the entire IRQ
 419          * space.
 420          */
 421         if (xen_pv_domain() && !xen_initial_domain())
 422                 return xen_allocate_irq_dynamic();
 423
 424         /* Legacy IRQ descriptors are already allocated by the arch. */
 425         if (gsi < NR_IRQS_LEGACY)
 426                 irq = gsi;
 427         else
 428                 irq = irq_alloc_desc_at(gsi, -1);
 429
 430         xen_irq_init(irq);
 431
 432         return irq;
 433 }
 434
 435 static void xen_free_irq(unsigned irq)
 436 {
 437         struct irq_info *info = irq_get_handler_data(irq);
 438
 439         if (WARN_ON(!info))
 440                 return;
 441
 442         list_del(&info->list);
 443
 444         irq_set_handler_data(irq, NULL);
 445
 446         WARN_ON(info->refcnt > 0);
 447
 448         kfree(info);
 449
 450         /* Legacy IRQ descriptors are managed by the arch. */
 451         if (irq < NR_IRQS_LEGACY)
 452                 return;
 453
 454         irq_free_desc(irq);
 455 }
 456
 457 static void xen_evtchn_close(unsigned int port)
 458 {
 459         struct evtchn_close close;
 460
 461         close.port = port;
 462         if (HYPERVISOR_event_channel_op(EVTCHNOP_close, &close) != 0)
 463                 BUG();
 464
 465         /* Closed ports are implicitly re-bound to VCPU0. */
 466         bind_evtchn_to_cpu(port, 0);
 467 }
 468
 469 static void pirq_query_unmask(int irq)
 470 {
 471         struct physdev_irq_status_query irq_status;
 472         struct irq_info *info = info_for_irq(irq);
 473
 474         BUG_ON(info->type != IRQT_PIRQ);
 475
 476         irq_status.irq = pirq_from_irq(irq);
 477         if (HYPERVISOR_physdev_op(PHYSDEVOP_irq_status_query, &irq_status))
 478                 irq_status.flags = 0;
 479
 480         info->u.pirq.flags &= ~PIRQ_NEEDS_EOI;
 481         if (irq_status.flags & XENIRQSTAT_needs_eoi)
 482                 info->u.pirq.flags |= PIRQ_NEEDS_EOI;
 483 }
 484
 485 static bool probing_irq(int irq)
 486 {
 487         struct irq_desc *desc = irq_to_desc(irq);
 488
 489         return desc && desc->action == NULL;
 490 }
 491
 492 static void eoi_pirq(struct irq_data *data)
 493 {
 494         int evtchn = evtchn_from_irq(data->irq);
 495         struct physdev_eoi eoi = { .irq = pirq_from_irq(data->irq) };
 496         int rc = 0;
 497
 498         irq_move_irq(data);
 499
 500         if (VALID_EVTCHN(evtchn))
 501                 clear_evtchn(evtchn);
 502
 503         if (pirq_needs_eoi(data->irq)) {
 504                 rc = HYPERVISOR_physdev_op(PHYSDEVOP_eoi, &eoi);
 505                 WARN_ON(rc);
 506         }
 507 }
 508
 509 static void mask_ack_pirq(struct irq_data *data)
 510 {
 511         disable_dynirq(data);
 512         eoi_pirq(data);
 513 }
 514
 515 static unsigned int __startup_pirq(unsigned int irq)
 516 {
 517         struct evtchn_bind_pirq bind_pirq;
 518         struct irq_info *info = info_for_irq(irq);
 519         int evtchn = evtchn_from_irq(irq);
 520         int rc;
 521
 522         BUG_ON(info->type != IRQT_PIRQ);
 523
 524         if (VALID_EVTCHN(evtchn))
 525                 goto out;
 526
 527         bind_pirq.pirq = pirq_from_irq(irq);
 528         /* NB. We are happy to share unless we are probing. */
 529         bind_pirq.flags = info->u.pirq.flags & PIRQ_SHAREABLE ?
 530                                         BIND_PIRQ__WILL_SHARE : 0;
 531         rc = HYPERVISOR_event_channel_op(EVTCHNOP_bind_pirq, &bind_pirq);
 532         if (rc != 0) {
 533                 if (!probing_irq(irq))
 534                         pr_info("Failed to obtain physical IRQ %d\n", irq);
 535                 return 0;
 536         }
 537         evtchn = bind_pirq.port;
 538
 539         pirq_query_unmask(irq);
 540
 541         rc = set_evtchn_to_irq(evtchn, irq);
 542         if (rc != 0) {
 543                 pr_err("irq%d: Failed to set port to irq mapping (%d)\n",
 544                        irq, rc);
 545                 xen_evtchn_close(evtchn);
 546                 return 0;
 547         }
 548         bind_evtchn_to_cpu(evtchn, 0);
 549         info->evtchn = evtchn;
 550
 551 out:
 552         unmask_evtchn(evtchn);
 553         eoi_pirq(irq_get_irq_data(irq));
 554
 555         return 0;
 556 }
 557
 558 static unsigned int startup_pirq(struct irq_data *data)
 559 {
 560         return __startup_pirq(data->irq);
 561 }
 562
 563 static void shutdown_pirq(struct irq_data *data)
 564 {
 565         unsigned int irq = data->irq;
 566         struct irq_info *info = info_for_irq(irq);
 567         unsigned evtchn = evtchn_from_irq(irq);
 568
 569         BUG_ON(info->type != IRQT_PIRQ);
 570
 571         if (!VALID_EVTCHN(evtchn))
 572                 return;
 573
 574         mask_evtchn(evtchn);
 575         xen_evtchn_close(evtchn);
 576         xen_irq_info_cleanup(info);
 577 }
 578
 579 static void enable_pirq(struct irq_data *data)
 580 {
 581         startup_pirq(data);
 582 }
 583
 584 static void disable_pirq(struct irq_data *data)
 585 {
 586         disable_dynirq(data);
 587 }
 588
 589 int xen_irq_from_gsi(unsigned gsi)
 590 {
 591         struct irq_info *info;
 592
 593         list_for_each_entry(info, &xen_irq_list_head, list) {
 594                 if (info->type != IRQT_PIRQ)
 595                         continue;
 596
 597                 if (info->u.pirq.gsi == gsi)
 598                         return info->irq;
 599         }
 600
 601         return -1;
 602 }
 603 EXPORT_SYMBOL_GPL(xen_irq_from_gsi);
 604
 605 static void __unbind_from_irq(unsigned int irq)
 606 {
 607         int evtchn = evtchn_from_irq(irq);
 608         struct irq_info *info = irq_get_handler_data(irq);
 609
 610         if (info->refcnt > 0) {
 611                 info->refcnt--;
 612                 if (info->refcnt != 0)
 613                         return;
 614         }
 615
 616         if (VALID_EVTCHN(evtchn)) {
 617                 unsigned int cpu = cpu_from_irq(irq);
 618
 619                 xen_evtchn_close(evtchn);
 620
 621                 switch (type_from_irq(irq)) {
 622                 case IRQT_VIRQ:
 623                         per_cpu(virq_to_irq, cpu)[virq_from_irq(irq)] = -1;
 624                         break;
 625                 case IRQT_IPI:
 626                         per_cpu(ipi_to_irq, cpu)[ipi_from_irq(irq)] = -1;
 627                         break;
 628                 default:
 629                         break;
 630                 }
 631
 632                 xen_irq_info_cleanup(info);
 633         }
 634
 635         BUG_ON(info_for_irq(irq)->type == IRQT_UNBOUND);
 636
 637         xen_free_irq(irq);
 638 }
 639
 640 /*
 641  * Do not make any assumptions regarding the relationship between the
 642  * IRQ number returned here and the Xen pirq argument.
 643  *
 644  * Note: We don't assign an event channel until the irq actually started
 645  * up.  Return an existing irq if we've already got one for the gsi.
 646  *
 647  * Shareable implies level triggered, not shareable implies edge
 648  * triggered here.
 649  */
 650 int xen_bind_pirq_gsi_to_irq(unsigned gsi,
 651                              unsigned pirq, int shareable, char *name)
 652 {
 653         int irq = -1;
 654         struct physdev_irq irq_op;
 655         int ret;
 656
 657         mutex_lock(&irq_mapping_update_lock);
 658
 659         irq = xen_irq_from_gsi(gsi);
 660         if (irq != -1) {
 661                 pr_info("%s: returning irq %d for gsi %u\n",
 662                         __func__, irq, gsi);
 663                 goto out;
 664         }
 665
 666         irq = xen_allocate_irq_gsi(gsi);
 667         if (irq < 0)
 668                 goto out;
 669
 670         irq_op.irq = irq;
 671         irq_op.vector = 0;
 672
 673         /* Only the privileged domain can do this. For non-priv, the pcifront
 674          * driver provides a PCI bus that does the call to do exactly
 675          * this in the priv domain. */
 676         if (xen_initial_domain() &&
 677             HYPERVISOR_physdev_op(PHYSDEVOP_alloc_irq_vector, &irq_op)) {
 678                 xen_free_irq(irq);
 679                 irq = -ENOSPC;
 680                 goto out;
 681         }
 682
 683         ret = xen_irq_info_pirq_setup(irq, 0, pirq, gsi, DOMID_SELF,
 684                                shareable ? PIRQ_SHAREABLE : 0);
 685         if (ret < 0) {
 686                 __unbind_from_irq(irq);
 687                 irq = ret;
 688                 goto out;
 689         }
 690
 691         pirq_query_unmask(irq);
 692         /* We try to use the handler with the appropriate semantic for the
 693          * type of interrupt: if the interrupt is an edge triggered
 694          * interrupt we use handle_edge_irq.
 695          *
 696          * On the other hand if the interrupt is level triggered we use
 697          * handle_fasteoi_irq like the native code does for this kind of
 698          * interrupts.
 699          *
 700          * Depending on the Xen version, pirq_needs_eoi might return true
 701          * not only for level triggered interrupts but for edge triggered
 702          * interrupts too. In any case Xen always honors the eoi mechanism,
 703          * not injecting any more pirqs of the same kind if the first one
 704          * hasn't received an eoi yet. Therefore using the fasteoi handler
 705          * is the right choice either way.
 706          */
 707         if (shareable)
 708                 irq_set_chip_and_handler_name(irq, &xen_pirq_chip,
 709                                 handle_fasteoi_irq, name);
 710         else
 711                 irq_set_chip_and_handler_name(irq, &xen_pirq_chip,
 712                                 handle_edge_irq, name);
 713
 714 out:
 715         mutex_unlock(&irq_mapping_update_lock);
 716
 717         return irq;
 718 }
 719
 720 #ifdef CONFIG_PCI_MSI
 721 int xen_allocate_pirq_msi(struct pci_dev *dev, struct msi_desc *msidesc)
 722 {
 723         int rc;
 724         struct physdev_get_free_pirq op_get_free_pirq;
 725
 726         op_get_free_pirq.type = MAP_PIRQ_TYPE_MSI;
 727         rc = HYPERVISOR_physdev_op(PHYSDEVOP_get_free_pirq, &op_get_free_pirq);
 728
 729         WARN_ONCE(rc == -ENOSYS,
 730                   "hypervisor does not support the PHYSDEVOP_get_free_pirq interface\n");
 731
 732         return rc ? -1 : op_get_free_pirq.pirq;
 733 }
 734
 735 int xen_bind_pirq_msi_to_irq(struct pci_dev *dev, struct msi_desc *msidesc,
 736                              int pirq, const char *name, domid_t domid)
 737 {
 738         int irq, ret;
 739
 740         mutex_lock(&irq_mapping_update_lock);
 741
 742         irq = xen_allocate_irq_dynamic();
 743         if (irq < 0)
 744                 goto out;
 745
 746         irq_set_chip_and_handler_name(irq, &xen_pirq_chip, handle_edge_irq,
 747                         name);
 748
 749         ret = xen_irq_info_pirq_setup(irq, 0, pirq, 0, domid, 0);
 750         if (ret < 0)
 751                 goto error_irq;
 752         ret = irq_set_msi_desc(irq, msidesc);
 753         if (ret < 0)
 754                 goto error_irq;
 755 out:
 756         mutex_unlock(&irq_mapping_update_lock);
 757         return irq;
 758 error_irq:
 759         __unbind_from_irq(irq);
 760         mutex_unlock(&irq_mapping_update_lock);
 761         return ret;
 762 }
 763 #endif
 764
 765 int xen_destroy_irq(int irq)
 766 {
 767         struct irq_desc *desc;
 768         struct physdev_unmap_pirq unmap_irq;
 769         struct irq_info *info = info_for_irq(irq);
 770         int rc = -ENOENT;
 771
 772         mutex_lock(&irq_mapping_update_lock);
 773
 774         desc = irq_to_desc(irq);
 775         if (!desc)
 776                 goto out;
 777
 778         if (xen_initial_domain()) {
 779                 unmap_irq.pirq = info->u.pirq.pirq;
 780                 unmap_irq.domid = info->u.pirq.domid;
 781                 rc = HYPERVISOR_physdev_op(PHYSDEVOP_unmap_pirq, &unmap_irq);
 782                 /* If another domain quits without making the pci_disable_msix
 783                  * call, the Xen hypervisor takes care of freeing the PIRQs
 784                  * (free_domain_pirqs).
 785                  */
 786                 if ((rc == -ESRCH && info->u.pirq.domid != DOMID_SELF))
 787                         pr_info("domain %d does not have %d anymore\n",
 788                                 info->u.pirq.domid, info->u.pirq.pirq);
 789                 else if (rc) {
 790                         pr_warn("unmap irq failed %d\n", rc);
 791                         goto out;
 792                 }
 793         }
 794
 795         xen_free_irq(irq);
 796
 797 out:
 798         mutex_unlock(&irq_mapping_update_lock);
 799         return rc;
 800 }
 801
 802 int xen_irq_from_pirq(unsigned pirq)
 803 {
 804         int irq;
 805
 806         struct irq_info *info;
 807
 808         mutex_lock(&irq_mapping_update_lock);
 809
 810         list_for_each_entry(info, &xen_irq_list_head, list) {
 811                 if (info->type != IRQT_PIRQ)
 812                         continue;
 813                 irq = info->irq;
 814                 if (info->u.pirq.pirq == pirq)
 815                         goto out;
 816         }
 817         irq = -1;
 818 out:
 819         mutex_unlock(&irq_mapping_update_lock);
 820
 821         return irq;
 822 }
 823
 824
 825 int xen_pirq_from_irq(unsigned irq)
 826 {
 827         return pirq_from_irq(irq);
 828 }
 829 EXPORT_SYMBOL_GPL(xen_pirq_from_irq);
 830
 831 int bind_evtchn_to_irq(unsigned int evtchn)
 832 {
 833         int irq;
 834         int ret;
 835
 836         if (evtchn >= xen_evtchn_max_channels())
 837                 return -ENOMEM;
 838
 839         mutex_lock(&irq_mapping_update_lock);
 840
 841         irq = get_evtchn_to_irq(evtchn);
 842
 843         if (irq == -1) {
 844                 irq = xen_allocate_irq_dynamic();
 845                 if (irq < 0)
 846                         goto out;
 847
 848                 irq_set_chip_and_handler_name(irq, &xen_dynamic_chip,
 849                                               handle_edge_irq, "event");
 850
 851                 ret = xen_irq_info_evtchn_setup(irq, evtchn);
 852                 if (ret < 0) {
 853                         __unbind_from_irq(irq);
 854                         irq = ret;
 855                         goto out;
 856                 }
 857         } else {
 858                 struct irq_info *info = info_for_irq(irq);
 859                 WARN_ON(info == NULL || info->type != IRQT_EVTCHN);
 860         }
 861
 862 out:
 863         mutex_unlock(&irq_mapping_update_lock);
 864
 865         return irq;
 866 }
 867 EXPORT_SYMBOL_GPL(bind_evtchn_to_irq);
 868
 869 static int bind_ipi_to_irq(unsigned int ipi, unsigned int cpu)
 870 {
 871         struct evtchn_bind_ipi bind_ipi;
 872         int evtchn, irq;
 873         int ret;
 874
 875         mutex_lock(&irq_mapping_update_lock);
 876
 877         irq = per_cpu(ipi_to_irq, cpu)[ipi];
 878
 879         if (irq == -1) {
 880                 irq = xen_allocate_irq_dynamic();
 881                 if (irq < 0)
 882                         goto out;
 883
 884                 irq_set_chip_and_handler_name(irq, &xen_percpu_chip,
 885                                               handle_percpu_irq, "ipi");
 886
 887                 bind_ipi.vcpu = cpu;
 888                 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_ipi,
 889                                                 &bind_ipi) != 0)
 890                         BUG();
 891                 evtchn = bind_ipi.port;
 892
 893                 ret = xen_irq_info_ipi_setup(cpu, irq, evtchn, ipi);
 894                 if (ret < 0) {
 895                         __unbind_from_irq(irq);
 896                         irq = ret;
 897                         goto out;
 898                 }
 899                 bind_evtchn_to_cpu(evtchn, cpu);
 900         } else {
 901                 struct irq_info *info = info_for_irq(irq);
 902                 WARN_ON(info == NULL || info->type != IRQT_IPI);
 903         }
 904
 905  out:
 906         mutex_unlock(&irq_mapping_update_lock);
 907         return irq;
 908 }
 909
 910 static int bind_interdomain_evtchn_to_irq(unsigned int remote_domain,
 911                                           unsigned int remote_port)
 912 {
 913         struct evtchn_bind_interdomain bind_interdomain;
 914         int err;
 915
 916         bind_interdomain.remote_dom  = remote_domain;
 917         bind_interdomain.remote_port = remote_port;
 918
 919         err = HYPERVISOR_event_channel_op(EVTCHNOP_bind_interdomain,
 920                                           &bind_interdomain);
 921
 922         return err ? : bind_evtchn_to_irq(bind_interdomain.local_port);
 923 }
 924
 925 static int find_virq(unsigned int virq, unsigned int cpu)
 926 {
 927         struct evtchn_status status;
 928         int port, rc = -ENOENT;
 929
 930         memset(&status, 0, sizeof(status));
 931         for (port = 0; port < xen_evtchn_max_channels(); port++) {
 932                 status.dom = DOMID_SELF;
 933                 status.port = port;
 934                 rc = HYPERVISOR_event_channel_op(EVTCHNOP_status, &status);
 935                 if (rc < 0)
 936                         continue;
 937                 if (status.status != EVTCHNSTAT_virq)
 938                         continue;
 939                 if (status.u.virq == virq && status.vcpu == cpu) {
 940                         rc = port;
 941                         break;
 942                 }
 943         }
 944         return rc;
 945 }
 946
 947 int bind_virq_to_irq(unsigned int virq, unsigned int cpu)
 948 {
 949         struct evtchn_bind_virq bind_virq;
 950         int evtchn, irq, ret;
 951
 952         mutex_lock(&irq_mapping_update_lock);
 953
 954         irq = per_cpu(virq_to_irq, cpu)[virq];
 955
 956         if (irq == -1) {
 957                 irq = xen_allocate_irq_dynamic();
 958                 if (irq < 0)
 959                         goto out;
 960
 961                 irq_set_chip_and_handler_name(irq, &xen_percpu_chip,
 962                                               handle_percpu_irq, "virq");
 963
 964                 bind_virq.virq = virq;
 965                 bind_virq.vcpu = cpu;
 966                 ret = HYPERVISOR_event_channel_op(EVTCHNOP_bind_virq,
 967                                                 &bind_virq);
 968                 if (ret == 0)
 969                         evtchn = bind_virq.port;
 970                 else {
 971                         if (ret == -EEXIST)
 972                                 ret = find_virq(virq, cpu);
 973                         BUG_ON(ret < 0);
 974                         evtchn = ret;
 975                 }
 976
 977                 ret = xen_irq_info_virq_setup(cpu, irq, evtchn, virq);
 978                 if (ret < 0) {
 979                         __unbind_from_irq(irq);
 980                         irq = ret;
 981                         goto out;
 982                 }
 983
 984                 bind_evtchn_to_cpu(evtchn, cpu);
 985         } else {
 986                 struct irq_info *info = info_for_irq(irq);
 987                 WARN_ON(info == NULL || info->type != IRQT_VIRQ);
 988         }
 989
 990 out:
 991         mutex_unlock(&irq_mapping_update_lock);
 992
 993         return irq;
 994 }
 995
 996 static void unbind_from_irq(unsigned int irq)
 997 {
 998         mutex_lock(&irq_mapping_update_lock);
 999         __unbind_from_irq(irq);
1000         mutex_unlock(&irq_mapping_update_lock);
1001 }
1002
1003 int bind_evtchn_to_irqhandler(unsigned int evtchn,
1004                               irq_handler_t handler,
1005                               unsigned long irqflags,
1006                               const char *devname, void *dev_id)
1007 {
1008         int irq, retval;
1009
1010         irq = bind_evtchn_to_irq(evtchn);
1011         if (irq < 0)
1012                 return irq;
1013         retval = request_irq(irq, handler, irqflags, devname, dev_id);
1014         if (retval != 0) {
1015                 unbind_from_irq(irq);
1016                 return retval;
1017         }
1018
1019         return irq;
1020 }
1021 EXPORT_SYMBOL_GPL(bind_evtchn_to_irqhandler);
1022
1023 int bind_interdomain_evtchn_to_irqhandler(unsigned int remote_domain,
1024                                           unsigned int remote_port,
1025                                           irq_handler_t handler,
1026                                           unsigned long irqflags,
1027                                           const char *devname,
1028                                           void *dev_id)
1029 {
1030         int irq, retval;
1031
1032         irq = bind_interdomain_evtchn_to_irq(remote_domain, remote_port);
1033         if (irq < 0)
1034                 return irq;
1035
1036         retval = request_irq(irq, handler, irqflags, devname, dev_id);
1037         if (retval != 0) {
1038                 unbind_from_irq(irq);
1039                 return retval;
1040         }
1041
1042         return irq;
1043 }
1044 EXPORT_SYMBOL_GPL(bind_interdomain_evtchn_to_irqhandler);
1045
1046 int bind_virq_to_irqhandler(unsigned int virq, unsigned int cpu,
1047                             irq_handler_t handler,
1048                             unsigned long irqflags, const char *devname, void *dev_id)
1049 {
1050         int irq, retval;
1051
1052         irq = bind_virq_to_irq(virq, cpu);
1053         if (irq < 0)
1054                 return irq;
1055         retval = request_irq(irq, handler, irqflags, devname, dev_id);
1056         if (retval != 0) {
1057                 unbind_from_irq(irq);
1058                 return retval;
1059         }
1060
1061         return irq;
1062 }
1063 EXPORT_SYMBOL_GPL(bind_virq_to_irqhandler);
1064
1065 int bind_ipi_to_irqhandler(enum ipi_vector ipi,
1066                            unsigned int cpu,
1067                            irq_handler_t handler,
1068                            unsigned long irqflags,
1069                            const char *devname,
1070                            void *dev_id)
1071 {
1072         int irq, retval;
1073
1074         irq = bind_ipi_to_irq(ipi, cpu);
1075         if (irq < 0)
1076                 return irq;
1077
1078         irqflags |= IRQF_NO_SUSPEND | IRQF_FORCE_RESUME | IRQF_EARLY_RESUME;
1079         retval = request_irq(irq, handler, irqflags, devname, dev_id);
1080         if (retval != 0) {
1081                 unbind_from_irq(irq);
1082                 return retval;
1083         }
1084
1085         return irq;
1086 }
1087
1088 void unbind_from_irqhandler(unsigned int irq, void *dev_id)
1089 {
1090         struct irq_info *info = irq_get_handler_data(irq);
1091
1092         if (WARN_ON(!info))
1093                 return;
1094         free_irq(irq, dev_id);
1095         unbind_from_irq(irq);
1096 }
1097 EXPORT_SYMBOL_GPL(unbind_from_irqhandler);
1098
1099 int evtchn_make_refcounted(unsigned int evtchn)
1100 {
1101         int irq = get_evtchn_to_irq(evtchn);
1102         struct irq_info *info;
1103
1104         if (irq == -1)
1105                 return -ENOENT;
1106
1107         info = irq_get_handler_data(irq);
1108
1109         if (!info)
1110                 return -ENOENT;
1111
1112         WARN_ON(info->refcnt != -1);
1113
1114         info->refcnt = 1;
1115
1116         return 0;
1117 }
1118 EXPORT_SYMBOL_GPL(evtchn_make_refcounted);
1119
1120 int evtchn_get(unsigned int evtchn)
1121 {
1122         int irq;
1123         struct irq_info *info;
1124         int err = -ENOENT;
1125
1126         if (evtchn >= xen_evtchn_max_channels())
1127                 return -EINVAL;
1128
1129         mutex_lock(&irq_mapping_update_lock);
1130
1131         irq = get_evtchn_to_irq(evtchn);
1132         if (irq == -1)
1133                 goto done;
1134
1135         info = irq_get_handler_data(irq);
1136
1137         if (!info)
1138                 goto done;
1139
1140         err = -EINVAL;
1141         if (info->refcnt <= 0)
1142                 goto done;
1143
1144         info->refcnt++;
1145         err = 0;
1146  done:
1147         mutex_unlock(&irq_mapping_update_lock);
1148
1149         return err;
1150 }
1151 EXPORT_SYMBOL_GPL(evtchn_get);
1152
1153 void evtchn_put(unsigned int evtchn)
1154 {
1155         int irq = get_evtchn_to_irq(evtchn);
1156         if (WARN_ON(irq == -1))
1157                 return;
1158         unbind_from_irq(irq);
1159 }
1160 EXPORT_SYMBOL_GPL(evtchn_put);
1161
1162 void xen_send_IPI_one(unsigned int cpu, enum ipi_vector vector)
1163 {
1164         int irq;
1165
1166 #ifdef CONFIG_X86
1167         if (unlikely(vector == XEN_NMI_VECTOR)) {
1168                 int rc =  HYPERVISOR_vcpu_op(VCPUOP_send_nmi, cpu, NULL);
1169                 if (rc < 0)
1170                         printk(KERN_WARNING "Sending nmi to CPU%d failed (rc:%d)\n", cpu, rc);
1171                 return;
1172         }
1173 #endif
1174         irq = per_cpu(ipi_to_irq, cpu)[vector];
1175         BUG_ON(irq < 0);
1176         notify_remote_via_irq(irq);
1177 }
1178
1179 static DEFINE_PER_CPU(unsigned, xed_nesting_count);
1180
1181 static void __xen_evtchn_do_upcall(void)
1182 {
1183         struct vcpu_info *vcpu_info = __this_cpu_read(xen_vcpu);
1184         int cpu = get_cpu();
1185         unsigned count;
1186
1187         do {
1188                 vcpu_info->evtchn_upcall_pending = 0;
1189
1190                 if (__this_cpu_inc_return(xed_nesting_count) - 1)
1191                         goto out;
1192
1193                 xen_evtchn_handle_events(cpu);
1194
1195                 BUG_ON(!irqs_disabled());
1196
1197                 count = __this_cpu_read(xed_nesting_count);
1198                 __this_cpu_write(xed_nesting_count, 0);
1199         } while (count != 1 || vcpu_info->evtchn_upcall_pending);
1200
1201 out:
1202
1203         put_cpu();
1204 }
1205
1206 void xen_evtchn_do_upcall(struct pt_regs *regs)
1207 {
1208         struct pt_regs *old_regs = set_irq_regs(regs);
1209
1210         irq_enter();
1211 #ifdef CONFIG_X86
1212         exit_idle();
1213 #endif
1214
1215         __xen_evtchn_do_upcall();
1216
1217         irq_exit();
1218         set_irq_regs(old_regs);
1219 }
1220
1221 void xen_hvm_evtchn_do_upcall(void)
1222 {
1223         __xen_evtchn_do_upcall();
1224 }
1225 EXPORT_SYMBOL_GPL(xen_hvm_evtchn_do_upcall);
1226
1227 /* Rebind a new event channel to an existing irq. */
1228 void rebind_evtchn_irq(int evtchn, int irq)
1229 {
1230         struct irq_info *info = info_for_irq(irq);
1231
1232         if (WARN_ON(!info))
1233                 return;
1234
1235         /* Make sure the irq is masked, since the new event channel
1236            will also be masked. */
1237         disable_irq(irq);
1238
1239         mutex_lock(&irq_mapping_update_lock);
1240
1241         /* After resume the irq<->evtchn mappings are all cleared out */
1242         BUG_ON(get_evtchn_to_irq(evtchn) != -1);
1243         /* Expect irq to have been bound before,
1244            so there should be a proper type */
1245         BUG_ON(info->type == IRQT_UNBOUND);
1246
1247         (void)xen_irq_info_evtchn_setup(irq, evtchn);
1248
1249         mutex_unlock(&irq_mapping_update_lock);
1250
1251         /* new event channels are always bound to cpu 0 */
1252         irq_set_affinity(irq, cpumask_of(0));
1253
1254         /* Unmask the event channel. */
1255         enable_irq(irq);
1256 }
1257
1258 /* Rebind an evtchn so that it gets delivered to a specific cpu */
1259 static int rebind_irq_to_cpu(unsigned irq, unsigned tcpu)
1260 {
1261         struct evtchn_bind_vcpu bind_vcpu;
1262         int evtchn = evtchn_from_irq(irq);
1263         int masked;
1264
1265         if (!VALID_EVTCHN(evtchn))
1266                 return -1;
1267
1268         /*
1269          * Events delivered via platform PCI interrupts are always
1270          * routed to vcpu 0 and hence cannot be rebound.
1271          */
1272         if (xen_hvm_domain() && !xen_have_vector_callback)
1273                 return -1;
1274
1275         /* Send future instances of this interrupt to other vcpu. */
1276         bind_vcpu.port = evtchn;
1277         bind_vcpu.vcpu = tcpu;
1278
1279         /*
1280          * Mask the event while changing the VCPU binding to prevent
1281          * it being delivered on an unexpected VCPU.
1282          */
1283         masked = test_and_set_mask(evtchn);
1284
1285         /*
1286          * If this fails, it usually just indicates that we're dealing with a
1287          * virq or IPI channel, which don't actually need to be rebound. Ignore
1288          * it, but don't do the xenlinux-level rebind in that case.
1289          */
1290         if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_vcpu, &bind_vcpu) >= 0)
1291                 bind_evtchn_to_cpu(evtchn, tcpu);
1292
1293         if (!masked)
1294                 unmask_evtchn(evtchn);
1295
1296         return 0;
1297 }
1298
1299 static int set_affinity_irq(struct irq_data *data, const struct cpumask *dest,
1300                             bool force)
1301 {
1302         unsigned tcpu = cpumask_first(dest);
1303
1304         return rebind_irq_to_cpu(data->irq, tcpu);
1305 }
1306
1307 static int retrigger_evtchn(int evtchn)
1308 {
1309         int masked;
1310
1311         if (!VALID_EVTCHN(evtchn))
1312                 return 0;
1313
1314         masked = test_and_set_mask(evtchn);
1315         set_evtchn(evtchn);
1316         if (!masked)
1317                 unmask_evtchn(evtchn);
1318
1319         return 1;
1320 }
1321
1322 int resend_irq_on_evtchn(unsigned int irq)
1323 {
1324         return retrigger_evtchn(evtchn_from_irq(irq));
1325 }
1326
1327 static void enable_dynirq(struct irq_data *data)
1328 {
1329         int evtchn = evtchn_from_irq(data->irq);
1330
1331         if (VALID_EVTCHN(evtchn))
1332                 unmask_evtchn(evtchn);
1333 }
1334
1335 static void disable_dynirq(struct irq_data *data)
1336 {
1337         int evtchn = evtchn_from_irq(data->irq);
1338
1339         if (VALID_EVTCHN(evtchn))
1340                 mask_evtchn(evtchn);
1341 }
1342
1343 static void ack_dynirq(struct irq_data *data)
1344 {
1345         int evtchn = evtchn_from_irq(data->irq);
1346
1347         irq_move_irq(data);
1348
1349         if (VALID_EVTCHN(evtchn))
1350                 clear_evtchn(evtchn);
1351 }
1352
1353 static void mask_ack_dynirq(struct irq_data *data)
1354 {
1355         disable_dynirq(data);
1356         ack_dynirq(data);
1357 }
1358
1359 static int retrigger_dynirq(struct irq_data *data)
1360 {
1361         return retrigger_evtchn(evtchn_from_irq(data->irq));
1362 }
1363
1364 static void restore_pirqs(void)
1365 {
1366         int pirq, rc, irq, gsi;
1367         struct physdev_map_pirq map_irq;
1368         struct irq_info *info;
1369
1370         list_for_each_entry(info, &xen_irq_list_head, list) {
1371                 if (info->type != IRQT_PIRQ)
1372                         continue;
1373
1374                 pirq = info->u.pirq.pirq;
1375                 gsi = info->u.pirq.gsi;
1376                 irq = info->irq;
1377
1378                 /* save/restore of PT devices doesn't work, so at this point the
1379                  * only devices present are GSI based emulated devices */
1380                 if (!gsi)
1381                         continue;
1382
1383                 map_irq.domid = DOMID_SELF;
1384                 map_irq.type = MAP_PIRQ_TYPE_GSI;
1385                 map_irq.index = gsi;
1386                 map_irq.pirq = pirq;
1387
1388                 rc = HYPERVISOR_physdev_op(PHYSDEVOP_map_pirq, &map_irq);
1389                 if (rc) {
1390                         pr_warn("xen map irq failed gsi=%d irq=%d pirq=%d rc=%d\n",
1391                                 gsi, irq, pirq, rc);
1392                         xen_free_irq(irq);
1393                         continue;
1394                 }
1395
1396                 printk(KERN_DEBUG "xen: --> irq=%d, pirq=%d\n", irq, map_irq.pirq);
1397
1398                 __startup_pirq(irq);
1399         }
1400 }
1401
1402 static void restore_cpu_virqs(unsigned int cpu)
1403 {
1404         struct evtchn_bind_virq bind_virq;
1405         int virq, irq, evtchn;
1406
1407         for (virq = 0; virq < NR_VIRQS; virq++) {
1408                 if ((irq = per_cpu(virq_to_irq, cpu)[virq]) == -1)
1409                         continue;
1410
1411                 BUG_ON(virq_from_irq(irq) != virq);
1412
1413                 /* Get a new binding from Xen. */
1414                 bind_virq.virq = virq;
1415                 bind_virq.vcpu = cpu;
1416                 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_virq,
1417                                                 &bind_virq) != 0)
1418                         BUG();
1419                 evtchn = bind_virq.port;
1420
1421                 /* Record the new mapping. */
1422                 (void)xen_irq_info_virq_setup(cpu, irq, evtchn, virq);
1423                 bind_evtchn_to_cpu(evtchn, cpu);
1424         }
1425 }
1426
1427 static void restore_cpu_ipis(unsigned int cpu)
1428 {
1429         struct evtchn_bind_ipi bind_ipi;
1430         int ipi, irq, evtchn;
1431
1432         for (ipi = 0; ipi < XEN_NR_IPIS; ipi++) {
1433                 if ((irq = per_cpu(ipi_to_irq, cpu)[ipi]) == -1)
1434                         continue;
1435
1436                 BUG_ON(ipi_from_irq(irq) != ipi);
1437
1438                 /* Get a new binding from Xen. */
1439                 bind_ipi.vcpu = cpu;
1440                 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_ipi,
1441                                                 &bind_ipi) != 0)
1442                         BUG();
1443                 evtchn = bind_ipi.port;
1444
1445                 /* Record the new mapping. */
1446                 (void)xen_irq_info_ipi_setup(cpu, irq, evtchn, ipi);
1447                 bind_evtchn_to_cpu(evtchn, cpu);
1448         }
1449 }
1450
1451 /* Clear an irq's pending state, in preparation for polling on it */
1452 void xen_clear_irq_pending(int irq)
1453 {
1454         int evtchn = evtchn_from_irq(irq);
1455
1456         if (VALID_EVTCHN(evtchn))
1457                 clear_evtchn(evtchn);
1458 }
1459 EXPORT_SYMBOL(xen_clear_irq_pending);
1460 void xen_set_irq_pending(int irq)
1461 {
1462         int evtchn = evtchn_from_irq(irq);
1463
1464         if (VALID_EVTCHN(evtchn))
1465                 set_evtchn(evtchn);
1466 }
1467
1468 bool xen_test_irq_pending(int irq)
1469 {
1470         int evtchn = evtchn_from_irq(irq);
1471         bool ret = false;
1472
1473         if (VALID_EVTCHN(evtchn))
1474                 ret = test_evtchn(evtchn);
1475
1476         return ret;
1477 }
1478
1479 /* Poll waiting for an irq to become pending with timeout.  In the usual case,
1480  * the irq will be disabled so it won't deliver an interrupt. */
1481 void xen_poll_irq_timeout(int irq, u64 timeout)
1482 {
1483         evtchn_port_t evtchn = evtchn_from_irq(irq);
1484
1485         if (VALID_EVTCHN(evtchn)) {
1486                 struct sched_poll poll;
1487
1488                 poll.nr_ports = 1;
1489                 poll.timeout = timeout;
1490                 set_xen_guest_handle(poll.ports, &evtchn);
1491
1492                 if (HYPERVISOR_sched_op(SCHEDOP_poll, &poll) != 0)
1493                         BUG();
1494         }
1495 }
1496 EXPORT_SYMBOL(xen_poll_irq_timeout);
1497 /* Poll waiting for an irq to become pending.  In the usual case, the
1498  * irq will be disabled so it won't deliver an interrupt. */
1499 void xen_poll_irq(int irq)
1500 {
1501         xen_poll_irq_timeout(irq, 0 /* no timeout */);
1502 }
1503
1504 /* Check whether the IRQ line is shared with other guests. */
1505 int xen_test_irq_shared(int irq)
1506 {
1507         struct irq_info *info = info_for_irq(irq);
1508         struct physdev_irq_status_query irq_status;
1509
1510         if (WARN_ON(!info))
1511                 return -ENOENT;
1512
1513         irq_status.irq = info->u.pirq.pirq;
1514
1515         if (HYPERVISOR_physdev_op(PHYSDEVOP_irq_status_query, &irq_status))
1516                 return 0;
1517         return !(irq_status.flags & XENIRQSTAT_shared);
1518 }
1519 EXPORT_SYMBOL_GPL(xen_test_irq_shared);
1520
1521 void xen_irq_resume(void)
1522 {
1523         unsigned int cpu, evtchn;
1524         struct irq_info *info;
1525
1526         /* New event-channel space is not 'live' yet. */
1527         for (evtchn = 0; evtchn < xen_evtchn_nr_channels(); evtchn++)
1528                 mask_evtchn(evtchn);
1529
1530         /* No IRQ <-> event-channel mappings. */
1531         list_for_each_entry(info, &xen_irq_list_head, list)
1532                 info->evtchn = 0; /* zap event-channel binding */
1533
1534         clear_evtchn_to_irq_all();
1535
1536         for_each_possible_cpu(cpu) {
1537                 restore_cpu_virqs(cpu);
1538                 restore_cpu_ipis(cpu);
1539         }
1540
1541         restore_pirqs();
1542 }
1543
1544 static struct irq_chip xen_dynamic_chip __read_mostly = {
1545         .name                   = "xen-dyn",
1546
1547         .irq_disable            = disable_dynirq,
1548         .irq_mask               = disable_dynirq,
1549         .irq_unmask             = enable_dynirq,
1550
1551         .irq_ack                = ack_dynirq,
1552         .irq_mask_ack           = mask_ack_dynirq,
1553
1554         .irq_set_affinity       = set_affinity_irq,
1555         .irq_retrigger          = retrigger_dynirq,
1556 };
1557
1558 static struct irq_chip xen_pirq_chip __read_mostly = {
1559         .name                   = "xen-pirq",
1560
1561         .irq_startup            = startup_pirq,
1562         .irq_shutdown           = shutdown_pirq,
1563         .irq_enable             = enable_pirq,
1564         .irq_disable            = disable_pirq,
1565
1566         .irq_mask               = disable_dynirq,
1567         .irq_unmask             = enable_dynirq,
1568
1569         .irq_ack                = eoi_pirq,
1570         .irq_eoi                = eoi_pirq,
1571         .irq_mask_ack           = mask_ack_pirq,
1572
1573         .irq_set_affinity       = set_affinity_irq,
1574
1575         .irq_retrigger          = retrigger_dynirq,
1576 };
1577
1578 static struct irq_chip xen_percpu_chip __read_mostly = {
1579         .name                   = "xen-percpu",
1580
1581         .irq_disable            = disable_dynirq,
1582         .irq_mask               = disable_dynirq,
1583         .irq_unmask             = enable_dynirq,
1584
1585         .irq_ack                = ack_dynirq,
1586 };
1587
1588 int xen_set_callback_via(uint64_t via)
1589 {
1590         struct xen_hvm_param a;
1591         a.domid = DOMID_SELF;
1592         a.index = HVM_PARAM_CALLBACK_IRQ;
1593         a.value = via;
1594         return HYPERVISOR_hvm_op(HVMOP_set_param, &a);
1595 }
1596 EXPORT_SYMBOL_GPL(xen_set_callback_via);
1597
1598 #ifdef CONFIG_XEN_PVHVM
1599 /* Vector callbacks are better than PCI interrupts to receive event
1600  * channel notifications because we can receive vector callbacks on any
1601  * vcpu and we don't need PCI support or APIC interactions. */
1602 void xen_callback_vector(void)
1603 {
1604         int rc;
1605         uint64_t callback_via;
1606         if (xen_have_vector_callback) {
1607                 callback_via = HVM_CALLBACK_VECTOR(HYPERVISOR_CALLBACK_VECTOR);
1608                 rc = xen_set_callback_via(callback_via);
1609                 if (rc) {
1610                         pr_err("Request for Xen HVM callback vector failed\n");
1611                         xen_have_vector_callback = 0;
1612                         return;
1613                 }
1614                 pr_info("Xen HVM callback vector for event delivery is enabled\n");
1615                 /* in the restore case the vector has already been allocated */
1616                 if (!test_bit(HYPERVISOR_CALLBACK_VECTOR, used_vectors))
1617                         alloc_intr_gate(HYPERVISOR_CALLBACK_VECTOR,
1618                                         xen_hvm_callback_vector);
1619         }
1620 }
1621 #else
1622 void xen_callback_vector(void) {}
1623 #endif
1624
1625 void __init xen_init_IRQ(void)
1626 {
1627         int i;
1628
1629         xen_evtchn_2l_init();
1630
1631         evtchn_to_irq = kcalloc(EVTCHN_ROW(xen_evtchn_max_channels()),
1632                                 sizeof(*evtchn_to_irq), GFP_KERNEL);
1633         BUG_ON(!evtchn_to_irq);
1634
1635         /* No event channels are 'live' right now. */
1636         for (i = 0; i < xen_evtchn_nr_channels(); i++)
1637                 mask_evtchn(i);
1638
1639         pirq_needs_eoi = pirq_needs_eoi_flag;
1640
1641 #ifdef CONFIG_X86
1642         if (xen_hvm_domain()) {
1643                 xen_callback_vector();
1644                 native_init_IRQ();
1645                 /* pci_xen_hvm_init must be called after native_init_IRQ so that
1646                  * __acpi_register_gsi can point at the right function */
1647                 pci_xen_hvm_init();
1648         } else {
1649                 int rc;
1650                 struct physdev_pirq_eoi_gmfn eoi_gmfn;
1651
1652                 irq_ctx_init(smp_processor_id());
1653                 if (xen_initial_domain())
1654                         pci_xen_initial_domain();
1655
1656                 pirq_eoi_map = (void *)__get_free_page(GFP_KERNEL|__GFP_ZERO);
1657                 eoi_gmfn.gmfn = virt_to_mfn(pirq_eoi_map);
1658                 rc = HYPERVISOR_physdev_op(PHYSDEVOP_pirq_eoi_gmfn_v2, &eoi_gmfn);
1659                 if (rc != 0) {
1660                         free_page((unsigned long) pirq_eoi_map);
1661                         pirq_eoi_map = NULL;
1662                 } else
1663                         pirq_needs_eoi = pirq_check_eoi_map;
1664         }
1665 #endif
1666 }