kernel/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/i8259.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 #include <xen/page.h>
  49
  50 #include <xen/xen.h>
  51 #include <xen/hvm.h>
  52 #include <xen/xen-ops.h>
  53 #include <xen/events.h>
  54 #include <xen/interface/xen.h>
  55 #include <xen/interface/event_channel.h>
  56 #include <xen/interface/hvm/hvm_op.h>
  57 #include <xen/interface/hvm/params.h>
  58 #include <xen/interface/physdev.h>
  59 #include <xen/interface/sched.h>
  60 #include <xen/interface/vcpu.h>
  61 #include <asm/hw_irq.h>
  62
  63 #include "events_internal.h"
  64
  65 const struct evtchn_ops *evtchn_ops;
  66
  67 /*
  68  * This lock protects updates to the following mapping and reference-count
  69  * arrays. The lock does not need to be acquired to read the mapping tables.
  70  */
  71 static DEFINE_MUTEX(irq_mapping_update_lock);
  72
  73 static LIST_HEAD(xen_irq_list_head);
  74
  75 /* IRQ <-> VIRQ mapping. */
  76 static DEFINE_PER_CPU(int [NR_VIRQS], virq_to_irq) = {[0 ... NR_VIRQS-1] = -1};
  77
  78 /* IRQ <-> IPI mapping */
  79 static DEFINE_PER_CPU(int [XEN_NR_IPIS], ipi_to_irq) = {[0 ... XEN_NR_IPIS-1] = -1};
  80
  81 int **evtchn_to_irq;
  82 #ifdef CONFIG_X86
  83 static unsigned long *pirq_eoi_map;
  84 #endif
  85 static bool (*pirq_needs_eoi)(unsigned irq);
  86
  87 #define EVTCHN_ROW(e)  (e / (PAGE_SIZE/sizeof(**evtchn_to_irq)))
  88 #define EVTCHN_COL(e)  (e % (PAGE_SIZE/sizeof(**evtchn_to_irq)))
  89 #define EVTCHN_PER_ROW (PAGE_SIZE / sizeof(**evtchn_to_irq))
  90
  91 /* Xen will never allocate port zero for any purpose. */
  92 #define VALID_EVTCHN(chn)       ((chn) != 0)
  93
  94 static struct irq_chip xen_dynamic_chip;
  95 static struct irq_chip xen_percpu_chip;
  96 static struct irq_chip xen_pirq_chip;
  97 static void enable_dynirq(struct irq_data *data);
  98 static void disable_dynirq(struct irq_data *data);
  99
 100 static void clear_evtchn_to_irq_row(unsigned row)
 101 {
 102         unsigned col;
 103
 104         for (col = 0; col < EVTCHN_PER_ROW; col++)
 105                 evtchn_to_irq[row][col] = -1;
 106 }
 107
 108 static void clear_evtchn_to_irq_all(void)
 109 {
 110         unsigned row;
 111
 112         for (row = 0; row < EVTCHN_ROW(xen_evtchn_max_channels()); row++) {
 113                 if (evtchn_to_irq[row] == NULL)
 114                         continue;
 115                 clear_evtchn_to_irq_row(row);
 116         }
 117 }
 118
 119 static int set_evtchn_to_irq(unsigned evtchn, unsigned irq)
 120 {
 121         unsigned row;
 122         unsigned col;
 123
 124         if (evtchn >= xen_evtchn_max_channels())
 125                 return -EINVAL;
 126
 127         row = EVTCHN_ROW(evtchn);
 128         col = EVTCHN_COL(evtchn);
 129
 130         if (evtchn_to_irq[row] == NULL) {
 131                 /* Unallocated irq entries return -1 anyway */
 132                 if (irq == -1)
 133                         return 0;
 134
 135                 evtchn_to_irq[row] = (int *)get_zeroed_page(GFP_KERNEL);
 136                 if (evtchn_to_irq[row] == NULL)
 137                         return -ENOMEM;
 138
 139                 clear_evtchn_to_irq_row(row);
 140         }
 141
 142         evtchn_to_irq[EVTCHN_ROW(evtchn)][EVTCHN_COL(evtchn)] = irq;
 143         return 0;
 144 }
 145
 146 int get_evtchn_to_irq(unsigned evtchn)
 147 {
 148         if (evtchn >= xen_evtchn_max_channels())
 149                 return -1;
 150         if (evtchn_to_irq[EVTCHN_ROW(evtchn)] == NULL)
 151                 return -1;
 152         return evtchn_to_irq[EVTCHN_ROW(evtchn)][EVTCHN_COL(evtchn)];
 153 }
 154
 155 /* Get info for IRQ */
 156 struct irq_info *info_for_irq(unsigned irq)
 157 {
 158         return irq_get_handler_data(irq);
 159 }
 160
 161 /* Constructors for packed IRQ information. */
 162 static int xen_irq_info_common_setup(struct irq_info *info,
 163                                      unsigned irq,
 164                                      enum xen_irq_type type,
 165                                      unsigned evtchn,
 166                                      unsigned short cpu)
 167 {
 168         int ret;
 169
 170         BUG_ON(info->type != IRQT_UNBOUND && info->type != type);
 171
 172         info->type = type;
 173         info->irq = irq;
 174         info->evtchn = evtchn;
 175         info->cpu = cpu;
 176
 177         ret = set_evtchn_to_irq(evtchn, irq);
 178         if (ret < 0)
 179                 return ret;
 180
 181         irq_clear_status_flags(irq, IRQ_NOREQUEST|IRQ_NOAUTOEN);
 182
 183         return xen_evtchn_port_setup(info);
 184 }
 185
 186 static int xen_irq_info_evtchn_setup(unsigned irq,
 187                                      unsigned evtchn)
 188 {
 189         struct irq_info *info = info_for_irq(irq);
 190
 191         return xen_irq_info_common_setup(info, irq, IRQT_EVTCHN, evtchn, 0);
 192 }
 193
 194 static int xen_irq_info_ipi_setup(unsigned cpu,
 195                                   unsigned irq,
 196                                   unsigned evtchn,
 197                                   enum ipi_vector ipi)
 198 {
 199         struct irq_info *info = info_for_irq(irq);
 200
 201         info->u.ipi = ipi;
 202
 203         per_cpu(ipi_to_irq, cpu)[ipi] = irq;
 204
 205         return xen_irq_info_common_setup(info, irq, IRQT_IPI, evtchn, 0);
 206 }
 207
 208 static int xen_irq_info_virq_setup(unsigned cpu,
 209                                    unsigned irq,
 210                                    unsigned evtchn,
 211                                    unsigned virq)
 212 {
 213         struct irq_info *info = info_for_irq(irq);
 214
 215         info->u.virq = virq;
 216
 217         per_cpu(virq_to_irq, cpu)[virq] = irq;
 218
 219         return xen_irq_info_common_setup(info, irq, IRQT_VIRQ, evtchn, 0);
 220 }
 221
 222 static int xen_irq_info_pirq_setup(unsigned irq,
 223                                    unsigned evtchn,
 224                                    unsigned pirq,
 225                                    unsigned gsi,
 226                                    uint16_t domid,
 227                                    unsigned char flags)
 228 {
 229         struct irq_info *info = info_for_irq(irq);
 230
 231         info->u.pirq.pirq = pirq;
 232         info->u.pirq.gsi = gsi;
 233         info->u.pirq.domid = domid;
 234         info->u.pirq.flags = flags;
 235
 236         return xen_irq_info_common_setup(info, irq, IRQT_PIRQ, evtchn, 0);
 237 }
 238
 239 static void xen_irq_info_cleanup(struct irq_info *info)
 240 {
 241         set_evtchn_to_irq(info->evtchn, -1);
 242         info->evtchn = 0;
 243 }
 244
 245 /*
 246  * Accessors for packed IRQ information.
 247  */
 248 unsigned int evtchn_from_irq(unsigned irq)
 249 {
 250         if (unlikely(WARN(irq >= nr_irqs, "Invalid irq %d!\n", irq)))
 251                 return 0;
 252
 253         return info_for_irq(irq)->evtchn;
 254 }
 255
 256 unsigned irq_from_evtchn(unsigned int evtchn)
 257 {
 258         return get_evtchn_to_irq(evtchn);
 259 }
 260 EXPORT_SYMBOL_GPL(irq_from_evtchn);
 261
 262 int irq_from_virq(unsigned int cpu, unsigned int virq)
 263 {
 264         return per_cpu(virq_to_irq, cpu)[virq];
 265 }
 266
 267 static enum ipi_vector ipi_from_irq(unsigned irq)
 268 {
 269         struct irq_info *info = info_for_irq(irq);
 270
 271         BUG_ON(info == NULL);
 272         BUG_ON(info->type != IRQT_IPI);
 273
 274         return info->u.ipi;
 275 }
 276
 277 static unsigned virq_from_irq(unsigned irq)
 278 {
 279         struct irq_info *info = info_for_irq(irq);
 280
 281         BUG_ON(info == NULL);
 282         BUG_ON(info->type != IRQT_VIRQ);
 283
 284         return info->u.virq;
 285 }
 286
 287 static unsigned pirq_from_irq(unsigned irq)
 288 {
 289         struct irq_info *info = info_for_irq(irq);
 290
 291         BUG_ON(info == NULL);
 292         BUG_ON(info->type != IRQT_PIRQ);
 293
 294         return info->u.pirq.pirq;
 295 }
 296
 297 static enum xen_irq_type type_from_irq(unsigned irq)
 298 {
 299         return info_for_irq(irq)->type;
 300 }
 301
 302 unsigned cpu_from_irq(unsigned irq)
 303 {
 304         return info_for_irq(irq)->cpu;
 305 }
 306
 307 unsigned int cpu_from_evtchn(unsigned int evtchn)
 308 {
 309         int irq = get_evtchn_to_irq(evtchn);
 310         unsigned ret = 0;
 311
 312         if (irq != -1)
 313                 ret = cpu_from_irq(irq);
 314
 315         return ret;
 316 }
 317
 318 #ifdef CONFIG_X86
 319 static bool pirq_check_eoi_map(unsigned irq)
 320 {
 321         return test_bit(pirq_from_irq(irq), pirq_eoi_map);
 322 }
 323 #endif
 324
 325 static bool pirq_needs_eoi_flag(unsigned irq)
 326 {
 327         struct irq_info *info = info_for_irq(irq);
 328         BUG_ON(info->type != IRQT_PIRQ);
 329
 330         return info->u.pirq.flags & PIRQ_NEEDS_EOI;
 331 }
 332
 333 static void bind_evtchn_to_cpu(unsigned int chn, unsigned int cpu)
 334 {
 335         int irq = get_evtchn_to_irq(chn);
 336         struct irq_info *info = info_for_irq(irq);
 337
 338         BUG_ON(irq == -1);
 339 #ifdef CONFIG_SMP
 340         cpumask_copy(irq_get_affinity_mask(irq), cpumask_of(cpu));
 341 #endif
 342         xen_evtchn_port_bind_to_cpu(info, cpu);
 343
 344         info->cpu = cpu;
 345 }
 346
 347 static void xen_evtchn_mask_all(void)
 348 {
 349         unsigned int evtchn;
 350
 351         for (evtchn = 0; evtchn < xen_evtchn_nr_channels(); evtchn++)
 352                 mask_evtchn(evtchn);
 353 }
 354
 355 /**
 356  * notify_remote_via_irq - send event to remote end of event channel via irq
 357  * @irq: irq of event channel to send event to
 358  *
 359  * Unlike notify_remote_via_evtchn(), this is safe to use across
 360  * save/restore. Notifications on a broken connection are silently
 361  * dropped.
 362  */
 363 void notify_remote_via_irq(int irq)
 364 {
 365         int evtchn = evtchn_from_irq(irq);
 366
 367         if (VALID_EVTCHN(evtchn))
 368                 notify_remote_via_evtchn(evtchn);
 369 }
 370 EXPORT_SYMBOL_GPL(notify_remote_via_irq);
 371
 372 static void xen_irq_init(unsigned irq)
 373 {
 374         struct irq_info *info;
 375 #ifdef CONFIG_SMP
 376         /* By default all event channels notify CPU#0. */
 377         cpumask_copy(irq_get_affinity_mask(irq), cpumask_of(0));
 378 #endif
 379
 380         info = kzalloc(sizeof(*info), GFP_KERNEL);
 381         if (info == NULL)
 382                 panic("Unable to allocate metadata for IRQ%d\n", irq);
 383
 384         info->type = IRQT_UNBOUND;
 385         info->refcnt = -1;
 386
 387         irq_set_handler_data(irq, info);
 388
 389         list_add_tail(&info->list, &xen_irq_list_head);
 390 }
 391
 392 static int __must_check xen_allocate_irqs_dynamic(int nvec)
 393 {
 394         int i, irq = irq_alloc_descs(-1, 0, nvec, -1);
 395
 396         if (irq >= 0) {
 397                 for (i = 0; i < nvec; i++)
 398                         xen_irq_init(irq + i);
 399         }
 400
 401         return irq;
 402 }
 403
 404 static inline int __must_check xen_allocate_irq_dynamic(void)
 405 {
 406
 407         return xen_allocate_irqs_dynamic(1);
 408 }
 409
 410 static int __must_check xen_allocate_irq_gsi(unsigned gsi)
 411 {
 412         int irq;
 413
 414         /*
 415          * A PV guest has no concept of a GSI (since it has no ACPI
 416          * nor access to/knowledge of the physical APICs). Therefore
 417          * all IRQs are dynamically allocated from the entire IRQ
 418          * space.
 419          */
 420         if (xen_pv_domain() && !xen_initial_domain())
 421                 return xen_allocate_irq_dynamic();
 422
 423         /* Legacy IRQ descriptors are already allocated by the arch. */
 424         if (gsi < nr_legacy_irqs())
 425                 irq = gsi;
 426         else
 427                 irq = irq_alloc_desc_at(gsi, -1);
 428
 429         xen_irq_init(irq);
 430
 431         return irq;
 432 }
 433
 434 static void xen_free_irq(unsigned irq)
 435 {
 436         struct irq_info *info = irq_get_handler_data(irq);
 437
 438         if (WARN_ON(!info))
 439                 return;
 440
 441         list_del(&info->list);
 442
 443         irq_set_handler_data(irq, NULL);
 444
 445         WARN_ON(info->refcnt > 0);
 446
 447         kfree(info);
 448
 449         /* Legacy IRQ descriptors are managed by the arch. */
 450         if (irq < nr_legacy_irqs())
 451                 return;
 452
 453         irq_free_desc(irq);
 454 }
 455
 456 static void xen_evtchn_close(unsigned int port)
 457 {
 458         struct evtchn_close close;
 459
 460         close.port = port;
 461         if (HYPERVISOR_event_channel_op(EVTCHNOP_close, &close) != 0)
 462                 BUG();
 463 }
 464
 465 static void pirq_query_unmask(int irq)
 466 {
 467         struct physdev_irq_status_query irq_status;
 468         struct irq_info *info = info_for_irq(irq);
 469
 470         BUG_ON(info->type != IRQT_PIRQ);
 471
 472         irq_status.irq = pirq_from_irq(irq);
 473         if (HYPERVISOR_physdev_op(PHYSDEVOP_irq_status_query, &irq_status))
 474                 irq_status.flags = 0;
 475
 476         info->u.pirq.flags &= ~PIRQ_NEEDS_EOI;
 477         if (irq_status.flags & XENIRQSTAT_needs_eoi)
 478                 info->u.pirq.flags |= PIRQ_NEEDS_EOI;
 479 }
 480
 481 static void eoi_pirq(struct irq_data *data)
 482 {
 483         int evtchn = evtchn_from_irq(data->irq);
 484         struct physdev_eoi eoi = { .irq = pirq_from_irq(data->irq) };
 485         int rc = 0;
 486
 487         irq_move_irq(data);
 488
 489         if (VALID_EVTCHN(evtchn))
 490                 clear_evtchn(evtchn);
 491
 492         if (pirq_needs_eoi(data->irq)) {
 493                 rc = HYPERVISOR_physdev_op(PHYSDEVOP_eoi, &eoi);
 494                 WARN_ON(rc);
 495         }
 496 }
 497
 498 static void mask_ack_pirq(struct irq_data *data)
 499 {
 500         disable_dynirq(data);
 501         eoi_pirq(data);
 502 }
 503
 504 static unsigned int __startup_pirq(unsigned int irq)
 505 {
 506         struct evtchn_bind_pirq bind_pirq;
 507         struct irq_info *info = info_for_irq(irq);
 508         int evtchn = evtchn_from_irq(irq);
 509         int rc;
 510
 511         BUG_ON(info->type != IRQT_PIRQ);
 512
 513         if (VALID_EVTCHN(evtchn))
 514                 goto out;
 515
 516         bind_pirq.pirq = pirq_from_irq(irq);
 517         /* NB. We are happy to share unless we are probing. */
 518         bind_pirq.flags = info->u.pirq.flags & PIRQ_SHAREABLE ?
 519                                         BIND_PIRQ__WILL_SHARE : 0;
 520         rc = HYPERVISOR_event_channel_op(EVTCHNOP_bind_pirq, &bind_pirq);
 521         if (rc != 0) {
 522                 pr_warn("Failed to obtain physical IRQ %d\n", irq);
 523                 return 0;
 524         }
 525         evtchn = bind_pirq.port;
 526
 527         pirq_query_unmask(irq);
 528
 529         rc = set_evtchn_to_irq(evtchn, irq);
 530         if (rc)
 531                 goto err;
 532
 533         info->evtchn = evtchn;
 534         bind_evtchn_to_cpu(evtchn, 0);
 535
 536         rc = xen_evtchn_port_setup(info);
 537         if (rc)
 538                 goto err;
 539
 540 out:
 541         unmask_evtchn(evtchn);
 542         eoi_pirq(irq_get_irq_data(irq));
 543
 544         return 0;
 545
 546 err:
 547         pr_err("irq%d: Failed to set port to irq mapping (%d)\n", irq, rc);
 548         xen_evtchn_close(evtchn);
 549         return 0;
 550 }
 551
 552 static unsigned int startup_pirq(struct irq_data *data)
 553 {
 554         return __startup_pirq(data->irq);
 555 }
 556
 557 static void shutdown_pirq(struct irq_data *data)
 558 {
 559         unsigned int irq = data->irq;
 560         struct irq_info *info = info_for_irq(irq);
 561         unsigned evtchn = evtchn_from_irq(irq);
 562
 563         BUG_ON(info->type != IRQT_PIRQ);
 564
 565         if (!VALID_EVTCHN(evtchn))
 566                 return;
 567
 568         mask_evtchn(evtchn);
 569         xen_evtchn_close(evtchn);
 570         xen_irq_info_cleanup(info);
 571 }
 572
 573 static void enable_pirq(struct irq_data *data)
 574 {
 575         startup_pirq(data);
 576 }
 577
 578 static void disable_pirq(struct irq_data *data)
 579 {
 580         disable_dynirq(data);
 581 }
 582
 583 int xen_irq_from_gsi(unsigned gsi)
 584 {
 585         struct irq_info *info;
 586
 587         list_for_each_entry(info, &xen_irq_list_head, list) {
 588                 if (info->type != IRQT_PIRQ)
 589                         continue;
 590
 591                 if (info->u.pirq.gsi == gsi)
 592                         return info->irq;
 593         }
 594
 595         return -1;
 596 }
 597 EXPORT_SYMBOL_GPL(xen_irq_from_gsi);
 598
 599 static void __unbind_from_irq(unsigned int irq)
 600 {
 601         int evtchn = evtchn_from_irq(irq);
 602         struct irq_info *info = irq_get_handler_data(irq);
 603
 604         if (info->refcnt > 0) {
 605                 info->refcnt--;
 606                 if (info->refcnt != 0)
 607                         return;
 608         }
 609
 610         if (VALID_EVTCHN(evtchn)) {
 611                 unsigned int cpu = cpu_from_irq(irq);
 612
 613                 xen_evtchn_close(evtchn);
 614
 615                 switch (type_from_irq(irq)) {
 616                 case IRQT_VIRQ:
 617                         per_cpu(virq_to_irq, cpu)[virq_from_irq(irq)] = -1;
 618                         break;
 619                 case IRQT_IPI:
 620                         per_cpu(ipi_to_irq, cpu)[ipi_from_irq(irq)] = -1;
 621                         break;
 622                 default:
 623                         break;
 624                 }
 625
 626                 xen_irq_info_cleanup(info);
 627         }
 628
 629         BUG_ON(info_for_irq(irq)->type == IRQT_UNBOUND);
 630
 631         xen_free_irq(irq);
 632 }
 633
 634 /*
 635  * Do not make any assumptions regarding the relationship between the
 636  * IRQ number returned here and the Xen pirq argument.
 637  *
 638  * Note: We don't assign an event channel until the irq actually started
 639  * up.  Return an existing irq if we've already got one for the gsi.
 640  *
 641  * Shareable implies level triggered, not shareable implies edge
 642  * triggered here.
 643  */
 644 int xen_bind_pirq_gsi_to_irq(unsigned gsi,
 645                              unsigned pirq, int shareable, char *name)
 646 {
 647         int irq = -1;
 648         struct physdev_irq irq_op;
 649         int ret;
 650
 651         mutex_lock(&irq_mapping_update_lock);
 652
 653         irq = xen_irq_from_gsi(gsi);
 654         if (irq != -1) {
 655                 pr_info("%s: returning irq %d for gsi %u\n",
 656                         __func__, irq, gsi);
 657                 goto out;
 658         }
 659
 660         irq = xen_allocate_irq_gsi(gsi);
 661         if (irq < 0)
 662                 goto out;
 663
 664         irq_op.irq = irq;
 665         irq_op.vector = 0;
 666
 667         /* Only the privileged domain can do this. For non-priv, the pcifront
 668          * driver provides a PCI bus that does the call to do exactly
 669          * this in the priv domain. */
 670         if (xen_initial_domain() &&
 671             HYPERVISOR_physdev_op(PHYSDEVOP_alloc_irq_vector, &irq_op)) {
 672                 xen_free_irq(irq);
 673                 irq = -ENOSPC;
 674                 goto out;
 675         }
 676
 677         ret = xen_irq_info_pirq_setup(irq, 0, pirq, gsi, DOMID_SELF,
 678                                shareable ? PIRQ_SHAREABLE : 0);
 679         if (ret < 0) {
 680                 __unbind_from_irq(irq);
 681                 irq = ret;
 682                 goto out;
 683         }
 684
 685         pirq_query_unmask(irq);
 686         /* We try to use the handler with the appropriate semantic for the
 687          * type of interrupt: if the interrupt is an edge triggered
 688          * interrupt we use handle_edge_irq.
 689          *
 690          * On the other hand if the interrupt is level triggered we use
 691          * handle_fasteoi_irq like the native code does for this kind of
 692          * interrupts.
 693          *
 694          * Depending on the Xen version, pirq_needs_eoi might return true
 695          * not only for level triggered interrupts but for edge triggered
 696          * interrupts too. In any case Xen always honors the eoi mechanism,
 697          * not injecting any more pirqs of the same kind if the first one
 698          * hasn't received an eoi yet. Therefore using the fasteoi handler
 699          * is the right choice either way.
 700          */
 701         if (shareable)
 702                 irq_set_chip_and_handler_name(irq, &xen_pirq_chip,
 703                                 handle_fasteoi_irq, name);
 704         else
 705                 irq_set_chip_and_handler_name(irq, &xen_pirq_chip,
 706                                 handle_edge_irq, name);
 707
 708 out:
 709         mutex_unlock(&irq_mapping_update_lock);
 710
 711         return irq;
 712 }
 713
 714 #ifdef CONFIG_PCI_MSI
 715 int xen_allocate_pirq_msi(struct pci_dev *dev, struct msi_desc *msidesc)
 716 {
 717         int rc;
 718         struct physdev_get_free_pirq op_get_free_pirq;
 719
 720         op_get_free_pirq.type = MAP_PIRQ_TYPE_MSI;
 721         rc = HYPERVISOR_physdev_op(PHYSDEVOP_get_free_pirq, &op_get_free_pirq);
 722
 723         WARN_ONCE(rc == -ENOSYS,
 724                   "hypervisor does not support the PHYSDEVOP_get_free_pirq interface\n");
 725
 726         return rc ? -1 : op_get_free_pirq.pirq;
 727 }
 728
 729 int xen_bind_pirq_msi_to_irq(struct pci_dev *dev, struct msi_desc *msidesc,
 730                              int pirq, int nvec, const char *name, domid_t domid)
 731 {
 732         int i, irq, ret;
 733
 734         mutex_lock(&irq_mapping_update_lock);
 735
 736         irq = xen_allocate_irqs_dynamic(nvec);
 737         if (irq < 0)
 738                 goto out;
 739
 740         for (i = 0; i < nvec; i++) {
 741                 irq_set_chip_and_handler_name(irq + i, &xen_pirq_chip, handle_edge_irq, name);
 742
 743                 ret = xen_irq_info_pirq_setup(irq + i, 0, pirq + i, 0, domid,
 744                                               i == 0 ? 0 : PIRQ_MSI_GROUP);
 745                 if (ret < 0)
 746                         goto error_irq;
 747         }
 748
 749         ret = irq_set_msi_desc(irq, msidesc);
 750         if (ret < 0)
 751                 goto error_irq;
 752 out:
 753         mutex_unlock(&irq_mapping_update_lock);
 754         return irq;
 755 error_irq:
 756         for (; i >= 0; i--)
 757                 __unbind_from_irq(irq + i);
 758         mutex_unlock(&irq_mapping_update_lock);
 759         return ret;
 760 }
 761 #endif
 762
 763 int xen_destroy_irq(int irq)
 764 {
 765         struct physdev_unmap_pirq unmap_irq;
 766         struct irq_info *info = info_for_irq(irq);
 767         int rc = -ENOENT;
 768
 769         mutex_lock(&irq_mapping_update_lock);
 770
 771         /*
 772          * If trying to remove a vector in a MSI group different
 773          * than the first one skip the PIRQ unmap unless this vector
 774          * is the first one in the group.
 775          */
 776         if (xen_initial_domain() && !(info->u.pirq.flags & PIRQ_MSI_GROUP)) {
 777                 unmap_irq.pirq = info->u.pirq.pirq;
 778                 unmap_irq.domid = info->u.pirq.domid;
 779                 rc = HYPERVISOR_physdev_op(PHYSDEVOP_unmap_pirq, &unmap_irq);
 780                 /* If another domain quits without making the pci_disable_msix
 781                  * call, the Xen hypervisor takes care of freeing the PIRQs
 782                  * (free_domain_pirqs).
 783                  */
 784                 if ((rc == -ESRCH && info->u.pirq.domid != DOMID_SELF))
 785                         pr_info("domain %d does not have %d anymore\n",
 786                                 info->u.pirq.domid, info->u.pirq.pirq);
 787                 else if (rc) {
 788                         pr_warn("unmap irq failed %d\n", rc);
 789                         goto out;
 790                 }
 791         }
 792
 793         xen_free_irq(irq);
 794
 795 out:
 796         mutex_unlock(&irq_mapping_update_lock);
 797         return rc;
 798 }
 799
 800 int xen_irq_from_pirq(unsigned pirq)
 801 {
 802         int irq;
 803
 804         struct irq_info *info;
 805
 806         mutex_lock(&irq_mapping_update_lock);
 807
 808         list_for_each_entry(info, &xen_irq_list_head, list) {
 809                 if (info->type != IRQT_PIRQ)
 810                         continue;
 811                 irq = info->irq;
 812                 if (info->u.pirq.pirq == pirq)
 813                         goto out;
 814         }
 815         irq = -1;
 816 out:
 817         mutex_unlock(&irq_mapping_update_lock);
 818
 819         return irq;
 820 }
 821
 822
 823 int xen_pirq_from_irq(unsigned irq)
 824 {
 825         return pirq_from_irq(irq);
 826 }
 827 EXPORT_SYMBOL_GPL(xen_pirq_from_irq);
 828
 829 int bind_evtchn_to_irq(unsigned int evtchn)
 830 {
 831         int irq;
 832         int ret;
 833
 834         if (evtchn >= xen_evtchn_max_channels())
 835                 return -ENOMEM;
 836
 837         mutex_lock(&irq_mapping_update_lock);
 838
 839         irq = get_evtchn_to_irq(evtchn);
 840
 841         if (irq == -1) {
 842                 irq = xen_allocate_irq_dynamic();
 843                 if (irq < 0)
 844                         goto out;
 845
 846                 irq_set_chip_and_handler_name(irq, &xen_dynamic_chip,
 847                                               handle_edge_irq, "event");
 848
 849                 ret = xen_irq_info_evtchn_setup(irq, evtchn);
 850                 if (ret < 0) {
 851                         __unbind_from_irq(irq);
 852                         irq = ret;
 853                         goto out;
 854                 }
 855                 /* New interdomain events are bound to VCPU 0. */
 856                 bind_evtchn_to_cpu(evtchn, 0);
 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 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 EXPORT_SYMBOL_GPL(bind_interdomain_evtchn_to_irq);
 925
 926 static int find_virq(unsigned int virq, unsigned int cpu)
 927 {
 928         struct evtchn_status status;
 929         int port, rc = -ENOENT;
 930
 931         memset(&status, 0, sizeof(status));
 932         for (port = 0; port < xen_evtchn_max_channels(); port++) {
 933                 status.dom = DOMID_SELF;
 934                 status.port = port;
 935                 rc = HYPERVISOR_event_channel_op(EVTCHNOP_status, &status);
 936                 if (rc < 0)
 937                         continue;
 938                 if (status.status != EVTCHNSTAT_virq)
 939                         continue;
 940                 if (status.u.virq == virq && status.vcpu == cpu) {
 941                         rc = port;
 942                         break;
 943                 }
 944         }
 945         return rc;
 946 }
 947
 948 /**
 949  * xen_evtchn_nr_channels - number of usable event channel ports
 950  *
 951  * This may be less than the maximum supported by the current
 952  * hypervisor ABI. Use xen_evtchn_max_channels() for the maximum
 953  * supported.
 954  */
 955 unsigned xen_evtchn_nr_channels(void)
 956 {
 957         return evtchn_ops->nr_channels();
 958 }
 959 EXPORT_SYMBOL_GPL(xen_evtchn_nr_channels);
 960
 961 int bind_virq_to_irq(unsigned int virq, unsigned int cpu, bool percpu)
 962 {
 963         struct evtchn_bind_virq bind_virq;
 964         int evtchn, irq, ret;
 965
 966         mutex_lock(&irq_mapping_update_lock);
 967
 968         irq = per_cpu(virq_to_irq, cpu)[virq];
 969
 970         if (irq == -1) {
 971                 irq = xen_allocate_irq_dynamic();
 972                 if (irq < 0)
 973                         goto out;
 974
 975                 if (percpu)
 976                         irq_set_chip_and_handler_name(irq, &xen_percpu_chip,
 977                                                       handle_percpu_irq, "virq");
 978                 else
 979                         irq_set_chip_and_handler_name(irq, &xen_dynamic_chip,
 980                                                       handle_edge_irq, "virq");
 981
 982                 bind_virq.virq = virq;
 983                 bind_virq.vcpu = cpu;
 984                 ret = HYPERVISOR_event_channel_op(EVTCHNOP_bind_virq,
 985                                                 &bind_virq);
 986                 if (ret == 0)
 987                         evtchn = bind_virq.port;
 988                 else {
 989                         if (ret == -EEXIST)
 990                                 ret = find_virq(virq, cpu);
 991                         BUG_ON(ret < 0);
 992                         evtchn = ret;
 993                 }
 994
 995                 ret = xen_irq_info_virq_setup(cpu, irq, evtchn, virq);
 996                 if (ret < 0) {
 997                         __unbind_from_irq(irq);
 998                         irq = ret;
 999                         goto out;
1000                 }
1001
1002                 bind_evtchn_to_cpu(evtchn, cpu);
1003         } else {
1004                 struct irq_info *info = info_for_irq(irq);
1005                 WARN_ON(info == NULL || info->type != IRQT_VIRQ);
1006         }
1007
1008 out:
1009         mutex_unlock(&irq_mapping_update_lock);
1010
1011         return irq;
1012 }
1013
1014 static void unbind_from_irq(unsigned int irq)
1015 {
1016         mutex_lock(&irq_mapping_update_lock);
1017         __unbind_from_irq(irq);
1018         mutex_unlock(&irq_mapping_update_lock);
1019 }
1020
1021 int bind_evtchn_to_irqhandler(unsigned int evtchn,
1022                               irq_handler_t handler,
1023                               unsigned long irqflags,
1024                               const char *devname, void *dev_id)
1025 {
1026         int irq, retval;
1027
1028         irq = bind_evtchn_to_irq(evtchn);
1029         if (irq < 0)
1030                 return irq;
1031         retval = request_irq(irq, handler, irqflags, devname, dev_id);
1032         if (retval != 0) {
1033                 unbind_from_irq(irq);
1034                 return retval;
1035         }
1036
1037         return irq;
1038 }
1039 EXPORT_SYMBOL_GPL(bind_evtchn_to_irqhandler);
1040
1041 int bind_interdomain_evtchn_to_irqhandler(unsigned int remote_domain,
1042                                           unsigned int remote_port,
1043                                           irq_handler_t handler,
1044                                           unsigned long irqflags,
1045                                           const char *devname,
1046                                           void *dev_id)
1047 {
1048         int irq, retval;
1049
1050         irq = bind_interdomain_evtchn_to_irq(remote_domain, remote_port);
1051         if (irq < 0)
1052                 return irq;
1053
1054         retval = request_irq(irq, handler, irqflags, devname, dev_id);
1055         if (retval != 0) {
1056                 unbind_from_irq(irq);
1057                 return retval;
1058         }
1059
1060         return irq;
1061 }
1062 EXPORT_SYMBOL_GPL(bind_interdomain_evtchn_to_irqhandler);
1063
1064 int bind_virq_to_irqhandler(unsigned int virq, unsigned int cpu,
1065                             irq_handler_t handler,
1066                             unsigned long irqflags, const char *devname, void *dev_id)
1067 {
1068         int irq, retval;
1069
1070         irq = bind_virq_to_irq(virq, cpu, irqflags & IRQF_PERCPU);
1071         if (irq < 0)
1072                 return irq;
1073         retval = request_irq(irq, handler, irqflags, devname, dev_id);
1074         if (retval != 0) {
1075                 unbind_from_irq(irq);
1076                 return retval;
1077         }
1078
1079         return irq;
1080 }
1081 EXPORT_SYMBOL_GPL(bind_virq_to_irqhandler);
1082
1083 int bind_ipi_to_irqhandler(enum ipi_vector ipi,
1084                            unsigned int cpu,
1085                            irq_handler_t handler,
1086                            unsigned long irqflags,
1087                            const char *devname,
1088                            void *dev_id)
1089 {
1090         int irq, retval;
1091
1092         irq = bind_ipi_to_irq(ipi, cpu);
1093         if (irq < 0)
1094                 return irq;
1095
1096         irqflags |= IRQF_NO_SUSPEND | IRQF_FORCE_RESUME | IRQF_EARLY_RESUME;
1097         retval = request_irq(irq, handler, irqflags, devname, dev_id);
1098         if (retval != 0) {
1099                 unbind_from_irq(irq);
1100                 return retval;
1101         }
1102
1103         return irq;
1104 }
1105
1106 void unbind_from_irqhandler(unsigned int irq, void *dev_id)
1107 {
1108         struct irq_info *info = irq_get_handler_data(irq);
1109
1110         if (WARN_ON(!info))
1111                 return;
1112         free_irq(irq, dev_id);
1113         unbind_from_irq(irq);
1114 }
1115 EXPORT_SYMBOL_GPL(unbind_from_irqhandler);
1116
1117 /**
1118  * xen_set_irq_priority() - set an event channel priority.
1119  * @irq:irq bound to an event channel.
1120  * @priority: priority between XEN_IRQ_PRIORITY_MAX and XEN_IRQ_PRIORITY_MIN.
1121  */
1122 int xen_set_irq_priority(unsigned irq, unsigned priority)
1123 {
1124         struct evtchn_set_priority set_priority;
1125
1126         set_priority.port = evtchn_from_irq(irq);
1127         set_priority.priority = priority;
1128
1129         return HYPERVISOR_event_channel_op(EVTCHNOP_set_priority,
1130                                            &set_priority);
1131 }
1132 EXPORT_SYMBOL_GPL(xen_set_irq_priority);
1133
1134 int evtchn_make_refcounted(unsigned int evtchn)
1135 {
1136         int irq = get_evtchn_to_irq(evtchn);
1137         struct irq_info *info;
1138
1139         if (irq == -1)
1140                 return -ENOENT;
1141
1142         info = irq_get_handler_data(irq);
1143
1144         if (!info)
1145                 return -ENOENT;
1146
1147         WARN_ON(info->refcnt != -1);
1148
1149         info->refcnt = 1;
1150
1151         return 0;
1152 }
1153 EXPORT_SYMBOL_GPL(evtchn_make_refcounted);
1154
1155 int evtchn_get(unsigned int evtchn)
1156 {
1157         int irq;
1158         struct irq_info *info;
1159         int err = -ENOENT;
1160
1161         if (evtchn >= xen_evtchn_max_channels())
1162                 return -EINVAL;
1163
1164         mutex_lock(&irq_mapping_update_lock);
1165
1166         irq = get_evtchn_to_irq(evtchn);
1167         if (irq == -1)
1168                 goto done;
1169
1170         info = irq_get_handler_data(irq);
1171
1172         if (!info)
1173                 goto done;
1174
1175         err = -EINVAL;
1176         if (info->refcnt <= 0)
1177                 goto done;
1178
1179         info->refcnt++;
1180         err = 0;
1181  done:
1182         mutex_unlock(&irq_mapping_update_lock);
1183
1184         return err;
1185 }
1186 EXPORT_SYMBOL_GPL(evtchn_get);
1187
1188 void evtchn_put(unsigned int evtchn)
1189 {
1190         int irq = get_evtchn_to_irq(evtchn);
1191         if (WARN_ON(irq == -1))
1192                 return;
1193         unbind_from_irq(irq);
1194 }
1195 EXPORT_SYMBOL_GPL(evtchn_put);
1196
1197 void xen_send_IPI_one(unsigned int cpu, enum ipi_vector vector)
1198 {
1199         int irq;
1200
1201 #ifdef CONFIG_X86
1202         if (unlikely(vector == XEN_NMI_VECTOR)) {
1203                 int rc =  HYPERVISOR_vcpu_op(VCPUOP_send_nmi, cpu, NULL);
1204                 if (rc < 0)
1205                         printk(KERN_WARNING "Sending nmi to CPU%d failed (rc:%d)\n", cpu, rc);
1206                 return;
1207         }
1208 #endif
1209         irq = per_cpu(ipi_to_irq, cpu)[vector];
1210         BUG_ON(irq < 0);
1211         notify_remote_via_irq(irq);
1212 }
1213
1214 static DEFINE_PER_CPU(unsigned, xed_nesting_count);
1215
1216 static void __xen_evtchn_do_upcall(void)
1217 {
1218         struct vcpu_info *vcpu_info = __this_cpu_read(xen_vcpu);
1219         int cpu = get_cpu();
1220         unsigned count;
1221
1222         do {
1223                 vcpu_info->evtchn_upcall_pending = 0;
1224
1225                 if (__this_cpu_inc_return(xed_nesting_count) - 1)
1226                         goto out;
1227
1228                 xen_evtchn_handle_events(cpu);
1229
1230                 BUG_ON(!irqs_disabled());
1231
1232                 count = __this_cpu_read(xed_nesting_count);
1233                 __this_cpu_write(xed_nesting_count, 0);
1234         } while (count != 1 || vcpu_info->evtchn_upcall_pending);
1235
1236 out:
1237
1238         put_cpu();
1239 }
1240
1241 void xen_evtchn_do_upcall(struct pt_regs *regs)
1242 {
1243         struct pt_regs *old_regs = set_irq_regs(regs);
1244
1245         irq_enter();
1246 #ifdef CONFIG_X86
1247         exit_idle();
1248         inc_irq_stat(irq_hv_callback_count);
1249 #endif
1250
1251         __xen_evtchn_do_upcall();
1252
1253         irq_exit();
1254         set_irq_regs(old_regs);
1255 }
1256
1257 void xen_hvm_evtchn_do_upcall(void)
1258 {
1259         __xen_evtchn_do_upcall();
1260 }
1261 EXPORT_SYMBOL_GPL(xen_hvm_evtchn_do_upcall);
1262
1263 /* Rebind a new event channel to an existing irq. */
1264 void rebind_evtchn_irq(int evtchn, int irq)
1265 {
1266         struct irq_info *info = info_for_irq(irq);
1267
1268         if (WARN_ON(!info))
1269                 return;
1270
1271         /* Make sure the irq is masked, since the new event channel
1272            will also be masked. */
1273         disable_irq(irq);
1274
1275         mutex_lock(&irq_mapping_update_lock);
1276
1277         /* After resume the irq<->evtchn mappings are all cleared out */
1278         BUG_ON(get_evtchn_to_irq(evtchn) != -1);
1279         /* Expect irq to have been bound before,
1280            so there should be a proper type */
1281         BUG_ON(info->type == IRQT_UNBOUND);
1282
1283         (void)xen_irq_info_evtchn_setup(irq, evtchn);
1284
1285         mutex_unlock(&irq_mapping_update_lock);
1286
1287         bind_evtchn_to_cpu(evtchn, info->cpu);
1288         /* This will be deferred until interrupt is processed */
1289         irq_set_affinity(irq, cpumask_of(info->cpu));
1290
1291         /* Unmask the event channel. */
1292         enable_irq(irq);
1293 }
1294
1295 /* Rebind an evtchn so that it gets delivered to a specific cpu */
1296 static int rebind_irq_to_cpu(unsigned irq, unsigned tcpu)
1297 {
1298         struct evtchn_bind_vcpu bind_vcpu;
1299         int evtchn = evtchn_from_irq(irq);
1300         int masked;
1301
1302         if (!VALID_EVTCHN(evtchn))
1303                 return -1;
1304
1305         if (!xen_support_evtchn_rebind())
1306                 return -1;
1307
1308         /* Send future instances of this interrupt to other vcpu. */
1309         bind_vcpu.port = evtchn;
1310         bind_vcpu.vcpu = tcpu;
1311
1312         /*
1313          * Mask the event while changing the VCPU binding to prevent
1314          * it being delivered on an unexpected VCPU.
1315          */
1316         masked = test_and_set_mask(evtchn);
1317
1318         /*
1319          * If this fails, it usually just indicates that we're dealing with a
1320          * virq or IPI channel, which don't actually need to be rebound. Ignore
1321          * it, but don't do the xenlinux-level rebind in that case.
1322          */
1323         if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_vcpu, &bind_vcpu) >= 0)
1324                 bind_evtchn_to_cpu(evtchn, tcpu);
1325
1326         if (!masked)
1327                 unmask_evtchn(evtchn);
1328
1329         return 0;
1330 }
1331
1332 static int set_affinity_irq(struct irq_data *data, const struct cpumask *dest,
1333                             bool force)
1334 {
1335         unsigned tcpu = cpumask_first_and(dest, cpu_online_mask);
1336
1337         return rebind_irq_to_cpu(data->irq, tcpu);
1338 }
1339
1340 static void enable_dynirq(struct irq_data *data)
1341 {
1342         int evtchn = evtchn_from_irq(data->irq);
1343
1344         if (VALID_EVTCHN(evtchn))
1345                 unmask_evtchn(evtchn);
1346 }
1347
1348 static void disable_dynirq(struct irq_data *data)
1349 {
1350         int evtchn = evtchn_from_irq(data->irq);
1351
1352         if (VALID_EVTCHN(evtchn))
1353                 mask_evtchn(evtchn);
1354 }
1355
1356 static void ack_dynirq(struct irq_data *data)
1357 {
1358         int evtchn = evtchn_from_irq(data->irq);
1359
1360         irq_move_irq(data);
1361
1362         if (VALID_EVTCHN(evtchn))
1363                 clear_evtchn(evtchn);
1364 }
1365
1366 static void mask_ack_dynirq(struct irq_data *data)
1367 {
1368         disable_dynirq(data);
1369         ack_dynirq(data);
1370 }
1371
1372 static int retrigger_dynirq(struct irq_data *data)
1373 {
1374         unsigned int evtchn = evtchn_from_irq(data->irq);
1375         int masked;
1376
1377         if (!VALID_EVTCHN(evtchn))
1378                 return 0;
1379
1380         masked = test_and_set_mask(evtchn);
1381         set_evtchn(evtchn);
1382         if (!masked)
1383                 unmask_evtchn(evtchn);
1384
1385         return 1;
1386 }
1387
1388 static void restore_pirqs(void)
1389 {
1390         int pirq, rc, irq, gsi;
1391         struct physdev_map_pirq map_irq;
1392         struct irq_info *info;
1393
1394         list_for_each_entry(info, &xen_irq_list_head, list) {
1395                 if (info->type != IRQT_PIRQ)
1396                         continue;
1397
1398                 pirq = info->u.pirq.pirq;
1399                 gsi = info->u.pirq.gsi;
1400                 irq = info->irq;
1401
1402                 /* save/restore of PT devices doesn't work, so at this point the
1403                  * only devices present are GSI based emulated devices */
1404                 if (!gsi)
1405                         continue;
1406
1407                 map_irq.domid = DOMID_SELF;
1408                 map_irq.type = MAP_PIRQ_TYPE_GSI;
1409                 map_irq.index = gsi;
1410                 map_irq.pirq = pirq;
1411
1412                 rc = HYPERVISOR_physdev_op(PHYSDEVOP_map_pirq, &map_irq);
1413                 if (rc) {
1414                         pr_warn("xen map irq failed gsi=%d irq=%d pirq=%d rc=%d\n",
1415                                 gsi, irq, pirq, rc);
1416                         xen_free_irq(irq);
1417                         continue;
1418                 }
1419
1420                 printk(KERN_DEBUG "xen: --> irq=%d, pirq=%d\n", irq, map_irq.pirq);
1421
1422                 __startup_pirq(irq);
1423         }
1424 }
1425
1426 static void restore_cpu_virqs(unsigned int cpu)
1427 {
1428         struct evtchn_bind_virq bind_virq;
1429         int virq, irq, evtchn;
1430
1431         for (virq = 0; virq < NR_VIRQS; virq++) {
1432                 if ((irq = per_cpu(virq_to_irq, cpu)[virq]) == -1)
1433                         continue;
1434
1435                 BUG_ON(virq_from_irq(irq) != virq);
1436
1437                 /* Get a new binding from Xen. */
1438                 bind_virq.virq = virq;
1439                 bind_virq.vcpu = cpu;
1440                 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_virq,
1441                                                 &bind_virq) != 0)
1442                         BUG();
1443                 evtchn = bind_virq.port;
1444
1445                 /* Record the new mapping. */
1446                 (void)xen_irq_info_virq_setup(cpu, irq, evtchn, virq);
1447                 bind_evtchn_to_cpu(evtchn, cpu);
1448         }
1449 }
1450
1451 static void restore_cpu_ipis(unsigned int cpu)
1452 {
1453         struct evtchn_bind_ipi bind_ipi;
1454         int ipi, irq, evtchn;
1455
1456         for (ipi = 0; ipi < XEN_NR_IPIS; ipi++) {
1457                 if ((irq = per_cpu(ipi_to_irq, cpu)[ipi]) == -1)
1458                         continue;
1459
1460                 BUG_ON(ipi_from_irq(irq) != ipi);
1461
1462                 /* Get a new binding from Xen. */
1463                 bind_ipi.vcpu = cpu;
1464                 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_ipi,
1465                                                 &bind_ipi) != 0)
1466                         BUG();
1467                 evtchn = bind_ipi.port;
1468
1469                 /* Record the new mapping. */
1470                 (void)xen_irq_info_ipi_setup(cpu, irq, evtchn, ipi);
1471                 bind_evtchn_to_cpu(evtchn, cpu);
1472         }
1473 }
1474
1475 /* Clear an irq's pending state, in preparation for polling on it */
1476 void xen_clear_irq_pending(int irq)
1477 {
1478         int evtchn = evtchn_from_irq(irq);
1479
1480         if (VALID_EVTCHN(evtchn))
1481                 clear_evtchn(evtchn);
1482 }
1483 EXPORT_SYMBOL(xen_clear_irq_pending);
1484 void xen_set_irq_pending(int irq)
1485 {
1486         int evtchn = evtchn_from_irq(irq);
1487
1488         if (VALID_EVTCHN(evtchn))
1489                 set_evtchn(evtchn);
1490 }
1491
1492 bool xen_test_irq_pending(int irq)
1493 {
1494         int evtchn = evtchn_from_irq(irq);
1495         bool ret = false;
1496
1497         if (VALID_EVTCHN(evtchn))
1498                 ret = test_evtchn(evtchn);
1499
1500         return ret;
1501 }
1502
1503 /* Poll waiting for an irq to become pending with timeout.  In the usual case,
1504  * the irq will be disabled so it won't deliver an interrupt. */
1505 void xen_poll_irq_timeout(int irq, u64 timeout)
1506 {
1507         evtchn_port_t evtchn = evtchn_from_irq(irq);
1508
1509         if (VALID_EVTCHN(evtchn)) {
1510                 struct sched_poll poll;
1511
1512                 poll.nr_ports = 1;
1513                 poll.timeout = timeout;
1514                 set_xen_guest_handle(poll.ports, &evtchn);
1515
1516                 if (HYPERVISOR_sched_op(SCHEDOP_poll, &poll) != 0)
1517                         BUG();
1518         }
1519 }
1520 EXPORT_SYMBOL(xen_poll_irq_timeout);
1521 /* Poll waiting for an irq to become pending.  In the usual case, the
1522  * irq will be disabled so it won't deliver an interrupt. */
1523 void xen_poll_irq(int irq)
1524 {
1525         xen_poll_irq_timeout(irq, 0 /* no timeout */);
1526 }
1527
1528 /* Check whether the IRQ line is shared with other guests. */
1529 int xen_test_irq_shared(int irq)
1530 {
1531         struct irq_info *info = info_for_irq(irq);
1532         struct physdev_irq_status_query irq_status;
1533
1534         if (WARN_ON(!info))
1535                 return -ENOENT;
1536
1537         irq_status.irq = info->u.pirq.pirq;
1538
1539         if (HYPERVISOR_physdev_op(PHYSDEVOP_irq_status_query, &irq_status))
1540                 return 0;
1541         return !(irq_status.flags & XENIRQSTAT_shared);
1542 }
1543 EXPORT_SYMBOL_GPL(xen_test_irq_shared);
1544
1545 void xen_irq_resume(void)
1546 {
1547         unsigned int cpu;
1548         struct irq_info *info;
1549
1550         /* New event-channel space is not 'live' yet. */
1551         xen_evtchn_mask_all();
1552         xen_evtchn_resume();
1553
1554         /* No IRQ <-> event-channel mappings. */
1555         list_for_each_entry(info, &xen_irq_list_head, list)
1556                 info->evtchn = 0; /* zap event-channel binding */
1557
1558         clear_evtchn_to_irq_all();
1559
1560         for_each_possible_cpu(cpu) {
1561                 restore_cpu_virqs(cpu);
1562                 restore_cpu_ipis(cpu);
1563         }
1564
1565         restore_pirqs();
1566 }
1567
1568 static struct irq_chip xen_dynamic_chip __read_mostly = {
1569         .name                   = "xen-dyn",
1570
1571         .irq_disable            = disable_dynirq,
1572         .irq_mask               = disable_dynirq,
1573         .irq_unmask             = enable_dynirq,
1574
1575         .irq_ack                = ack_dynirq,
1576         .irq_mask_ack           = mask_ack_dynirq,
1577
1578         .irq_set_affinity       = set_affinity_irq,
1579         .irq_retrigger          = retrigger_dynirq,
1580 };
1581
1582 static struct irq_chip xen_pirq_chip __read_mostly = {
1583         .name                   = "xen-pirq",
1584
1585         .irq_startup            = startup_pirq,
1586         .irq_shutdown           = shutdown_pirq,
1587         .irq_enable             = enable_pirq,
1588         .irq_disable            = disable_pirq,
1589
1590         .irq_mask               = disable_dynirq,
1591         .irq_unmask             = enable_dynirq,
1592
1593         .irq_ack                = eoi_pirq,
1594         .irq_eoi                = eoi_pirq,
1595         .irq_mask_ack           = mask_ack_pirq,
1596
1597         .irq_set_affinity       = set_affinity_irq,
1598
1599         .irq_retrigger          = retrigger_dynirq,
1600 };
1601
1602 static struct irq_chip xen_percpu_chip __read_mostly = {
1603         .name                   = "xen-percpu",
1604
1605         .irq_disable            = disable_dynirq,
1606         .irq_mask               = disable_dynirq,
1607         .irq_unmask             = enable_dynirq,
1608
1609         .irq_ack                = ack_dynirq,
1610 };
1611
1612 int xen_set_callback_via(uint64_t via)
1613 {
1614         struct xen_hvm_param a;
1615         a.domid = DOMID_SELF;
1616         a.index = HVM_PARAM_CALLBACK_IRQ;
1617         a.value = via;
1618         return HYPERVISOR_hvm_op(HVMOP_set_param, &a);
1619 }
1620 EXPORT_SYMBOL_GPL(xen_set_callback_via);
1621
1622 #ifdef CONFIG_XEN_PVHVM
1623 /* Vector callbacks are better than PCI interrupts to receive event
1624  * channel notifications because we can receive vector callbacks on any
1625  * vcpu and we don't need PCI support or APIC interactions. */
1626 void xen_callback_vector(void)
1627 {
1628         int rc;
1629         uint64_t callback_via;
1630         if (xen_have_vector_callback) {
1631                 callback_via = HVM_CALLBACK_VECTOR(HYPERVISOR_CALLBACK_VECTOR);
1632                 rc = xen_set_callback_via(callback_via);
1633                 if (rc) {
1634                         pr_err("Request for Xen HVM callback vector failed\n");
1635                         xen_have_vector_callback = 0;
1636                         return;
1637                 }
1638                 pr_info("Xen HVM callback vector for event delivery is enabled\n");
1639                 /* in the restore case the vector has already been allocated */
1640                 if (!test_bit(HYPERVISOR_CALLBACK_VECTOR, used_vectors))
1641                         alloc_intr_gate(HYPERVISOR_CALLBACK_VECTOR,
1642                                         xen_hvm_callback_vector);
1643         }
1644 }
1645 #else
1646 void xen_callback_vector(void) {}
1647 #endif
1648
1649 #undef MODULE_PARAM_PREFIX
1650 #define MODULE_PARAM_PREFIX "xen."
1651
1652 static bool fifo_events = true;
1653 module_param(fifo_events, bool, 0);
1654
1655 void __init xen_init_IRQ(void)
1656 {
1657         int ret = -EINVAL;
1658
1659         if (fifo_events)
1660                 ret = xen_evtchn_fifo_init();
1661         if (ret < 0)
1662                 xen_evtchn_2l_init();
1663
1664         evtchn_to_irq = kcalloc(EVTCHN_ROW(xen_evtchn_max_channels()),
1665                                 sizeof(*evtchn_to_irq), GFP_KERNEL);
1666         BUG_ON(!evtchn_to_irq);
1667
1668         /* No event channels are 'live' right now. */
1669         xen_evtchn_mask_all();
1670
1671         pirq_needs_eoi = pirq_needs_eoi_flag;
1672
1673 #ifdef CONFIG_X86
1674         if (xen_pv_domain()) {
1675                 irq_ctx_init(smp_processor_id());
1676                 if (xen_initial_domain())
1677                         pci_xen_initial_domain();
1678         }
1679         if (xen_feature(XENFEAT_hvm_callback_vector))
1680                 xen_callback_vector();
1681
1682         if (xen_hvm_domain()) {
1683                 native_init_IRQ();
1684                 /* pci_xen_hvm_init must be called after native_init_IRQ so that
1685                  * __acpi_register_gsi can point at the right function */
1686                 pci_xen_hvm_init();
1687         } else {
1688                 int rc;
1689                 struct physdev_pirq_eoi_gmfn eoi_gmfn;
1690
1691                 pirq_eoi_map = (void *)__get_free_page(GFP_KERNEL|__GFP_ZERO);
1692                 eoi_gmfn.gmfn = virt_to_gfn(pirq_eoi_map);
1693                 rc = HYPERVISOR_physdev_op(PHYSDEVOP_pirq_eoi_gmfn_v2, &eoi_gmfn);
1694                 /* TODO: No PVH support for PIRQ EOI */
1695                 if (rc != 0) {
1696                         free_page((unsigned long) pirq_eoi_map);
1697                         pirq_eoi_map = NULL;
1698                 } else
1699                         pirq_needs_eoi = pirq_check_eoi_map;
1700         }
1701 #endif
1702 }