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/pci.h>
  43 #include <xen/page.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 >= 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_get_irq_data(irq)->affinity, cpumask_of(cpu));
 340 #endif
 341         xen_evtchn_port_bind_to_cpu(info, cpu);
 342
 343         info->cpu = cpu;
 344 }
 345
 346 static void xen_evtchn_mask_all(void)
 347 {
 348         unsigned int evtchn;
 349
 350         for (evtchn = 0; evtchn < xen_evtchn_nr_channels(); evtchn++)
 351                 mask_evtchn(evtchn);
 352 }
 353
 354 /**
 355  * notify_remote_via_irq - send event to remote end of event channel via irq
 356  * @irq: irq of event channel to send event to
 357  *
 358  * Unlike notify_remote_via_evtchn(), this is safe to use across
 359  * save/restore. Notifications on a broken connection are silently
 360  * dropped.
 361  */
 362 void notify_remote_via_irq(int irq)
 363 {
 364         int evtchn = evtchn_from_irq(irq);
 365
 366         if (VALID_EVTCHN(evtchn))
 367                 notify_remote_via_evtchn(evtchn);
 368 }
 369 EXPORT_SYMBOL_GPL(notify_remote_via_irq);
 370
 371 static void xen_irq_init(unsigned irq)
 372 {
 373         struct irq_info *info;
 374 #ifdef CONFIG_SMP
 375         /* By default all event channels notify CPU#0. */
 376         cpumask_copy(irq_get_irq_data(irq)->affinity, cpumask_of(0));
 377 #endif
 378
 379         info = kzalloc(sizeof(*info), GFP_KERNEL);
 380         if (info == NULL)
 381                 panic("Unable to allocate metadata for IRQ%d\n", irq);
 382
 383         info->type = IRQT_UNBOUND;
 384         info->refcnt = -1;
 385
 386         irq_set_handler_data(irq, info);
 387
 388         list_add_tail(&info->list, &xen_irq_list_head);
 389 }
 390
 391 static int __must_check xen_allocate_irqs_dynamic(int nvec)
 392 {
 393         int i, irq = irq_alloc_descs(-1, 0, nvec, -1);
 394
 395         if (irq >= 0) {
 396                 for (i = 0; i < nvec; i++)
 397                         xen_irq_init(irq + i);
 398         }
 399
 400         return irq;
 401 }
 402
 403 static inline int __must_check xen_allocate_irq_dynamic(void)
 404 {
 405
 406         return xen_allocate_irqs_dynamic(1);
 407 }
 408
 409 static int __must_check xen_allocate_irq_gsi(unsigned gsi)
 410 {
 411         int irq;
 412
 413         /*
 414          * A PV guest has no concept of a GSI (since it has no ACPI
 415          * nor access to/knowledge of the physical APICs). Therefore
 416          * all IRQs are dynamically allocated from the entire IRQ
 417          * space.
 418          */
 419         if (xen_pv_domain() && !xen_initial_domain())
 420                 return xen_allocate_irq_dynamic();
 421
 422         /* Legacy IRQ descriptors are already allocated by the arch. */
 423         if (gsi < NR_IRQS_LEGACY)
 424                 irq = gsi;
 425         else
 426                 irq = irq_alloc_desc_at(gsi, -1);
 427
 428         xen_irq_init(irq);
 429
 430         return irq;
 431 }
 432
 433 static void xen_free_irq(unsigned irq)
 434 {
 435         struct irq_info *info = irq_get_handler_data(irq);
 436
 437         if (WARN_ON(!info))
 438                 return;
 439
 440         list_del(&info->list);
 441
 442         irq_set_handler_data(irq, NULL);
 443
 444         WARN_ON(info->refcnt > 0);
 445
 446         kfree(info);
 447
 448         /* Legacy IRQ descriptors are managed by the arch. */
 449         if (irq < NR_IRQS_LEGACY)
 450                 return;
 451
 452         irq_free_desc(irq);
 453 }
 454
 455 static void xen_evtchn_close(unsigned int port, unsigned int cpu)
 456 {
 457         struct evtchn_close close;
 458
 459         xen_evtchn_op_close(port, cpu);
 460
 461         close.port = port;
 462         if (HYPERVISOR_event_channel_op(EVTCHNOP_close, &close) != 0)
 463                 BUG();
 464 }
 465
 466 static void pirq_query_unmask(int irq)
 467 {
 468         struct physdev_irq_status_query irq_status;
 469         struct irq_info *info = info_for_irq(irq);
 470
 471         BUG_ON(info->type != IRQT_PIRQ);
 472
 473         irq_status.irq = pirq_from_irq(irq);
 474         if (HYPERVISOR_physdev_op(PHYSDEVOP_irq_status_query, &irq_status))
 475                 irq_status.flags = 0;
 476
 477         info->u.pirq.flags &= ~PIRQ_NEEDS_EOI;
 478         if (irq_status.flags & XENIRQSTAT_needs_eoi)
 479                 info->u.pirq.flags |= PIRQ_NEEDS_EOI;
 480 }
 481
 482 static void eoi_pirq(struct irq_data *data)
 483 {
 484         int evtchn = evtchn_from_irq(data->irq);
 485         struct physdev_eoi eoi = { .irq = pirq_from_irq(data->irq) };
 486         int rc = 0;
 487
 488         irq_move_irq(data);
 489
 490         if (VALID_EVTCHN(evtchn))
 491                 clear_evtchn(evtchn);
 492
 493         if (pirq_needs_eoi(data->irq)) {
 494                 rc = HYPERVISOR_physdev_op(PHYSDEVOP_eoi, &eoi);
 495                 WARN_ON(rc);
 496         }
 497 }
 498
 499 static void mask_ack_pirq(struct irq_data *data)
 500 {
 501         disable_dynirq(data);
 502         eoi_pirq(data);
 503 }
 504
 505 static unsigned int __startup_pirq(unsigned int irq)
 506 {
 507         struct evtchn_bind_pirq bind_pirq;
 508         struct irq_info *info = info_for_irq(irq);
 509         int evtchn = evtchn_from_irq(irq);
 510         int rc;
 511
 512         BUG_ON(info->type != IRQT_PIRQ);
 513
 514         if (VALID_EVTCHN(evtchn))
 515                 goto out;
 516
 517         bind_pirq.pirq = pirq_from_irq(irq);
 518         /* NB. We are happy to share unless we are probing. */
 519         bind_pirq.flags = info->u.pirq.flags & PIRQ_SHAREABLE ?
 520                                         BIND_PIRQ__WILL_SHARE : 0;
 521         rc = HYPERVISOR_event_channel_op(EVTCHNOP_bind_pirq, &bind_pirq);
 522         if (rc != 0) {
 523                 pr_warn("Failed to obtain physical IRQ %d\n", irq);
 524                 return 0;
 525         }
 526         evtchn = bind_pirq.port;
 527
 528         pirq_query_unmask(irq);
 529
 530         rc = set_evtchn_to_irq(evtchn, irq);
 531         if (rc)
 532                 goto err;
 533
 534         info->evtchn = evtchn;
 535         bind_evtchn_to_cpu(evtchn, 0);
 536
 537         rc = xen_evtchn_port_setup(info);
 538         if (rc)
 539                 goto err;
 540
 541 out:
 542         unmask_evtchn(evtchn);
 543         eoi_pirq(irq_get_irq_data(irq));
 544
 545         return 0;
 546
 547 err:
 548         pr_err("irq%d: Failed to set port to irq mapping (%d)\n", irq, rc);
 549         xen_evtchn_close(evtchn, NR_CPUS);
 550         return 0;
 551 }
 552
 553 static unsigned int startup_pirq(struct irq_data *data)
 554 {
 555         return __startup_pirq(data->irq);
 556 }
 557
 558 static void shutdown_pirq(struct irq_data *data)
 559 {
 560         unsigned int irq = data->irq;
 561         struct irq_info *info = info_for_irq(irq);
 562         unsigned evtchn = evtchn_from_irq(irq);
 563
 564         BUG_ON(info->type != IRQT_PIRQ);
 565
 566         if (!VALID_EVTCHN(evtchn))
 567                 return;
 568
 569         mask_evtchn(evtchn);
 570         xen_evtchn_close(evtchn, cpu_from_evtchn(evtchn));
 571         xen_irq_info_cleanup(info);
 572 }
 573
 574 static void enable_pirq(struct irq_data *data)
 575 {
 576         startup_pirq(data);
 577 }
 578
 579 static void disable_pirq(struct irq_data *data)
 580 {
 581         disable_dynirq(data);
 582 }
 583
 584 int xen_irq_from_gsi(unsigned gsi)
 585 {
 586         struct irq_info *info;
 587
 588         list_for_each_entry(info, &xen_irq_list_head, list) {
 589                 if (info->type != IRQT_PIRQ)
 590                         continue;
 591
 592                 if (info->u.pirq.gsi == gsi)
 593                         return info->irq;
 594         }
 595
 596         return -1;
 597 }
 598 EXPORT_SYMBOL_GPL(xen_irq_from_gsi);
 599
 600 static void __unbind_from_irq(unsigned int irq)
 601 {
 602         int evtchn = evtchn_from_irq(irq);
 603         struct irq_info *info = irq_get_handler_data(irq);
 604
 605         if (info->refcnt > 0) {
 606                 info->refcnt--;
 607                 if (info->refcnt != 0)
 608                         return;
 609         }
 610
 611         if (VALID_EVTCHN(evtchn)) {
 612                 unsigned int cpu = cpu_from_irq(irq);
 613
 614                 xen_evtchn_close(evtchn, cpu);
 615
 616                 switch (type_from_irq(irq)) {
 617                 case IRQT_VIRQ:
 618                         per_cpu(virq_to_irq, cpu)[virq_from_irq(irq)] = -1;
 619                         break;
 620                 case IRQT_IPI:
 621                         per_cpu(ipi_to_irq, cpu)[ipi_from_irq(irq)] = -1;
 622                         break;
 623                 default:
 624                         break;
 625                 }
 626
 627                 xen_irq_info_cleanup(info);
 628         }
 629
 630         BUG_ON(info_for_irq(irq)->type == IRQT_UNBOUND);
 631
 632         xen_free_irq(irq);
 633 }
 634
 635 /*
 636  * Do not make any assumptions regarding the relationship between the
 637  * IRQ number returned here and the Xen pirq argument.
 638  *
 639  * Note: We don't assign an event channel until the irq actually started
 640  * up.  Return an existing irq if we've already got one for the gsi.
 641  *
 642  * Shareable implies level triggered, not shareable implies edge
 643  * triggered here.
 644  */
 645 int xen_bind_pirq_gsi_to_irq(unsigned gsi,
 646                              unsigned pirq, int shareable, char *name)
 647 {
 648         int irq = -1;
 649         struct physdev_irq irq_op;
 650         int ret;
 651
 652         mutex_lock(&irq_mapping_update_lock);
 653
 654         irq = xen_irq_from_gsi(gsi);
 655         if (irq != -1) {
 656                 pr_info("%s: returning irq %d for gsi %u\n",
 657                         __func__, irq, gsi);
 658                 goto out;
 659         }
 660
 661         irq = xen_allocate_irq_gsi(gsi);
 662         if (irq < 0)
 663                 goto out;
 664
 665         irq_op.irq = irq;
 666         irq_op.vector = 0;
 667
 668         /* Only the privileged domain can do this. For non-priv, the pcifront
 669          * driver provides a PCI bus that does the call to do exactly
 670          * this in the priv domain. */
 671         if (xen_initial_domain() &&
 672             HYPERVISOR_physdev_op(PHYSDEVOP_alloc_irq_vector, &irq_op)) {
 673                 xen_free_irq(irq);
 674                 irq = -ENOSPC;
 675                 goto out;
 676         }
 677
 678         ret = xen_irq_info_pirq_setup(irq, 0, pirq, gsi, DOMID_SELF,
 679                                shareable ? PIRQ_SHAREABLE : 0);
 680         if (ret < 0) {
 681                 __unbind_from_irq(irq);
 682                 irq = ret;
 683                 goto out;
 684         }
 685
 686         pirq_query_unmask(irq);
 687         /* We try to use the handler with the appropriate semantic for the
 688          * type of interrupt: if the interrupt is an edge triggered
 689          * interrupt we use handle_edge_irq.
 690          *
 691          * On the other hand if the interrupt is level triggered we use
 692          * handle_fasteoi_irq like the native code does for this kind of
 693          * interrupts.
 694          *
 695          * Depending on the Xen version, pirq_needs_eoi might return true
 696          * not only for level triggered interrupts but for edge triggered
 697          * interrupts too. In any case Xen always honors the eoi mechanism,
 698          * not injecting any more pirqs of the same kind if the first one
 699          * hasn't received an eoi yet. Therefore using the fasteoi handler
 700          * is the right choice either way.
 701          */
 702         if (shareable)
 703                 irq_set_chip_and_handler_name(irq, &xen_pirq_chip,
 704                                 handle_fasteoi_irq, name);
 705         else
 706                 irq_set_chip_and_handler_name(irq, &xen_pirq_chip,
 707                                 handle_edge_irq, name);
 708
 709 out:
 710         mutex_unlock(&irq_mapping_update_lock);
 711
 712         return irq;
 713 }
 714
 715 #ifdef CONFIG_PCI_MSI
 716 int xen_allocate_pirq_msi(struct pci_dev *dev, struct msi_desc *msidesc)
 717 {
 718         int rc;
 719         struct physdev_get_free_pirq op_get_free_pirq;
 720
 721         op_get_free_pirq.type = MAP_PIRQ_TYPE_MSI;
 722         rc = HYPERVISOR_physdev_op(PHYSDEVOP_get_free_pirq, &op_get_free_pirq);
 723
 724         WARN_ONCE(rc == -ENOSYS,
 725                   "hypervisor does not support the PHYSDEVOP_get_free_pirq interface\n");
 726
 727         return rc ? -1 : op_get_free_pirq.pirq;
 728 }
 729
 730 int xen_bind_pirq_msi_to_irq(struct pci_dev *dev, struct msi_desc *msidesc,
 731                              int pirq, int nvec, const char *name, domid_t domid)
 732 {
 733         int i, irq, ret;
 734
 735         mutex_lock(&irq_mapping_update_lock);
 736
 737         irq = xen_allocate_irqs_dynamic(nvec);
 738         if (irq < 0)
 739                 goto out;
 740
 741         for (i = 0; i < nvec; i++) {
 742                 irq_set_chip_and_handler_name(irq + i, &xen_pirq_chip, handle_edge_irq, name);
 743
 744                 ret = xen_irq_info_pirq_setup(irq + i, 0, pirq + i, 0, domid,
 745                                               i == 0 ? 0 : PIRQ_MSI_GROUP);
 746                 if (ret < 0)
 747                         goto error_irq;
 748         }
 749
 750         ret = irq_set_msi_desc(irq, msidesc);
 751         if (ret < 0)
 752                 goto error_irq;
 753 out:
 754         mutex_unlock(&irq_mapping_update_lock);
 755         return irq;
 756 error_irq:
 757         for (; i >= 0; i--)
 758                 __unbind_from_irq(irq + i);
 759         mutex_unlock(&irq_mapping_update_lock);
 760         return ret;
 761 }
 762 #endif
 763
 764 int xen_destroy_irq(int irq)
 765 {
 766         struct physdev_unmap_pirq unmap_irq;
 767         struct irq_info *info = info_for_irq(irq);
 768         int rc = -ENOENT;
 769
 770         mutex_lock(&irq_mapping_update_lock);
 771
 772         /*
 773          * If trying to remove a vector in a MSI group different
 774          * than the first one skip the PIRQ unmap unless this vector
 775          * is the first one in the group.
 776          */
 777         if (xen_initial_domain() && !(info->u.pirq.flags & PIRQ_MSI_GROUP)) {
 778                 unmap_irq.pirq = info->u.pirq.pirq;
 779                 unmap_irq.domid = info->u.pirq.domid;
 780                 rc = HYPERVISOR_physdev_op(PHYSDEVOP_unmap_pirq, &unmap_irq);
 781                 /* If another domain quits without making the pci_disable_msix
 782                  * call, the Xen hypervisor takes care of freeing the PIRQs
 783                  * (free_domain_pirqs).
 784                  */
 785                 if ((rc == -ESRCH && info->u.pirq.domid != DOMID_SELF))
 786                         pr_info("domain %d does not have %d anymore\n",
 787                                 info->u.pirq.domid, info->u.pirq.pirq);
 788                 else if (rc) {
 789                         pr_warn("unmap irq failed %d\n", rc);
 790                         goto out;
 791                 }
 792         }
 793
 794         xen_free_irq(irq);
 795
 796 out:
 797         mutex_unlock(&irq_mapping_update_lock);
 798         return rc;
 799 }
 800
 801 int xen_irq_from_pirq(unsigned pirq)
 802 {
 803         int irq;
 804
 805         struct irq_info *info;
 806
 807         mutex_lock(&irq_mapping_update_lock);
 808
 809         list_for_each_entry(info, &xen_irq_list_head, list) {
 810                 if (info->type != IRQT_PIRQ)
 811                         continue;
 812                 irq = info->irq;
 813                 if (info->u.pirq.pirq == pirq)
 814                         goto out;
 815         }
 816         irq = -1;
 817 out:
 818         mutex_unlock(&irq_mapping_update_lock);
 819
 820         return irq;
 821 }
 822
 823
 824 int xen_pirq_from_irq(unsigned irq)
 825 {
 826         return pirq_from_irq(irq);
 827 }
 828 EXPORT_SYMBOL_GPL(xen_pirq_from_irq);
 829
 830 int bind_evtchn_to_irq(unsigned int evtchn)
 831 {
 832         int irq;
 833         int ret;
 834
 835         if (evtchn >= xen_evtchn_max_channels())
 836                 return -ENOMEM;
 837
 838         mutex_lock(&irq_mapping_update_lock);
 839
 840         irq = get_evtchn_to_irq(evtchn);
 841
 842         if (irq == -1) {
 843                 irq = xen_allocate_irq_dynamic();
 844                 if (irq < 0)
 845                         goto out;
 846
 847                 irq_set_chip_and_handler_name(irq, &xen_dynamic_chip,
 848                                               handle_edge_irq, "event");
 849
 850                 ret = xen_irq_info_evtchn_setup(irq, evtchn);
 851                 if (ret < 0) {
 852                         __unbind_from_irq(irq);
 853                         irq = ret;
 854                         goto out;
 855                 }
 856                 /* New interdomain events are bound to VCPU 0. */
 857                 bind_evtchn_to_cpu(evtchn, 0);
 858         } else {
 859                 struct irq_info *info = info_for_irq(irq);
 860                 WARN_ON(info == NULL || info->type != IRQT_EVTCHN);
 861         }
 862
 863 out:
 864         mutex_unlock(&irq_mapping_update_lock);
 865
 866         return irq;
 867 }
 868 EXPORT_SYMBOL_GPL(bind_evtchn_to_irq);
 869
 870 static int bind_ipi_to_irq(unsigned int ipi, unsigned int cpu)
 871 {
 872         struct evtchn_bind_ipi bind_ipi;
 873         int evtchn, irq;
 874         int ret;
 875
 876         mutex_lock(&irq_mapping_update_lock);
 877
 878         irq = per_cpu(ipi_to_irq, cpu)[ipi];
 879
 880         if (irq == -1) {
 881                 irq = xen_allocate_irq_dynamic();
 882                 if (irq < 0)
 883                         goto out;
 884
 885                 irq_set_chip_and_handler_name(irq, &xen_percpu_chip,
 886                                               handle_percpu_irq, "ipi");
 887
 888                 bind_ipi.vcpu = cpu;
 889                 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_ipi,
 890                                                 &bind_ipi) != 0)
 891                         BUG();
 892                 evtchn = bind_ipi.port;
 893
 894                 ret = xen_irq_info_ipi_setup(cpu, irq, evtchn, ipi);
 895                 if (ret < 0) {
 896                         __unbind_from_irq(irq);
 897                         irq = ret;
 898                         goto out;
 899                 }
 900                 bind_evtchn_to_cpu(evtchn, cpu);
 901         } else {
 902                 struct irq_info *info = info_for_irq(irq);
 903                 WARN_ON(info == NULL || info->type != IRQT_IPI);
 904         }
 905
 906  out:
 907         mutex_unlock(&irq_mapping_update_lock);
 908         return irq;
 909 }
 910
 911 int bind_interdomain_evtchn_to_irq(unsigned int remote_domain,
 912                                    unsigned int remote_port)
 913 {
 914         struct evtchn_bind_interdomain bind_interdomain;
 915         int err;
 916
 917         bind_interdomain.remote_dom  = remote_domain;
 918         bind_interdomain.remote_port = remote_port;
 919
 920         err = HYPERVISOR_event_channel_op(EVTCHNOP_bind_interdomain,
 921                                           &bind_interdomain);
 922
 923         return err ? : bind_evtchn_to_irq(bind_interdomain.local_port);
 924 }
 925 EXPORT_SYMBOL_GPL(bind_interdomain_evtchn_to_irq);
 926
 927 static int find_virq(unsigned int virq, unsigned int cpu)
 928 {
 929         struct evtchn_status status;
 930         int port, rc = -ENOENT;
 931
 932         memset(&status, 0, sizeof(status));
 933         for (port = 0; port < xen_evtchn_max_channels(); port++) {
 934                 status.dom = DOMID_SELF;
 935                 status.port = port;
 936                 rc = HYPERVISOR_event_channel_op(EVTCHNOP_status, &status);
 937                 if (rc < 0)
 938                         continue;
 939                 if (status.status != EVTCHNSTAT_virq)
 940                         continue;
 941                 if (status.u.virq == virq && status.vcpu == cpu) {
 942                         rc = port;
 943                         break;
 944                 }
 945         }
 946         return rc;
 947 }
 948
 949 /**
 950  * xen_evtchn_nr_channels - number of usable event channel ports
 951  *
 952  * This may be less than the maximum supported by the current
 953  * hypervisor ABI. Use xen_evtchn_max_channels() for the maximum
 954  * supported.
 955  */
 956 unsigned xen_evtchn_nr_channels(void)
 957 {
 958         return evtchn_ops->nr_channels();
 959 }
 960 EXPORT_SYMBOL_GPL(xen_evtchn_nr_channels);
 961
 962 int bind_virq_to_irq(unsigned int virq, unsigned int cpu, bool percpu)
 963 {
 964         struct evtchn_bind_virq bind_virq;
 965         int evtchn, irq, ret;
 966
 967         mutex_lock(&irq_mapping_update_lock);
 968
 969         irq = per_cpu(virq_to_irq, cpu)[virq];
 970
 971         if (irq == -1) {
 972                 irq = xen_allocate_irq_dynamic();
 973                 if (irq < 0)
 974                         goto out;
 975
 976                 if (percpu)
 977                         irq_set_chip_and_handler_name(irq, &xen_percpu_chip,
 978                                                       handle_percpu_irq, "virq");
 979                 else
 980                         irq_set_chip_and_handler_name(irq, &xen_dynamic_chip,
 981                                                       handle_edge_irq, "virq");
 982
 983                 bind_virq.virq = virq;
 984                 bind_virq.vcpu = cpu;
 985                 ret = HYPERVISOR_event_channel_op(EVTCHNOP_bind_virq,
 986                                                 &bind_virq);
 987                 if (ret == 0)
 988                         evtchn = bind_virq.port;
 989                 else {
 990                         if (ret == -EEXIST)
 991                                 ret = find_virq(virq, cpu);
 992                         BUG_ON(ret < 0);
 993                         evtchn = ret;
 994                 }
 995
 996                 ret = xen_irq_info_virq_setup(cpu, irq, evtchn, virq);
 997                 if (ret < 0) {
 998                         __unbind_from_irq(irq);
 999                         irq = ret;
1000                         goto out;
1001                 }
1002
1003                 bind_evtchn_to_cpu(evtchn, cpu);
1004         } else {
1005                 struct irq_info *info = info_for_irq(irq);
1006                 WARN_ON(info == NULL || info->type != IRQT_VIRQ);
1007         }
1008
1009 out:
1010         mutex_unlock(&irq_mapping_update_lock);
1011
1012         return irq;
1013 }
1014
1015 static void unbind_from_irq(unsigned int irq)
1016 {
1017         mutex_lock(&irq_mapping_update_lock);
1018         __unbind_from_irq(irq);
1019         mutex_unlock(&irq_mapping_update_lock);
1020 }
1021
1022 int bind_evtchn_to_irqhandler(unsigned int evtchn,
1023                               irq_handler_t handler,
1024                               unsigned long irqflags,
1025                               const char *devname, void *dev_id)
1026 {
1027         int irq, retval;
1028
1029         irq = bind_evtchn_to_irq(evtchn);
1030         if (irq < 0)
1031                 return irq;
1032         retval = request_irq(irq, handler, irqflags, devname, dev_id);
1033         if (retval != 0) {
1034                 unbind_from_irq(irq);
1035                 return retval;
1036         }
1037
1038         return irq;
1039 }
1040 EXPORT_SYMBOL_GPL(bind_evtchn_to_irqhandler);
1041
1042 int bind_interdomain_evtchn_to_irqhandler(unsigned int remote_domain,
1043                                           unsigned int remote_port,
1044                                           irq_handler_t handler,
1045                                           unsigned long irqflags,
1046                                           const char *devname,
1047                                           void *dev_id)
1048 {
1049         int irq, retval;
1050
1051         irq = bind_interdomain_evtchn_to_irq(remote_domain, remote_port);
1052         if (irq < 0)
1053                 return irq;
1054
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_interdomain_evtchn_to_irqhandler);
1064
1065 int bind_virq_to_irqhandler(unsigned int virq, unsigned int cpu,
1066                             irq_handler_t handler,
1067                             unsigned long irqflags, const char *devname, void *dev_id)
1068 {
1069         int irq, retval;
1070
1071         irq = bind_virq_to_irq(virq, cpu, irqflags & IRQF_PERCPU);
1072         if (irq < 0)
1073                 return irq;
1074         retval = request_irq(irq, handler, irqflags, devname, dev_id);
1075         if (retval != 0) {
1076                 unbind_from_irq(irq);
1077                 return retval;
1078         }
1079
1080         return irq;
1081 }
1082 EXPORT_SYMBOL_GPL(bind_virq_to_irqhandler);
1083
1084 int bind_ipi_to_irqhandler(enum ipi_vector ipi,
1085                            unsigned int cpu,
1086                            irq_handler_t handler,
1087                            unsigned long irqflags,
1088                            const char *devname,
1089                            void *dev_id)
1090 {
1091         int irq, retval;
1092
1093         irq = bind_ipi_to_irq(ipi, cpu);
1094         if (irq < 0)
1095                 return irq;
1096
1097         irqflags |= IRQF_NO_SUSPEND | IRQF_FORCE_RESUME | IRQF_EARLY_RESUME;
1098         retval = request_irq(irq, handler, irqflags, devname, dev_id);
1099         if (retval != 0) {
1100                 unbind_from_irq(irq);
1101                 return retval;
1102         }
1103
1104         return irq;
1105 }
1106
1107 void unbind_from_irqhandler(unsigned int irq, void *dev_id)
1108 {
1109         struct irq_info *info = irq_get_handler_data(irq);
1110
1111         if (WARN_ON(!info))
1112                 return;
1113         free_irq(irq, dev_id);
1114         unbind_from_irq(irq);
1115 }
1116 EXPORT_SYMBOL_GPL(unbind_from_irqhandler);
1117
1118 /**
1119  * xen_set_irq_priority() - set an event channel priority.
1120  * @irq:irq bound to an event channel.
1121  * @priority: priority between XEN_IRQ_PRIORITY_MAX and XEN_IRQ_PRIORITY_MIN.
1122  */
1123 int xen_set_irq_priority(unsigned irq, unsigned priority)
1124 {
1125         struct evtchn_set_priority set_priority;
1126
1127         set_priority.port = evtchn_from_irq(irq);
1128         set_priority.priority = priority;
1129
1130         return HYPERVISOR_event_channel_op(EVTCHNOP_set_priority,
1131                                            &set_priority);
1132 }
1133 EXPORT_SYMBOL_GPL(xen_set_irq_priority);
1134
1135 int evtchn_make_refcounted(unsigned int evtchn)
1136 {
1137         int irq = get_evtchn_to_irq(evtchn);
1138         struct irq_info *info;
1139
1140         if (irq == -1)
1141                 return -ENOENT;
1142
1143         info = irq_get_handler_data(irq);
1144
1145         if (!info)
1146                 return -ENOENT;
1147
1148         WARN_ON(info->refcnt != -1);
1149
1150         info->refcnt = 1;
1151
1152         return 0;
1153 }
1154 EXPORT_SYMBOL_GPL(evtchn_make_refcounted);
1155
1156 int evtchn_get(unsigned int evtchn)
1157 {
1158         int irq;
1159         struct irq_info *info;
1160         int err = -ENOENT;
1161
1162         if (evtchn >= xen_evtchn_max_channels())
1163                 return -EINVAL;
1164
1165         mutex_lock(&irq_mapping_update_lock);
1166
1167         irq = get_evtchn_to_irq(evtchn);
1168         if (irq == -1)
1169                 goto done;
1170
1171         info = irq_get_handler_data(irq);
1172
1173         if (!info)
1174                 goto done;
1175
1176         err = -EINVAL;
1177         if (info->refcnt <= 0)
1178                 goto done;
1179
1180         info->refcnt++;
1181         err = 0;
1182  done:
1183         mutex_unlock(&irq_mapping_update_lock);
1184
1185         return err;
1186 }
1187 EXPORT_SYMBOL_GPL(evtchn_get);
1188
1189 void evtchn_put(unsigned int evtchn)
1190 {
1191         int irq = get_evtchn_to_irq(evtchn);
1192         if (WARN_ON(irq == -1))
1193                 return;
1194         unbind_from_irq(irq);
1195 }
1196 EXPORT_SYMBOL_GPL(evtchn_put);
1197
1198 void xen_send_IPI_one(unsigned int cpu, enum ipi_vector vector)
1199 {
1200         int irq;
1201
1202 #ifdef CONFIG_X86
1203         if (unlikely(vector == XEN_NMI_VECTOR)) {
1204                 int rc =  HYPERVISOR_vcpu_op(VCPUOP_send_nmi, cpu, NULL);
1205                 if (rc < 0)
1206                         printk(KERN_WARNING "Sending nmi to CPU%d failed (rc:%d)\n", cpu, rc);
1207                 return;
1208         }
1209 #endif
1210         irq = per_cpu(ipi_to_irq, cpu)[vector];
1211         BUG_ON(irq < 0);
1212         notify_remote_via_irq(irq);
1213 }
1214
1215 static DEFINE_PER_CPU(unsigned, xed_nesting_count);
1216
1217 static void __xen_evtchn_do_upcall(void)
1218 {
1219         struct vcpu_info *vcpu_info = __this_cpu_read(xen_vcpu);
1220         int cpu = get_cpu();
1221         unsigned count;
1222
1223         do {
1224                 vcpu_info->evtchn_upcall_pending = 0;
1225
1226                 if (__this_cpu_inc_return(xed_nesting_count) - 1)
1227                         goto out;
1228
1229                 xen_evtchn_handle_events(cpu);
1230
1231                 BUG_ON(!irqs_disabled());
1232
1233                 count = __this_cpu_read(xed_nesting_count);
1234                 __this_cpu_write(xed_nesting_count, 0);
1235         } while (count != 1 || vcpu_info->evtchn_upcall_pending);
1236
1237 out:
1238
1239         put_cpu();
1240 }
1241
1242 void xen_evtchn_do_upcall(struct pt_regs *regs)
1243 {
1244         struct pt_regs *old_regs = set_irq_regs(regs);
1245
1246         irq_enter();
1247 #ifdef CONFIG_X86
1248         exit_idle();
1249         inc_irq_stat(irq_hv_callback_count);
1250 #endif
1251
1252         __xen_evtchn_do_upcall();
1253
1254         irq_exit();
1255         set_irq_regs(old_regs);
1256 }
1257
1258 void xen_hvm_evtchn_do_upcall(void)
1259 {
1260         __xen_evtchn_do_upcall();
1261 }
1262 EXPORT_SYMBOL_GPL(xen_hvm_evtchn_do_upcall);
1263
1264 /* Rebind a new event channel to an existing irq. */
1265 void rebind_evtchn_irq(int evtchn, int irq)
1266 {
1267         struct irq_info *info = info_for_irq(irq);
1268
1269         if (WARN_ON(!info))
1270                 return;
1271
1272         /* Make sure the irq is masked, since the new event channel
1273            will also be masked. */
1274         disable_irq(irq);
1275
1276         mutex_lock(&irq_mapping_update_lock);
1277
1278         /* After resume the irq<->evtchn mappings are all cleared out */
1279         BUG_ON(get_evtchn_to_irq(evtchn) != -1);
1280         /* Expect irq to have been bound before,
1281            so there should be a proper type */
1282         BUG_ON(info->type == IRQT_UNBOUND);
1283
1284         (void)xen_irq_info_evtchn_setup(irq, evtchn);
1285
1286         mutex_unlock(&irq_mapping_update_lock);
1287
1288         bind_evtchn_to_cpu(evtchn, info->cpu);
1289         /* This will be deferred until interrupt is processed */
1290         irq_set_affinity(irq, cpumask_of(info->cpu));
1291
1292         /* Unmask the event channel. */
1293         enable_irq(irq);
1294 }
1295
1296 /* Rebind an evtchn so that it gets delivered to a specific cpu */
1297 static int rebind_irq_to_cpu(unsigned irq, unsigned tcpu)
1298 {
1299         struct evtchn_bind_vcpu bind_vcpu;
1300         int evtchn = evtchn_from_irq(irq);
1301         int masked;
1302
1303         if (!VALID_EVTCHN(evtchn))
1304                 return -1;
1305
1306         /*
1307          * Events delivered via platform PCI interrupts are always
1308          * routed to vcpu 0 and hence cannot be rebound.
1309          */
1310         if (xen_hvm_domain() && !xen_have_vector_callback)
1311                 return -1;
1312
1313         /* Send future instances of this interrupt to other vcpu. */
1314         bind_vcpu.port = evtchn;
1315         bind_vcpu.vcpu = tcpu;
1316
1317         /*
1318          * Mask the event while changing the VCPU binding to prevent
1319          * it being delivered on an unexpected VCPU.
1320          */
1321         masked = test_and_set_mask(evtchn);
1322
1323         /*
1324          * If this fails, it usually just indicates that we're dealing with a
1325          * virq or IPI channel, which don't actually need to be rebound. Ignore
1326          * it, but don't do the xenlinux-level rebind in that case.
1327          */
1328         if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_vcpu, &bind_vcpu) >= 0)
1329                 bind_evtchn_to_cpu(evtchn, tcpu);
1330
1331         if (!masked)
1332                 unmask_evtchn(evtchn);
1333
1334         return 0;
1335 }
1336
1337 static int set_affinity_irq(struct irq_data *data, const struct cpumask *dest,
1338                             bool force)
1339 {
1340         unsigned tcpu = cpumask_first_and(dest, cpu_online_mask);
1341
1342         return rebind_irq_to_cpu(data->irq, tcpu);
1343 }
1344
1345 static void enable_dynirq(struct irq_data *data)
1346 {
1347         int evtchn = evtchn_from_irq(data->irq);
1348
1349         if (VALID_EVTCHN(evtchn))
1350                 unmask_evtchn(evtchn);
1351 }
1352
1353 static void disable_dynirq(struct irq_data *data)
1354 {
1355         int evtchn = evtchn_from_irq(data->irq);
1356
1357         if (VALID_EVTCHN(evtchn))
1358                 mask_evtchn(evtchn);
1359 }
1360
1361 static void ack_dynirq(struct irq_data *data)
1362 {
1363         int evtchn = evtchn_from_irq(data->irq);
1364
1365         irq_move_irq(data);
1366
1367         if (VALID_EVTCHN(evtchn))
1368                 clear_evtchn(evtchn);
1369 }
1370
1371 static void mask_ack_dynirq(struct irq_data *data)
1372 {
1373         disable_dynirq(data);
1374         ack_dynirq(data);
1375 }
1376
1377 static int retrigger_dynirq(struct irq_data *data)
1378 {
1379         unsigned int evtchn = evtchn_from_irq(data->irq);
1380         int masked;
1381
1382         if (!VALID_EVTCHN(evtchn))
1383                 return 0;
1384
1385         masked = test_and_set_mask(evtchn);
1386         set_evtchn(evtchn);
1387         if (!masked)
1388                 unmask_evtchn(evtchn);
1389
1390         return 1;
1391 }
1392
1393 static void restore_pirqs(void)
1394 {
1395         int pirq, rc, irq, gsi;
1396         struct physdev_map_pirq map_irq;
1397         struct irq_info *info;
1398
1399         list_for_each_entry(info, &xen_irq_list_head, list) {
1400                 if (info->type != IRQT_PIRQ)
1401                         continue;
1402
1403                 pirq = info->u.pirq.pirq;
1404                 gsi = info->u.pirq.gsi;
1405                 irq = info->irq;
1406
1407                 /* save/restore of PT devices doesn't work, so at this point the
1408                  * only devices present are GSI based emulated devices */
1409                 if (!gsi)
1410                         continue;
1411
1412                 map_irq.domid = DOMID_SELF;
1413                 map_irq.type = MAP_PIRQ_TYPE_GSI;
1414                 map_irq.index = gsi;
1415                 map_irq.pirq = pirq;
1416
1417                 rc = HYPERVISOR_physdev_op(PHYSDEVOP_map_pirq, &map_irq);
1418                 if (rc) {
1419                         pr_warn("xen map irq failed gsi=%d irq=%d pirq=%d rc=%d\n",
1420                                 gsi, irq, pirq, rc);
1421                         xen_free_irq(irq);
1422                         continue;
1423                 }
1424
1425                 printk(KERN_DEBUG "xen: --> irq=%d, pirq=%d\n", irq, map_irq.pirq);
1426
1427                 __startup_pirq(irq);
1428         }
1429 }
1430
1431 static void restore_cpu_virqs(unsigned int cpu)
1432 {
1433         struct evtchn_bind_virq bind_virq;
1434         int virq, irq, evtchn;
1435
1436         for (virq = 0; virq < NR_VIRQS; virq++) {
1437                 if ((irq = per_cpu(virq_to_irq, cpu)[virq]) == -1)
1438                         continue;
1439
1440                 BUG_ON(virq_from_irq(irq) != virq);
1441
1442                 /* Get a new binding from Xen. */
1443                 bind_virq.virq = virq;
1444                 bind_virq.vcpu = cpu;
1445                 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_virq,
1446                                                 &bind_virq) != 0)
1447                         BUG();
1448                 evtchn = bind_virq.port;
1449
1450                 /* Record the new mapping. */
1451                 (void)xen_irq_info_virq_setup(cpu, irq, evtchn, virq);
1452                 bind_evtchn_to_cpu(evtchn, cpu);
1453         }
1454 }
1455
1456 static void restore_cpu_ipis(unsigned int cpu)
1457 {
1458         struct evtchn_bind_ipi bind_ipi;
1459         int ipi, irq, evtchn;
1460
1461         for (ipi = 0; ipi < XEN_NR_IPIS; ipi++) {
1462                 if ((irq = per_cpu(ipi_to_irq, cpu)[ipi]) == -1)
1463                         continue;
1464
1465                 BUG_ON(ipi_from_irq(irq) != ipi);
1466
1467                 /* Get a new binding from Xen. */
1468                 bind_ipi.vcpu = cpu;
1469                 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_ipi,
1470                                                 &bind_ipi) != 0)
1471                         BUG();
1472                 evtchn = bind_ipi.port;
1473
1474                 /* Record the new mapping. */
1475                 (void)xen_irq_info_ipi_setup(cpu, irq, evtchn, ipi);
1476                 bind_evtchn_to_cpu(evtchn, cpu);
1477         }
1478 }
1479
1480 /* Clear an irq's pending state, in preparation for polling on it */
1481 void xen_clear_irq_pending(int irq)
1482 {
1483         int evtchn = evtchn_from_irq(irq);
1484
1485         if (VALID_EVTCHN(evtchn))
1486                 clear_evtchn(evtchn);
1487 }
1488 EXPORT_SYMBOL(xen_clear_irq_pending);
1489 void xen_set_irq_pending(int irq)
1490 {
1491         int evtchn = evtchn_from_irq(irq);
1492
1493         if (VALID_EVTCHN(evtchn))
1494                 set_evtchn(evtchn);
1495 }
1496
1497 bool xen_test_irq_pending(int irq)
1498 {
1499         int evtchn = evtchn_from_irq(irq);
1500         bool ret = false;
1501
1502         if (VALID_EVTCHN(evtchn))
1503                 ret = test_evtchn(evtchn);
1504
1505         return ret;
1506 }
1507
1508 /* Poll waiting for an irq to become pending with timeout.  In the usual case,
1509  * the irq will be disabled so it won't deliver an interrupt. */
1510 void xen_poll_irq_timeout(int irq, u64 timeout)
1511 {
1512         evtchn_port_t evtchn = evtchn_from_irq(irq);
1513
1514         if (VALID_EVTCHN(evtchn)) {
1515                 struct sched_poll poll;
1516
1517                 poll.nr_ports = 1;
1518                 poll.timeout = timeout;
1519                 set_xen_guest_handle(poll.ports, &evtchn);
1520
1521                 if (HYPERVISOR_sched_op(SCHEDOP_poll, &poll) != 0)
1522                         BUG();
1523         }
1524 }
1525 EXPORT_SYMBOL(xen_poll_irq_timeout);
1526 /* Poll waiting for an irq to become pending.  In the usual case, the
1527  * irq will be disabled so it won't deliver an interrupt. */
1528 void xen_poll_irq(int irq)
1529 {
1530         xen_poll_irq_timeout(irq, 0 /* no timeout */);
1531 }
1532
1533 /* Check whether the IRQ line is shared with other guests. */
1534 int xen_test_irq_shared(int irq)
1535 {
1536         struct irq_info *info = info_for_irq(irq);
1537         struct physdev_irq_status_query irq_status;
1538
1539         if (WARN_ON(!info))
1540                 return -ENOENT;
1541
1542         irq_status.irq = info->u.pirq.pirq;
1543
1544         if (HYPERVISOR_physdev_op(PHYSDEVOP_irq_status_query, &irq_status))
1545                 return 0;
1546         return !(irq_status.flags & XENIRQSTAT_shared);
1547 }
1548 EXPORT_SYMBOL_GPL(xen_test_irq_shared);
1549
1550 void xen_irq_resume(void)
1551 {
1552         unsigned int cpu;
1553         struct irq_info *info;
1554
1555         /* New event-channel space is not 'live' yet. */
1556         xen_evtchn_mask_all();
1557         xen_evtchn_resume();
1558
1559         /* No IRQ <-> event-channel mappings. */
1560         list_for_each_entry(info, &xen_irq_list_head, list)
1561                 info->evtchn = 0; /* zap event-channel binding */
1562
1563         clear_evtchn_to_irq_all();
1564
1565         for_each_possible_cpu(cpu) {
1566                 restore_cpu_virqs(cpu);
1567                 restore_cpu_ipis(cpu);
1568         }
1569
1570         restore_pirqs();
1571 }
1572
1573 static struct irq_chip xen_dynamic_chip __read_mostly = {
1574         .name                   = "xen-dyn",
1575
1576         .irq_disable            = disable_dynirq,
1577         .irq_mask               = disable_dynirq,
1578         .irq_unmask             = enable_dynirq,
1579
1580         .irq_ack                = ack_dynirq,
1581         .irq_mask_ack           = mask_ack_dynirq,
1582
1583         .irq_set_affinity       = set_affinity_irq,
1584         .irq_retrigger          = retrigger_dynirq,
1585 };
1586
1587 static struct irq_chip xen_pirq_chip __read_mostly = {
1588         .name                   = "xen-pirq",
1589
1590         .irq_startup            = startup_pirq,
1591         .irq_shutdown           = shutdown_pirq,
1592         .irq_enable             = enable_pirq,
1593         .irq_disable            = disable_pirq,
1594
1595         .irq_mask               = disable_dynirq,
1596         .irq_unmask             = enable_dynirq,
1597
1598         .irq_ack                = eoi_pirq,
1599         .irq_eoi                = eoi_pirq,
1600         .irq_mask_ack           = mask_ack_pirq,
1601
1602         .irq_set_affinity       = set_affinity_irq,
1603
1604         .irq_retrigger          = retrigger_dynirq,
1605 };
1606
1607 static struct irq_chip xen_percpu_chip __read_mostly = {
1608         .name                   = "xen-percpu",
1609
1610         .irq_disable            = disable_dynirq,
1611         .irq_mask               = disable_dynirq,
1612         .irq_unmask             = enable_dynirq,
1613
1614         .irq_ack                = ack_dynirq,
1615 };
1616
1617 int xen_set_callback_via(uint64_t via)
1618 {
1619         struct xen_hvm_param a;
1620         a.domid = DOMID_SELF;
1621         a.index = HVM_PARAM_CALLBACK_IRQ;
1622         a.value = via;
1623         return HYPERVISOR_hvm_op(HVMOP_set_param, &a);
1624 }
1625 EXPORT_SYMBOL_GPL(xen_set_callback_via);
1626
1627 #ifdef CONFIG_XEN_PVHVM
1628 /* Vector callbacks are better than PCI interrupts to receive event
1629  * channel notifications because we can receive vector callbacks on any
1630  * vcpu and we don't need PCI support or APIC interactions. */
1631 void xen_callback_vector(void)
1632 {
1633         int rc;
1634         uint64_t callback_via;
1635         if (xen_have_vector_callback) {
1636                 callback_via = HVM_CALLBACK_VECTOR(HYPERVISOR_CALLBACK_VECTOR);
1637                 rc = xen_set_callback_via(callback_via);
1638                 if (rc) {
1639                         pr_err("Request for Xen HVM callback vector failed\n");
1640                         xen_have_vector_callback = 0;
1641                         return;
1642                 }
1643                 pr_info("Xen HVM callback vector for event delivery is enabled\n");
1644                 /* in the restore case the vector has already been allocated */
1645                 if (!test_bit(HYPERVISOR_CALLBACK_VECTOR, used_vectors))
1646                         alloc_intr_gate(HYPERVISOR_CALLBACK_VECTOR,
1647                                         xen_hvm_callback_vector);
1648         }
1649 }
1650 #else
1651 void xen_callback_vector(void) {}
1652 #endif
1653
1654 #undef MODULE_PARAM_PREFIX
1655 #define MODULE_PARAM_PREFIX "xen."
1656
1657 static bool fifo_events = true;
1658 module_param(fifo_events, bool, 0);
1659
1660 void __init xen_init_IRQ(void)
1661 {
1662         int ret = -EINVAL;
1663
1664         if (fifo_events)
1665                 ret = xen_evtchn_fifo_init();
1666         if (ret < 0)
1667                 xen_evtchn_2l_init();
1668
1669         evtchn_to_irq = kcalloc(EVTCHN_ROW(xen_evtchn_max_channels()),
1670                                 sizeof(*evtchn_to_irq), GFP_KERNEL);
1671         BUG_ON(!evtchn_to_irq);
1672
1673         /* No event channels are 'live' right now. */
1674         xen_evtchn_mask_all();
1675
1676         pirq_needs_eoi = pirq_needs_eoi_flag;
1677
1678 #ifdef CONFIG_X86
1679         if (xen_pv_domain()) {
1680                 irq_ctx_init(smp_processor_id());
1681                 if (xen_initial_domain())
1682                         pci_xen_initial_domain();
1683         }
1684         if (xen_feature(XENFEAT_hvm_callback_vector))
1685                 xen_callback_vector();
1686
1687         if (xen_hvm_domain()) {
1688                 native_init_IRQ();
1689                 /* pci_xen_hvm_init must be called after native_init_IRQ so that
1690                  * __acpi_register_gsi can point at the right function */
1691                 pci_xen_hvm_init();
1692         } else {
1693                 int rc;
1694                 struct physdev_pirq_eoi_gmfn eoi_gmfn;
1695
1696                 pirq_eoi_map = (void *)__get_free_page(GFP_KERNEL|__GFP_ZERO);
1697                 eoi_gmfn.gmfn = virt_to_mfn(pirq_eoi_map);
1698                 rc = HYPERVISOR_physdev_op(PHYSDEVOP_pirq_eoi_gmfn_v2, &eoi_gmfn);
1699                 /* TODO: No PVH support for PIRQ EOI */
1700                 if (rc != 0) {
1701                         free_page((unsigned long) pirq_eoi_map);
1702                         pirq_eoi_map = NULL;
1703                 } else
1704                         pirq_needs_eoi = pirq_check_eoi_map;
1705         }
1706 #endif
1707 }