drivers/xen/events/events_base.c

   1 /*
   2  * Xen event channels
   3  *
   4  * Xen models interrupts with abstract event channels.  Because each
   5  * domain gets 1024 event channels, but NR_IRQ is not that large, we
   6  * must dynamically map irqs<->event channels.  The event channels
   7  * interface with the rest of the kernel by defining a xen interrupt
   8  * chip.  When an event is received, it is mapped to an irq and sent
   9  * through the normal interrupt processing path.
  10  *
  11  * There are four kinds of events which can be mapped to an event
  12  * channel:
  13  *
  14  * 1. Inter-domain notifications.  This includes all the virtual
  15  *    device events, since they're driven by front-ends in another domain
  16  *    (typically dom0).
  17  * 2. VIRQs, typically used for timers.  These are per-cpu events.
  18  * 3. IPIs.
  19  * 4. PIRQs - Hardware interrupts.
  20  *
  21  * Jeremy Fitzhardinge <jeremy@xensource.com>, XenSource Inc, 2007
  22  */
  23
  24 #define pr_fmt(fmt) "xen:" KBUILD_MODNAME ": " fmt
  25
  26 #include <linux/linkage.h>
  27 #include <linux/interrupt.h>
  28 #include <linux/irq.h>
  29 #include <linux/module.h>
  30 #include <linux/string.h>
  31 #include <linux/bootmem.h>
  32 #include <linux/slab.h>
  33 #include <linux/irqnr.h>
  34 #include <linux/pci.h>
  35
  36 #ifdef CONFIG_X86
  37 #include <asm/desc.h>
  38 #include <asm/ptrace.h>
  39 #include <asm/irq.h>
  40 #include <asm/idle.h>
  41 #include <asm/io_apic.h>
  42 #include <asm/xen/page.h>
  43 #include <asm/xen/pci.h>
  44 #endif
  45 #include <asm/sync_bitops.h>
  46 #include <asm/xen/hypercall.h>
  47 #include <asm/xen/hypervisor.h>
  48
  49 #include <xen/xen.h>
  50 #include <xen/hvm.h>
  51 #include <xen/xen-ops.h>
  52 #include <xen/events.h>
  53 #include <xen/interface/xen.h>
  54 #include <xen/interface/event_channel.h>
  55 #include <xen/interface/hvm/hvm_op.h>
  56 #include <xen/interface/hvm/params.h>
  57 #include <xen/interface/physdev.h>
  58 #include <xen/interface/sched.h>
  59 #include <xen/interface/vcpu.h>
  60 #include <asm/hw_irq.h>
  61
  62 #include "events_internal.h"
  63
  64 const struct evtchn_ops *evtchn_ops;
  65
  66 /*
  67  * This lock protects updates to the following mapping and reference-count
  68  * arrays. The lock does not need to be acquired to read the mapping tables.
  69  */
  70 static DEFINE_MUTEX(irq_mapping_update_lock);
  71
  72 static LIST_HEAD(xen_irq_list_head);
  73
  74 /* IRQ <-> VIRQ mapping. */
  75 static DEFINE_PER_CPU(int [NR_VIRQS], virq_to_irq) = {[0 ... NR_VIRQS-1] = -1};
  76
  77 /* IRQ <-> IPI mapping */
  78 static DEFINE_PER_CPU(int [XEN_NR_IPIS], ipi_to_irq) = {[0 ... XEN_NR_IPIS-1] = -1};
  79
  80 int **evtchn_to_irq;
  81 #ifdef CONFIG_X86
  82 static unsigned long *pirq_eoi_map;
  83 #endif
  84 static bool (*pirq_needs_eoi)(unsigned irq);
  85
  86 #define EVTCHN_ROW(e)  (e / (PAGE_SIZE/sizeof(**evtchn_to_irq)))
  87 #define EVTCHN_COL(e)  (e % (PAGE_SIZE/sizeof(**evtchn_to_irq)))
  88 #define EVTCHN_PER_ROW (PAGE_SIZE / sizeof(**evtchn_to_irq))
  89
  90 /* Xen will never allocate port zero for any purpose. */
  91 #define VALID_EVTCHN(chn)       ((chn) != 0)
  92
  93 static struct irq_chip xen_dynamic_chip;
  94 static struct irq_chip xen_percpu_chip;
  95 static struct irq_chip xen_pirq_chip;
  96 static void enable_dynirq(struct irq_data *data);
  97 static void disable_dynirq(struct irq_data *data);
  98
  99 static void clear_evtchn_to_irq_row(unsigned row)
 100 {
 101         unsigned col;
 102
 103         for (col = 0; col < EVTCHN_PER_ROW; col++)
 104                 evtchn_to_irq[row][col] = -1;
 105 }
 106
 107 static void clear_evtchn_to_irq_all(void)
 108 {
 109         unsigned row;
 110
 111         for (row = 0; row < EVTCHN_ROW(xen_evtchn_max_channels()); row++) {
 112                 if (evtchn_to_irq[row] == NULL)
 113                         continue;
 114                 clear_evtchn_to_irq_row(row);
 115         }
 116 }
 117
 118 static int set_evtchn_to_irq(unsigned evtchn, unsigned irq)
 119 {
 120         unsigned row;
 121         unsigned col;
 122
 123         if (evtchn >= xen_evtchn_max_channels())
 124                 return -EINVAL;
 125
 126         row = EVTCHN_ROW(evtchn);
 127         col = EVTCHN_COL(evtchn);
 128
 129         if (evtchn_to_irq[row] == NULL) {
 130                 /* Unallocated irq entries return -1 anyway */
 131                 if (irq == -1)
 132                         return 0;
 133
 134                 evtchn_to_irq[row] = (int *)get_zeroed_page(GFP_KERNEL);
 135                 if (evtchn_to_irq[row] == NULL)
 136                         return -ENOMEM;
 137
 138                 clear_evtchn_to_irq_row(row);
 139         }
 140
 141         evtchn_to_irq[EVTCHN_ROW(evtchn)][EVTCHN_COL(evtchn)] = irq;
 142         return 0;
 143 }
 144
 145 int get_evtchn_to_irq(unsigned evtchn)
 146 {
 147         if (evtchn >= xen_evtchn_max_channels())
 148                 return -1;
 149         if (evtchn_to_irq[EVTCHN_ROW(evtchn)] == NULL)
 150                 return -1;
 151         return evtchn_to_irq[EVTCHN_ROW(evtchn)][EVTCHN_COL(evtchn)];
 152 }
 153
 154 /* Get info for IRQ */
 155 struct irq_info *info_for_irq(unsigned irq)
 156 {
 157         return irq_get_handler_data(irq);
 158 }
 159
 160 /* Constructors for packed IRQ information. */
 161 static int xen_irq_info_common_setup(struct irq_info *info,
 162                                      unsigned irq,
 163                                      enum xen_irq_type type,
 164                                      unsigned evtchn,
 165                                      unsigned short cpu)
 166 {
 167         int ret;
 168
 169         BUG_ON(info->type != IRQT_UNBOUND && info->type != type);
 170
 171         info->type = type;
 172         info->irq = irq;
 173         info->evtchn = evtchn;
 174         info->cpu = cpu;
 175
 176         ret = set_evtchn_to_irq(evtchn, irq);
 177         if (ret < 0)
 178                 return ret;
 179
 180         irq_clear_status_flags(irq, IRQ_NOREQUEST|IRQ_NOAUTOEN);
 181
 182         return xen_evtchn_port_setup(info);
 183 }
 184
 185 static int xen_irq_info_evtchn_setup(unsigned irq,
 186                                      unsigned evtchn)
 187 {
 188         struct irq_info *info = info_for_irq(irq);
 189
 190         return xen_irq_info_common_setup(info, irq, IRQT_EVTCHN, evtchn, 0);
 191 }
 192
 193 static int xen_irq_info_ipi_setup(unsigned cpu,
 194                                   unsigned irq,
 195                                   unsigned evtchn,
 196                                   enum ipi_vector ipi)
 197 {
 198         struct irq_info *info = info_for_irq(irq);
 199
 200         info->u.ipi = ipi;
 201
 202         per_cpu(ipi_to_irq, cpu)[ipi] = irq;
 203
 204         return xen_irq_info_common_setup(info, irq, IRQT_IPI, evtchn, 0);
 205 }
 206
 207 static int xen_irq_info_virq_setup(unsigned cpu,
 208                                    unsigned irq,
 209                                    unsigned evtchn,
 210                                    unsigned virq)
 211 {
 212         struct irq_info *info = info_for_irq(irq);
 213
 214         info->u.virq = virq;
 215
 216         per_cpu(virq_to_irq, cpu)[virq] = irq;
 217
 218         return xen_irq_info_common_setup(info, irq, IRQT_VIRQ, evtchn, 0);
 219 }
 220
 221 static int xen_irq_info_pirq_setup(unsigned irq,
 222                                    unsigned evtchn,
 223                                    unsigned pirq,
 224                                    unsigned gsi,
 225                                    uint16_t domid,
 226                                    unsigned char flags)
 227 {
 228         struct irq_info *info = info_for_irq(irq);
 229
 230         info->u.pirq.pirq = pirq;
 231         info->u.pirq.gsi = gsi;
 232         info->u.pirq.domid = domid;
 233         info->u.pirq.flags = flags;
 234
 235         return xen_irq_info_common_setup(info, irq, IRQT_PIRQ, evtchn, 0);
 236 }
 237
 238 static void xen_irq_info_cleanup(struct irq_info *info)
 239 {
 240         set_evtchn_to_irq(info->evtchn, -1);
 241         info->evtchn = 0;
 242 }
 243
 244 /*
 245  * Accessors for packed IRQ information.
 246  */
 247 unsigned int evtchn_from_irq(unsigned irq)
 248 {
 249         if (unlikely(WARN(irq < 0 || irq >= nr_irqs, "Invalid irq %d!\n", irq)))
 250                 return 0;
 251
 252         return info_for_irq(irq)->evtchn;
 253 }
 254
 255 unsigned irq_from_evtchn(unsigned int evtchn)
 256 {
 257         return get_evtchn_to_irq(evtchn);
 258 }
 259 EXPORT_SYMBOL_GPL(irq_from_evtchn);
 260
 261 int irq_from_virq(unsigned int cpu, unsigned int virq)
 262 {
 263         return per_cpu(virq_to_irq, cpu)[virq];
 264 }
 265
 266 static enum ipi_vector ipi_from_irq(unsigned irq)
 267 {
 268         struct irq_info *info = info_for_irq(irq);
 269
 270         BUG_ON(info == NULL);
 271         BUG_ON(info->type != IRQT_IPI);
 272
 273         return info->u.ipi;
 274 }
 275
 276 static unsigned virq_from_irq(unsigned irq)
 277 {
 278         struct irq_info *info = info_for_irq(irq);
 279
 280         BUG_ON(info == NULL);
 281         BUG_ON(info->type != IRQT_VIRQ);
 282
 283         return info->u.virq;
 284 }
 285
 286 static unsigned pirq_from_irq(unsigned irq)
 287 {
 288         struct irq_info *info = info_for_irq(irq);
 289
 290         BUG_ON(info == NULL);
 291         BUG_ON(info->type != IRQT_PIRQ);
 292
 293         return info->u.pirq.pirq;
 294 }
 295
 296 static enum xen_irq_type type_from_irq(unsigned irq)
 297 {
 298         return info_for_irq(irq)->type;
 299 }
 300
 301 unsigned cpu_from_irq(unsigned irq)
 302 {
 303         return info_for_irq(irq)->cpu;
 304 }
 305
 306 unsigned int cpu_from_evtchn(unsigned int evtchn)
 307 {
 308         int irq = get_evtchn_to_irq(evtchn);
 309         unsigned ret = 0;
 310
 311         if (irq != -1)
 312                 ret = cpu_from_irq(irq);
 313
 314         return ret;
 315 }
 316
 317 #ifdef CONFIG_X86
 318 static bool pirq_check_eoi_map(unsigned irq)
 319 {
 320         return test_bit(pirq_from_irq(irq), pirq_eoi_map);
 321 }
 322 #endif
 323
 324 static bool pirq_needs_eoi_flag(unsigned irq)
 325 {
 326         struct irq_info *info = info_for_irq(irq);
 327         BUG_ON(info->type != IRQT_PIRQ);
 328
 329         return info->u.pirq.flags & PIRQ_NEEDS_EOI;
 330 }
 331
 332 static void bind_evtchn_to_cpu(unsigned int chn, unsigned int cpu)
 333 {
 334         int irq = get_evtchn_to_irq(chn);
 335         struct irq_info *info = info_for_irq(irq);
 336
 337         BUG_ON(irq == -1);
 338 #ifdef CONFIG_SMP
 339         cpumask_copy(irq_to_desc(irq)->irq_data.affinity, cpumask_of(cpu));
 340 #endif
 341
 342         xen_evtchn_port_bind_to_cpu(info, cpu);
 343
 344         info->cpu = cpu;
 345 }
 346
 347 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         struct irq_desc *desc = irq_to_desc(irq);
 377
 378         /* By default all event channels notify CPU#0. */
 379         cpumask_copy(desc->irq_data.affinity, cpumask_of(0));
 380 #endif
 381
 382         info = kzalloc(sizeof(*info), GFP_KERNEL);
 383         if (info == NULL)
 384                 panic("Unable to allocate metadata for IRQ%d\n", irq);
 385
 386         info->type = IRQT_UNBOUND;
 387         info->refcnt = -1;
 388
 389         irq_set_handler_data(irq, info);
 390
 391         list_add_tail(&info->list, &xen_irq_list_head);
 392 }
 393
 394 static int __must_check xen_allocate_irq_dynamic(void)
 395 {
 396         int first = 0;
 397         int irq;
 398
 399 #ifdef CONFIG_X86_IO_APIC
 400         /*
 401          * For an HVM guest or domain 0 which see "real" (emulated or
 402          * actual respectively) GSIs we allocate dynamic IRQs
 403          * e.g. those corresponding to event channels or MSIs
 404          * etc. from the range above those "real" GSIs to avoid
 405          * collisions.
 406          */
 407         if (xen_initial_domain() || xen_hvm_domain())
 408                 first = get_nr_irqs_gsi();
 409 #endif
 410
 411         irq = irq_alloc_desc_from(first, -1);
 412
 413         if (irq >= 0)
 414                 xen_irq_init(irq);
 415
 416         return irq;
 417 }
 418
 419 static int __must_check xen_allocate_irq_gsi(unsigned gsi)
 420 {
 421         int irq;
 422
 423         /*
 424          * A PV guest has no concept of a GSI (since it has no ACPI
 425          * nor access to/knowledge of the physical APICs). Therefore
 426          * all IRQs are dynamically allocated from the entire IRQ
 427          * space.
 428          */
 429         if (xen_pv_domain() && !xen_initial_domain())
 430                 return xen_allocate_irq_dynamic();
 431
 432         /* Legacy IRQ descriptors are already allocated by the arch. */
 433         if (gsi < NR_IRQS_LEGACY)
 434                 irq = gsi;
 435         else
 436                 irq = irq_alloc_desc_at(gsi, -1);
 437
 438         xen_irq_init(irq);
 439
 440         return irq;
 441 }
 442
 443 static void xen_free_irq(unsigned irq)
 444 {
 445         struct irq_info *info = irq_get_handler_data(irq);
 446
 447         if (WARN_ON(!info))
 448                 return;
 449
 450         list_del(&info->list);
 451
 452         irq_set_handler_data(irq, NULL);
 453
 454         WARN_ON(info->refcnt > 0);
 455
 456         kfree(info);
 457
 458         /* Legacy IRQ descriptors are managed by the arch. */
 459         if (irq < NR_IRQS_LEGACY)
 460                 return;
 461
 462         irq_free_desc(irq);
 463 }
 464
 465 static void xen_evtchn_close(unsigned int port)
 466 {
 467         struct evtchn_close close;
 468
 469         close.port = port;
 470         if (HYPERVISOR_event_channel_op(EVTCHNOP_close, &close) != 0)
 471                 BUG();
 472
 473         /* Closed ports are implicitly re-bound to VCPU0. */
 474         bind_evtchn_to_cpu(port, 0);
 475 }
 476
 477 static void pirq_query_unmask(int irq)
 478 {
 479         struct physdev_irq_status_query irq_status;
 480         struct irq_info *info = info_for_irq(irq);
 481
 482         BUG_ON(info->type != IRQT_PIRQ);
 483
 484         irq_status.irq = pirq_from_irq(irq);
 485         if (HYPERVISOR_physdev_op(PHYSDEVOP_irq_status_query, &irq_status))
 486                 irq_status.flags = 0;
 487
 488         info->u.pirq.flags &= ~PIRQ_NEEDS_EOI;
 489         if (irq_status.flags & XENIRQSTAT_needs_eoi)
 490                 info->u.pirq.flags |= PIRQ_NEEDS_EOI;
 491 }
 492
 493 static bool probing_irq(int irq)
 494 {
 495         struct irq_desc *desc = irq_to_desc(irq);
 496
 497         return desc && desc->action == NULL;
 498 }
 499
 500 static void eoi_pirq(struct irq_data *data)
 501 {
 502         int evtchn = evtchn_from_irq(data->irq);
 503         struct physdev_eoi eoi = { .irq = pirq_from_irq(data->irq) };
 504         int rc = 0;
 505
 506         irq_move_irq(data);
 507
 508         if (VALID_EVTCHN(evtchn))
 509                 clear_evtchn(evtchn);
 510
 511         if (pirq_needs_eoi(data->irq)) {
 512                 rc = HYPERVISOR_physdev_op(PHYSDEVOP_eoi, &eoi);
 513                 WARN_ON(rc);
 514         }
 515 }
 516
 517 static void mask_ack_pirq(struct irq_data *data)
 518 {
 519         disable_dynirq(data);
 520         eoi_pirq(data);
 521 }
 522
 523 static unsigned int __startup_pirq(unsigned int irq)
 524 {
 525         struct evtchn_bind_pirq bind_pirq;
 526         struct irq_info *info = info_for_irq(irq);
 527         int evtchn = evtchn_from_irq(irq);
 528         int rc;
 529
 530         BUG_ON(info->type != IRQT_PIRQ);
 531
 532         if (VALID_EVTCHN(evtchn))
 533                 goto out;
 534
 535         bind_pirq.pirq = pirq_from_irq(irq);
 536         /* NB. We are happy to share unless we are probing. */
 537         bind_pirq.flags = info->u.pirq.flags & PIRQ_SHAREABLE ?
 538                                         BIND_PIRQ__WILL_SHARE : 0;
 539         rc = HYPERVISOR_event_channel_op(EVTCHNOP_bind_pirq, &bind_pirq);
 540         if (rc != 0) {
 541                 if (!probing_irq(irq))
 542                         pr_info("Failed to obtain physical IRQ %d\n", irq);
 543                 return 0;
 544         }
 545         evtchn = bind_pirq.port;
 546
 547         pirq_query_unmask(irq);
 548
 549         rc = set_evtchn_to_irq(evtchn, irq);
 550         if (rc != 0) {
 551                 pr_err("irq%d: Failed to set port to irq mapping (%d)\n",
 552                        irq, rc);
 553                 xen_evtchn_close(evtchn);
 554                 return 0;
 555         }
 556         bind_evtchn_to_cpu(evtchn, 0);
 557         info->evtchn = evtchn;
 558
 559 out:
 560         unmask_evtchn(evtchn);
 561         eoi_pirq(irq_get_irq_data(irq));
 562
 563         return 0;
 564 }
 565
 566 static unsigned int startup_pirq(struct irq_data *data)
 567 {
 568         return __startup_pirq(data->irq);
 569 }
 570
 571 static void shutdown_pirq(struct irq_data *data)
 572 {
 573         unsigned int irq = data->irq;
 574         struct irq_info *info = info_for_irq(irq);
 575         unsigned evtchn = evtchn_from_irq(irq);
 576
 577         BUG_ON(info->type != IRQT_PIRQ);
 578
 579         if (!VALID_EVTCHN(evtchn))
 580                 return;
 581
 582         mask_evtchn(evtchn);
 583         xen_evtchn_close(evtchn);
 584         xen_irq_info_cleanup(info);
 585 }
 586
 587 static void enable_pirq(struct irq_data *data)
 588 {
 589         startup_pirq(data);
 590 }
 591
 592 static void disable_pirq(struct irq_data *data)
 593 {
 594         disable_dynirq(data);
 595 }
 596
 597 int xen_irq_from_gsi(unsigned gsi)
 598 {
 599         struct irq_info *info;
 600
 601         list_for_each_entry(info, &xen_irq_list_head, list) {
 602                 if (info->type != IRQT_PIRQ)
 603                         continue;
 604
 605                 if (info->u.pirq.gsi == gsi)
 606                         return info->irq;
 607         }
 608
 609         return -1;
 610 }
 611 EXPORT_SYMBOL_GPL(xen_irq_from_gsi);
 612
 613 static void __unbind_from_irq(unsigned int irq)
 614 {
 615         int evtchn = evtchn_from_irq(irq);
 616         struct irq_info *info = irq_get_handler_data(irq);
 617
 618         if (info->refcnt > 0) {
 619                 info->refcnt--;
 620                 if (info->refcnt != 0)
 621                         return;
 622         }
 623
 624         if (VALID_EVTCHN(evtchn)) {
 625                 unsigned int cpu = cpu_from_irq(irq);
 626
 627                 xen_evtchn_close(evtchn);
 628
 629                 switch (type_from_irq(irq)) {
 630                 case IRQT_VIRQ:
 631                         per_cpu(virq_to_irq, cpu)[virq_from_irq(irq)] = -1;
 632                         break;
 633                 case IRQT_IPI:
 634                         per_cpu(ipi_to_irq, cpu)[ipi_from_irq(irq)] = -1;
 635                         break;
 636                 default:
 637                         break;
 638                 }
 639
 640                 xen_irq_info_cleanup(info);
 641         }
 642
 643         BUG_ON(info_for_irq(irq)->type == IRQT_UNBOUND);
 644
 645         xen_free_irq(irq);
 646 }
 647
 648 /*
 649  * Do not make any assumptions regarding the relationship between the
 650  * IRQ number returned here and the Xen pirq argument.
 651  *
 652  * Note: We don't assign an event channel until the irq actually started
 653  * up.  Return an existing irq if we've already got one for the gsi.
 654  *
 655  * Shareable implies level triggered, not shareable implies edge
 656  * triggered here.
 657  */
 658 int xen_bind_pirq_gsi_to_irq(unsigned gsi,
 659                              unsigned pirq, int shareable, char *name)
 660 {
 661         int irq = -1;
 662         struct physdev_irq irq_op;
 663         int ret;
 664
 665         mutex_lock(&irq_mapping_update_lock);
 666
 667         irq = xen_irq_from_gsi(gsi);
 668         if (irq != -1) {
 669                 pr_info("%s: returning irq %d for gsi %u\n",
 670                         __func__, irq, gsi);
 671                 goto out;
 672         }
 673
 674         irq = xen_allocate_irq_gsi(gsi);
 675         if (irq < 0)
 676                 goto out;
 677
 678         irq_op.irq = irq;
 679         irq_op.vector = 0;
 680
 681         /* Only the privileged domain can do this. For non-priv, the pcifront
 682          * driver provides a PCI bus that does the call to do exactly
 683          * this in the priv domain. */
 684         if (xen_initial_domain() &&
 685             HYPERVISOR_physdev_op(PHYSDEVOP_alloc_irq_vector, &irq_op)) {
 686                 xen_free_irq(irq);
 687                 irq = -ENOSPC;
 688                 goto out;
 689         }
 690
 691         ret = xen_irq_info_pirq_setup(irq, 0, pirq, gsi, DOMID_SELF,
 692                                shareable ? PIRQ_SHAREABLE : 0);
 693         if (ret < 0) {
 694                 __unbind_from_irq(irq);
 695                 irq = ret;
 696                 goto out;
 697         }
 698
 699         pirq_query_unmask(irq);
 700         /* We try to use the handler with the appropriate semantic for the
 701          * type of interrupt: if the interrupt is an edge triggered
 702          * interrupt we use handle_edge_irq.
 703          *
 704          * On the other hand if the interrupt is level triggered we use
 705          * handle_fasteoi_irq like the native code does for this kind of
 706          * interrupts.
 707          *
 708          * Depending on the Xen version, pirq_needs_eoi might return true
 709          * not only for level triggered interrupts but for edge triggered
 710          * interrupts too. In any case Xen always honors the eoi mechanism,
 711          * not injecting any more pirqs of the same kind if the first one
 712          * hasn't received an eoi yet. Therefore using the fasteoi handler
 713          * is the right choice either way.
 714          */
 715         if (shareable)
 716                 irq_set_chip_and_handler_name(irq, &xen_pirq_chip,
 717                                 handle_fasteoi_irq, name);
 718         else
 719                 irq_set_chip_and_handler_name(irq, &xen_pirq_chip,
 720                                 handle_edge_irq, name);
 721
 722 out:
 723         mutex_unlock(&irq_mapping_update_lock);
 724
 725         return irq;
 726 }
 727
 728 #ifdef CONFIG_PCI_MSI
 729 int xen_allocate_pirq_msi(struct pci_dev *dev, struct msi_desc *msidesc)
 730 {
 731         int rc;
 732         struct physdev_get_free_pirq op_get_free_pirq;
 733
 734         op_get_free_pirq.type = MAP_PIRQ_TYPE_MSI;
 735         rc = HYPERVISOR_physdev_op(PHYSDEVOP_get_free_pirq, &op_get_free_pirq);
 736
 737         WARN_ONCE(rc == -ENOSYS,
 738                   "hypervisor does not support the PHYSDEVOP_get_free_pirq interface\n");
 739
 740         return rc ? -1 : op_get_free_pirq.pirq;
 741 }
 742
 743 int xen_bind_pirq_msi_to_irq(struct pci_dev *dev, struct msi_desc *msidesc,
 744                              int pirq, const char *name, domid_t domid)
 745 {
 746         int irq, ret;
 747
 748         mutex_lock(&irq_mapping_update_lock);
 749
 750         irq = xen_allocate_irq_dynamic();
 751         if (irq < 0)
 752                 goto out;
 753
 754         irq_set_chip_and_handler_name(irq, &xen_pirq_chip, handle_edge_irq,
 755                         name);
 756
 757         ret = xen_irq_info_pirq_setup(irq, 0, pirq, 0, domid, 0);
 758         if (ret < 0)
 759                 goto error_irq;
 760         ret = irq_set_msi_desc(irq, msidesc);
 761         if (ret < 0)
 762                 goto error_irq;
 763 out:
 764         mutex_unlock(&irq_mapping_update_lock);
 765         return irq;
 766 error_irq:
 767         __unbind_from_irq(irq);
 768         mutex_unlock(&irq_mapping_update_lock);
 769         return ret;
 770 }
 771 #endif
 772
 773 int xen_destroy_irq(int irq)
 774 {
 775         struct irq_desc *desc;
 776         struct physdev_unmap_pirq unmap_irq;
 777         struct irq_info *info = info_for_irq(irq);
 778         int rc = -ENOENT;
 779
 780         mutex_lock(&irq_mapping_update_lock);
 781
 782         desc = irq_to_desc(irq);
 783         if (!desc)
 784                 goto out;
 785
 786         if (xen_initial_domain()) {
 787                 unmap_irq.pirq = info->u.pirq.pirq;
 788                 unmap_irq.domid = info->u.pirq.domid;
 789                 rc = HYPERVISOR_physdev_op(PHYSDEVOP_unmap_pirq, &unmap_irq);
 790                 /* If another domain quits without making the pci_disable_msix
 791                  * call, the Xen hypervisor takes care of freeing the PIRQs
 792                  * (free_domain_pirqs).
 793                  */
 794                 if ((rc == -ESRCH && info->u.pirq.domid != DOMID_SELF))
 795                         pr_info("domain %d does not have %d anymore\n",
 796                                 info->u.pirq.domid, info->u.pirq.pirq);
 797                 else if (rc) {
 798                         pr_warn("unmap irq failed %d\n", rc);
 799                         goto out;
 800                 }
 801         }
 802
 803         xen_free_irq(irq);
 804
 805 out:
 806         mutex_unlock(&irq_mapping_update_lock);
 807         return rc;
 808 }
 809
 810 int xen_irq_from_pirq(unsigned pirq)
 811 {
 812         int irq;
 813
 814         struct irq_info *info;
 815
 816         mutex_lock(&irq_mapping_update_lock);
 817
 818         list_for_each_entry(info, &xen_irq_list_head, list) {
 819                 if (info->type != IRQT_PIRQ)
 820                         continue;
 821                 irq = info->irq;
 822                 if (info->u.pirq.pirq == pirq)
 823                         goto out;
 824         }
 825         irq = -1;
 826 out:
 827         mutex_unlock(&irq_mapping_update_lock);
 828
 829         return irq;
 830 }
 831
 832
 833 int xen_pirq_from_irq(unsigned irq)
 834 {
 835         return pirq_from_irq(irq);
 836 }
 837 EXPORT_SYMBOL_GPL(xen_pirq_from_irq);
 838
 839 int bind_evtchn_to_irq(unsigned int evtchn)
 840 {
 841         int irq;
 842         int ret;
 843
 844         if (evtchn >= xen_evtchn_max_channels())
 845                 return -ENOMEM;
 846
 847         mutex_lock(&irq_mapping_update_lock);
 848
 849         irq = get_evtchn_to_irq(evtchn);
 850
 851         if (irq == -1) {
 852                 irq = xen_allocate_irq_dynamic();
 853                 if (irq < 0)
 854                         goto out;
 855
 856                 irq_set_chip_and_handler_name(irq, &xen_dynamic_chip,
 857                                               handle_edge_irq, "event");
 858
 859                 ret = xen_irq_info_evtchn_setup(irq, evtchn);
 860                 if (ret < 0) {
 861                         __unbind_from_irq(irq);
 862                         irq = ret;
 863                         goto out;
 864                 }
 865         } else {
 866                 struct irq_info *info = info_for_irq(irq);
 867                 WARN_ON(info == NULL || info->type != IRQT_EVTCHN);
 868         }
 869
 870 out:
 871         mutex_unlock(&irq_mapping_update_lock);
 872
 873         return irq;
 874 }
 875 EXPORT_SYMBOL_GPL(bind_evtchn_to_irq);
 876
 877 static int bind_ipi_to_irq(unsigned int ipi, unsigned int cpu)
 878 {
 879         struct evtchn_bind_ipi bind_ipi;
 880         int evtchn, irq;
 881         int ret;
 882
 883         mutex_lock(&irq_mapping_update_lock);
 884
 885         irq = per_cpu(ipi_to_irq, cpu)[ipi];
 886
 887         if (irq == -1) {
 888                 irq = xen_allocate_irq_dynamic();
 889                 if (irq < 0)
 890                         goto out;
 891
 892                 irq_set_chip_and_handler_name(irq, &xen_percpu_chip,
 893                                               handle_percpu_irq, "ipi");
 894
 895                 bind_ipi.vcpu = cpu;
 896                 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_ipi,
 897                                                 &bind_ipi) != 0)
 898                         BUG();
 899                 evtchn = bind_ipi.port;
 900
 901                 ret = xen_irq_info_ipi_setup(cpu, irq, evtchn, ipi);
 902                 if (ret < 0) {
 903                         __unbind_from_irq(irq);
 904                         irq = ret;
 905                         goto out;
 906                 }
 907                 bind_evtchn_to_cpu(evtchn, cpu);
 908         } else {
 909                 struct irq_info *info = info_for_irq(irq);
 910                 WARN_ON(info == NULL || info->type != IRQT_IPI);
 911         }
 912
 913  out:
 914         mutex_unlock(&irq_mapping_update_lock);
 915         return irq;
 916 }
 917
 918 static int bind_interdomain_evtchn_to_irq(unsigned int remote_domain,
 919                                           unsigned int remote_port)
 920 {
 921         struct evtchn_bind_interdomain bind_interdomain;
 922         int err;
 923
 924         bind_interdomain.remote_dom  = remote_domain;
 925         bind_interdomain.remote_port = remote_port;
 926
 927         err = HYPERVISOR_event_channel_op(EVTCHNOP_bind_interdomain,
 928                                           &bind_interdomain);
 929
 930         return err ? : bind_evtchn_to_irq(bind_interdomain.local_port);
 931 }
 932
 933 static int find_virq(unsigned int virq, unsigned int cpu)
 934 {
 935         struct evtchn_status status;
 936         int port, rc = -ENOENT;
 937
 938         memset(&status, 0, sizeof(status));
 939         for (port = 0; port < xen_evtchn_max_channels(); port++) {
 940                 status.dom = DOMID_SELF;
 941                 status.port = port;
 942                 rc = HYPERVISOR_event_channel_op(EVTCHNOP_status, &status);
 943                 if (rc < 0)
 944                         continue;
 945                 if (status.status != EVTCHNSTAT_virq)
 946                         continue;
 947                 if (status.u.virq == virq && status.vcpu == cpu) {
 948                         rc = port;
 949                         break;
 950                 }
 951         }
 952         return rc;
 953 }
 954
 955 int bind_virq_to_irq(unsigned int virq, unsigned int cpu)
 956 {
 957         struct evtchn_bind_virq bind_virq;
 958         int evtchn, irq, ret;
 959
 960         mutex_lock(&irq_mapping_update_lock);
 961
 962         irq = per_cpu(virq_to_irq, cpu)[virq];
 963
 964         if (irq == -1) {
 965                 irq = xen_allocate_irq_dynamic();
 966                 if (irq < 0)
 967                         goto out;
 968
 969                 irq_set_chip_and_handler_name(irq, &xen_percpu_chip,
 970                                               handle_percpu_irq, "virq");
 971
 972                 bind_virq.virq = virq;
 973                 bind_virq.vcpu = cpu;
 974                 ret = HYPERVISOR_event_channel_op(EVTCHNOP_bind_virq,
 975                                                 &bind_virq);
 976                 if (ret == 0)
 977                         evtchn = bind_virq.port;
 978                 else {
 979                         if (ret == -EEXIST)
 980                                 ret = find_virq(virq, cpu);
 981                         BUG_ON(ret < 0);
 982                         evtchn = ret;
 983                 }
 984
 985                 ret = xen_irq_info_virq_setup(cpu, irq, evtchn, virq);
 986                 if (ret < 0) {
 987                         __unbind_from_irq(irq);
 988                         irq = ret;
 989                         goto out;
 990                 }
 991
 992                 bind_evtchn_to_cpu(evtchn, cpu);
 993         } else {
 994                 struct irq_info *info = info_for_irq(irq);
 995                 WARN_ON(info == NULL || info->type != IRQT_VIRQ);
 996         }
 997
 998 out:
 999         mutex_unlock(&irq_mapping_update_lock);
1000
1001         return irq;
1002 }
1003
1004 static void unbind_from_irq(unsigned int irq)
1005 {
1006         mutex_lock(&irq_mapping_update_lock);
1007         __unbind_from_irq(irq);
1008         mutex_unlock(&irq_mapping_update_lock);
1009 }
1010
1011 int bind_evtchn_to_irqhandler(unsigned int evtchn,
1012                               irq_handler_t handler,
1013                               unsigned long irqflags,
1014                               const char *devname, void *dev_id)
1015 {
1016         int irq, retval;
1017
1018         irq = bind_evtchn_to_irq(evtchn);
1019         if (irq < 0)
1020                 return irq;
1021         retval = request_irq(irq, handler, irqflags, devname, dev_id);
1022         if (retval != 0) {
1023                 unbind_from_irq(irq);
1024                 return retval;
1025         }
1026
1027         return irq;
1028 }
1029 EXPORT_SYMBOL_GPL(bind_evtchn_to_irqhandler);
1030
1031 int bind_interdomain_evtchn_to_irqhandler(unsigned int remote_domain,
1032                                           unsigned int remote_port,
1033                                           irq_handler_t handler,
1034                                           unsigned long irqflags,
1035                                           const char *devname,
1036                                           void *dev_id)
1037 {
1038         int irq, retval;
1039
1040         irq = bind_interdomain_evtchn_to_irq(remote_domain, remote_port);
1041         if (irq < 0)
1042                 return irq;
1043
1044         retval = request_irq(irq, handler, irqflags, devname, dev_id);
1045         if (retval != 0) {
1046                 unbind_from_irq(irq);
1047                 return retval;
1048         }
1049
1050         return irq;
1051 }
1052 EXPORT_SYMBOL_GPL(bind_interdomain_evtchn_to_irqhandler);
1053
1054 int bind_virq_to_irqhandler(unsigned int virq, unsigned int cpu,
1055                             irq_handler_t handler,
1056                             unsigned long irqflags, const char *devname, void *dev_id)
1057 {
1058         int irq, retval;
1059
1060         irq = bind_virq_to_irq(virq, cpu);
1061         if (irq < 0)
1062                 return irq;
1063         retval = request_irq(irq, handler, irqflags, devname, dev_id);
1064         if (retval != 0) {
1065                 unbind_from_irq(irq);
1066                 return retval;
1067         }
1068
1069         return irq;
1070 }
1071 EXPORT_SYMBOL_GPL(bind_virq_to_irqhandler);
1072
1073 int bind_ipi_to_irqhandler(enum ipi_vector ipi,
1074                            unsigned int cpu,
1075                            irq_handler_t handler,
1076                            unsigned long irqflags,
1077                            const char *devname,
1078                            void *dev_id)
1079 {
1080         int irq, retval;
1081
1082         irq = bind_ipi_to_irq(ipi, cpu);
1083         if (irq < 0)
1084                 return irq;
1085
1086         irqflags |= IRQF_NO_SUSPEND | IRQF_FORCE_RESUME | IRQF_EARLY_RESUME;
1087         retval = request_irq(irq, handler, irqflags, devname, dev_id);
1088         if (retval != 0) {
1089                 unbind_from_irq(irq);
1090                 return retval;
1091         }
1092
1093         return irq;
1094 }
1095
1096 void unbind_from_irqhandler(unsigned int irq, void *dev_id)
1097 {
1098         struct irq_info *info = irq_get_handler_data(irq);
1099
1100         if (WARN_ON(!info))
1101                 return;
1102         free_irq(irq, dev_id);
1103         unbind_from_irq(irq);
1104 }
1105 EXPORT_SYMBOL_GPL(unbind_from_irqhandler);
1106
1107 int evtchn_make_refcounted(unsigned int evtchn)
1108 {
1109         int irq = get_evtchn_to_irq(evtchn);
1110         struct irq_info *info;
1111
1112         if (irq == -1)
1113                 return -ENOENT;
1114
1115         info = irq_get_handler_data(irq);
1116
1117         if (!info)
1118                 return -ENOENT;
1119
1120         WARN_ON(info->refcnt != -1);
1121
1122         info->refcnt = 1;
1123
1124         return 0;
1125 }
1126 EXPORT_SYMBOL_GPL(evtchn_make_refcounted);
1127
1128 int evtchn_get(unsigned int evtchn)
1129 {
1130         int irq;
1131         struct irq_info *info;
1132         int err = -ENOENT;
1133
1134         if (evtchn >= xen_evtchn_max_channels())
1135                 return -EINVAL;
1136
1137         mutex_lock(&irq_mapping_update_lock);
1138
1139         irq = get_evtchn_to_irq(evtchn);
1140         if (irq == -1)
1141                 goto done;
1142
1143         info = irq_get_handler_data(irq);
1144
1145         if (!info)
1146                 goto done;
1147
1148         err = -EINVAL;
1149         if (info->refcnt <= 0)
1150                 goto done;
1151
1152         info->refcnt++;
1153         err = 0;
1154  done:
1155         mutex_unlock(&irq_mapping_update_lock);
1156
1157         return err;
1158 }
1159 EXPORT_SYMBOL_GPL(evtchn_get);
1160
1161 void evtchn_put(unsigned int evtchn)
1162 {
1163         int irq = get_evtchn_to_irq(evtchn);
1164         if (WARN_ON(irq == -1))
1165                 return;
1166         unbind_from_irq(irq);
1167 }
1168 EXPORT_SYMBOL_GPL(evtchn_put);
1169
1170 void xen_send_IPI_one(unsigned int cpu, enum ipi_vector vector)
1171 {
1172         int irq;
1173
1174 #ifdef CONFIG_X86
1175         if (unlikely(vector == XEN_NMI_VECTOR)) {
1176                 int rc =  HYPERVISOR_vcpu_op(VCPUOP_send_nmi, cpu, NULL);
1177                 if (rc < 0)
1178                         printk(KERN_WARNING "Sending nmi to CPU%d failed (rc:%d)\n", cpu, rc);
1179                 return;
1180         }
1181 #endif
1182         irq = per_cpu(ipi_to_irq, cpu)[vector];
1183         BUG_ON(irq < 0);
1184         notify_remote_via_irq(irq);
1185 }
1186
1187 static DEFINE_PER_CPU(unsigned, xed_nesting_count);
1188
1189 static void __xen_evtchn_do_upcall(void)
1190 {
1191         struct vcpu_info *vcpu_info = __this_cpu_read(xen_vcpu);
1192         int cpu = get_cpu();
1193         unsigned count;
1194
1195         do {
1196                 vcpu_info->evtchn_upcall_pending = 0;
1197
1198                 if (__this_cpu_inc_return(xed_nesting_count) - 1)
1199                         goto out;
1200
1201                 xen_evtchn_handle_events(cpu);
1202
1203                 BUG_ON(!irqs_disabled());
1204
1205                 count = __this_cpu_read(xed_nesting_count);
1206                 __this_cpu_write(xed_nesting_count, 0);
1207         } while (count != 1 || vcpu_info->evtchn_upcall_pending);
1208
1209 out:
1210
1211         put_cpu();
1212 }
1213
1214 void xen_evtchn_do_upcall(struct pt_regs *regs)
1215 {
1216         struct pt_regs *old_regs = set_irq_regs(regs);
1217
1218         irq_enter();
1219 #ifdef CONFIG_X86
1220         exit_idle();
1221 #endif
1222
1223         __xen_evtchn_do_upcall();
1224
1225         irq_exit();
1226         set_irq_regs(old_regs);
1227 }
1228
1229 void xen_hvm_evtchn_do_upcall(void)
1230 {
1231         __xen_evtchn_do_upcall();
1232 }
1233 EXPORT_SYMBOL_GPL(xen_hvm_evtchn_do_upcall);
1234
1235 /* Rebind a new event channel to an existing irq. */
1236 void rebind_evtchn_irq(int evtchn, int irq)
1237 {
1238         struct irq_info *info = info_for_irq(irq);
1239
1240         if (WARN_ON(!info))
1241                 return;
1242
1243         /* Make sure the irq is masked, since the new event channel
1244            will also be masked. */
1245         disable_irq(irq);
1246
1247         mutex_lock(&irq_mapping_update_lock);
1248
1249         /* After resume the irq<->evtchn mappings are all cleared out */
1250         BUG_ON(get_evtchn_to_irq(evtchn) != -1);
1251         /* Expect irq to have been bound before,
1252            so there should be a proper type */
1253         BUG_ON(info->type == IRQT_UNBOUND);
1254
1255         (void)xen_irq_info_evtchn_setup(irq, evtchn);
1256
1257         mutex_unlock(&irq_mapping_update_lock);
1258
1259         /* new event channels are always bound to cpu 0 */
1260         irq_set_affinity(irq, cpumask_of(0));
1261
1262         /* Unmask the event channel. */
1263         enable_irq(irq);
1264 }
1265
1266 /* Rebind an evtchn so that it gets delivered to a specific cpu */
1267 static int rebind_irq_to_cpu(unsigned irq, unsigned tcpu)
1268 {
1269         struct evtchn_bind_vcpu bind_vcpu;
1270         int evtchn = evtchn_from_irq(irq);
1271         int masked;
1272
1273         if (!VALID_EVTCHN(evtchn))
1274                 return -1;
1275
1276         /*
1277          * Events delivered via platform PCI interrupts are always
1278          * routed to vcpu 0 and hence cannot be rebound.
1279          */
1280         if (xen_hvm_domain() && !xen_have_vector_callback)
1281                 return -1;
1282
1283         /* Send future instances of this interrupt to other vcpu. */
1284         bind_vcpu.port = evtchn;
1285         bind_vcpu.vcpu = tcpu;
1286
1287         /*
1288          * Mask the event while changing the VCPU binding to prevent
1289          * it being delivered on an unexpected VCPU.
1290          */
1291         masked = test_and_set_mask(evtchn);
1292
1293         /*
1294          * If this fails, it usually just indicates that we're dealing with a
1295          * virq or IPI channel, which don't actually need to be rebound. Ignore
1296          * it, but don't do the xenlinux-level rebind in that case.
1297          */
1298         if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_vcpu, &bind_vcpu) >= 0)
1299                 bind_evtchn_to_cpu(evtchn, tcpu);
1300
1301         if (!masked)
1302                 unmask_evtchn(evtchn);
1303
1304         return 0;
1305 }
1306
1307 static int set_affinity_irq(struct irq_data *data, const struct cpumask *dest,
1308                             bool force)
1309 {
1310         unsigned tcpu = cpumask_first(dest);
1311
1312         return rebind_irq_to_cpu(data->irq, tcpu);
1313 }
1314
1315 static int retrigger_evtchn(int evtchn)
1316 {
1317         int masked;
1318
1319         if (!VALID_EVTCHN(evtchn))
1320                 return 0;
1321
1322         masked = test_and_set_mask(evtchn);
1323         set_evtchn(evtchn);
1324         if (!masked)
1325                 unmask_evtchn(evtchn);
1326
1327         return 1;
1328 }
1329
1330 int resend_irq_on_evtchn(unsigned int irq)
1331 {
1332         return retrigger_evtchn(evtchn_from_irq(irq));
1333 }
1334
1335 static void enable_dynirq(struct irq_data *data)
1336 {
1337         int evtchn = evtchn_from_irq(data->irq);
1338
1339         if (VALID_EVTCHN(evtchn))
1340                 unmask_evtchn(evtchn);
1341 }
1342
1343 static void disable_dynirq(struct irq_data *data)
1344 {
1345         int evtchn = evtchn_from_irq(data->irq);
1346
1347         if (VALID_EVTCHN(evtchn))
1348                 mask_evtchn(evtchn);
1349 }
1350
1351 static void ack_dynirq(struct irq_data *data)
1352 {
1353         int evtchn = evtchn_from_irq(data->irq);
1354
1355         irq_move_irq(data);
1356
1357         if (VALID_EVTCHN(evtchn))
1358                 clear_evtchn(evtchn);
1359 }
1360
1361 static void mask_ack_dynirq(struct irq_data *data)
1362 {
1363         disable_dynirq(data);
1364         ack_dynirq(data);
1365 }
1366
1367 static int retrigger_dynirq(struct irq_data *data)
1368 {
1369         return retrigger_evtchn(evtchn_from_irq(data->irq));
1370 }
1371
1372 static void restore_pirqs(void)
1373 {
1374         int pirq, rc, irq, gsi;
1375         struct physdev_map_pirq map_irq;
1376         struct irq_info *info;
1377
1378         list_for_each_entry(info, &xen_irq_list_head, list) {
1379                 if (info->type != IRQT_PIRQ)
1380                         continue;
1381
1382                 pirq = info->u.pirq.pirq;
1383                 gsi = info->u.pirq.gsi;
1384                 irq = info->irq;
1385
1386                 /* save/restore of PT devices doesn't work, so at this point the
1387                  * only devices present are GSI based emulated devices */
1388                 if (!gsi)
1389                         continue;
1390
1391                 map_irq.domid = DOMID_SELF;
1392                 map_irq.type = MAP_PIRQ_TYPE_GSI;
1393                 map_irq.index = gsi;
1394                 map_irq.pirq = pirq;
1395
1396                 rc = HYPERVISOR_physdev_op(PHYSDEVOP_map_pirq, &map_irq);
1397                 if (rc) {
1398                         pr_warn("xen map irq failed gsi=%d irq=%d pirq=%d rc=%d\n",
1399                                 gsi, irq, pirq, rc);
1400                         xen_free_irq(irq);
1401                         continue;
1402                 }
1403
1404                 printk(KERN_DEBUG "xen: --> irq=%d, pirq=%d\n", irq, map_irq.pirq);
1405
1406                 __startup_pirq(irq);
1407         }
1408 }
1409
1410 static void restore_cpu_virqs(unsigned int cpu)
1411 {
1412         struct evtchn_bind_virq bind_virq;
1413         int virq, irq, evtchn;
1414
1415         for (virq = 0; virq < NR_VIRQS; virq++) {
1416                 if ((irq = per_cpu(virq_to_irq, cpu)[virq]) == -1)
1417                         continue;
1418
1419                 BUG_ON(virq_from_irq(irq) != virq);
1420
1421                 /* Get a new binding from Xen. */
1422                 bind_virq.virq = virq;
1423                 bind_virq.vcpu = cpu;
1424                 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_virq,
1425                                                 &bind_virq) != 0)
1426                         BUG();
1427                 evtchn = bind_virq.port;
1428
1429                 /* Record the new mapping. */
1430                 (void)xen_irq_info_virq_setup(cpu, irq, evtchn, virq);
1431                 bind_evtchn_to_cpu(evtchn, cpu);
1432         }
1433 }
1434
1435 static void restore_cpu_ipis(unsigned int cpu)
1436 {
1437         struct evtchn_bind_ipi bind_ipi;
1438         int ipi, irq, evtchn;
1439
1440         for (ipi = 0; ipi < XEN_NR_IPIS; ipi++) {
1441                 if ((irq = per_cpu(ipi_to_irq, cpu)[ipi]) == -1)
1442                         continue;
1443
1444                 BUG_ON(ipi_from_irq(irq) != ipi);
1445
1446                 /* Get a new binding from Xen. */
1447                 bind_ipi.vcpu = cpu;
1448                 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_ipi,
1449                                                 &bind_ipi) != 0)
1450                         BUG();
1451                 evtchn = bind_ipi.port;
1452
1453                 /* Record the new mapping. */
1454                 (void)xen_irq_info_ipi_setup(cpu, irq, evtchn, ipi);
1455                 bind_evtchn_to_cpu(evtchn, cpu);
1456         }
1457 }
1458
1459 /* Clear an irq's pending state, in preparation for polling on it */
1460 void xen_clear_irq_pending(int irq)
1461 {
1462         int evtchn = evtchn_from_irq(irq);
1463
1464         if (VALID_EVTCHN(evtchn))
1465                 clear_evtchn(evtchn);
1466 }
1467 EXPORT_SYMBOL(xen_clear_irq_pending);
1468 void xen_set_irq_pending(int irq)
1469 {
1470         int evtchn = evtchn_from_irq(irq);
1471
1472         if (VALID_EVTCHN(evtchn))
1473                 set_evtchn(evtchn);
1474 }
1475
1476 bool xen_test_irq_pending(int irq)
1477 {
1478         int evtchn = evtchn_from_irq(irq);
1479         bool ret = false;
1480
1481         if (VALID_EVTCHN(evtchn))
1482                 ret = test_evtchn(evtchn);
1483
1484         return ret;
1485 }
1486
1487 /* Poll waiting for an irq to become pending with timeout.  In the usual case,
1488  * the irq will be disabled so it won't deliver an interrupt. */
1489 void xen_poll_irq_timeout(int irq, u64 timeout)
1490 {
1491         evtchn_port_t evtchn = evtchn_from_irq(irq);
1492
1493         if (VALID_EVTCHN(evtchn)) {
1494                 struct sched_poll poll;
1495
1496                 poll.nr_ports = 1;
1497                 poll.timeout = timeout;
1498                 set_xen_guest_handle(poll.ports, &evtchn);
1499
1500                 if (HYPERVISOR_sched_op(SCHEDOP_poll, &poll) != 0)
1501                         BUG();
1502         }
1503 }
1504 EXPORT_SYMBOL(xen_poll_irq_timeout);
1505 /* Poll waiting for an irq to become pending.  In the usual case, the
1506  * irq will be disabled so it won't deliver an interrupt. */
1507 void xen_poll_irq(int irq)
1508 {
1509         xen_poll_irq_timeout(irq, 0 /* no timeout */);
1510 }
1511
1512 /* Check whether the IRQ line is shared with other guests. */
1513 int xen_test_irq_shared(int irq)
1514 {
1515         struct irq_info *info = info_for_irq(irq);
1516         struct physdev_irq_status_query irq_status;
1517
1518         if (WARN_ON(!info))
1519                 return -ENOENT;
1520
1521         irq_status.irq = info->u.pirq.pirq;
1522
1523         if (HYPERVISOR_physdev_op(PHYSDEVOP_irq_status_query, &irq_status))
1524                 return 0;
1525         return !(irq_status.flags & XENIRQSTAT_shared);
1526 }
1527 EXPORT_SYMBOL_GPL(xen_test_irq_shared);
1528
1529 void xen_irq_resume(void)
1530 {
1531         unsigned int cpu;
1532         struct irq_info *info;
1533
1534         /* New event-channel space is not 'live' yet. */
1535         xen_evtchn_mask_all();
1536
1537         /* No IRQ <-> event-channel mappings. */
1538         list_for_each_entry(info, &xen_irq_list_head, list)
1539                 info->evtchn = 0; /* zap event-channel binding */
1540
1541         clear_evtchn_to_irq_all();
1542
1543         for_each_possible_cpu(cpu) {
1544                 restore_cpu_virqs(cpu);
1545                 restore_cpu_ipis(cpu);
1546         }
1547
1548         restore_pirqs();
1549 }
1550
1551 static struct irq_chip xen_dynamic_chip __read_mostly = {
1552         .name                   = "xen-dyn",
1553
1554         .irq_disable            = disable_dynirq,
1555         .irq_mask               = disable_dynirq,
1556         .irq_unmask             = enable_dynirq,
1557
1558         .irq_ack                = ack_dynirq,
1559         .irq_mask_ack           = mask_ack_dynirq,
1560
1561         .irq_set_affinity       = set_affinity_irq,
1562         .irq_retrigger          = retrigger_dynirq,
1563 };
1564
1565 static struct irq_chip xen_pirq_chip __read_mostly = {
1566         .name                   = "xen-pirq",
1567
1568         .irq_startup            = startup_pirq,
1569         .irq_shutdown           = shutdown_pirq,
1570         .irq_enable             = enable_pirq,
1571         .irq_disable            = disable_pirq,
1572
1573         .irq_mask               = disable_dynirq,
1574         .irq_unmask             = enable_dynirq,
1575
1576         .irq_ack                = eoi_pirq,
1577         .irq_eoi                = eoi_pirq,
1578         .irq_mask_ack           = mask_ack_pirq,
1579
1580         .irq_set_affinity       = set_affinity_irq,
1581
1582         .irq_retrigger          = retrigger_dynirq,
1583 };
1584
1585 static struct irq_chip xen_percpu_chip __read_mostly = {
1586         .name                   = "xen-percpu",
1587
1588         .irq_disable            = disable_dynirq,
1589         .irq_mask               = disable_dynirq,
1590         .irq_unmask             = enable_dynirq,
1591
1592         .irq_ack                = ack_dynirq,
1593 };
1594
1595 int xen_set_callback_via(uint64_t via)
1596 {
1597         struct xen_hvm_param a;
1598         a.domid = DOMID_SELF;
1599         a.index = HVM_PARAM_CALLBACK_IRQ;
1600         a.value = via;
1601         return HYPERVISOR_hvm_op(HVMOP_set_param, &a);
1602 }
1603 EXPORT_SYMBOL_GPL(xen_set_callback_via);
1604
1605 #ifdef CONFIG_XEN_PVHVM
1606 /* Vector callbacks are better than PCI interrupts to receive event
1607  * channel notifications because we can receive vector callbacks on any
1608  * vcpu and we don't need PCI support or APIC interactions. */
1609 void xen_callback_vector(void)
1610 {
1611         int rc;
1612         uint64_t callback_via;
1613         if (xen_have_vector_callback) {
1614                 callback_via = HVM_CALLBACK_VECTOR(HYPERVISOR_CALLBACK_VECTOR);
1615                 rc = xen_set_callback_via(callback_via);
1616                 if (rc) {
1617                         pr_err("Request for Xen HVM callback vector failed\n");
1618                         xen_have_vector_callback = 0;
1619                         return;
1620                 }
1621                 pr_info("Xen HVM callback vector for event delivery is enabled\n");
1622                 /* in the restore case the vector has already been allocated */
1623                 if (!test_bit(HYPERVISOR_CALLBACK_VECTOR, used_vectors))
1624                         alloc_intr_gate(HYPERVISOR_CALLBACK_VECTOR,
1625                                         xen_hvm_callback_vector);
1626         }
1627 }
1628 #else
1629 void xen_callback_vector(void) {}
1630 #endif
1631
1632 void __init xen_init_IRQ(void)
1633 {
1634         xen_evtchn_2l_init();
1635
1636         evtchn_to_irq = kcalloc(EVTCHN_ROW(xen_evtchn_max_channels()),
1637                                 sizeof(*evtchn_to_irq), GFP_KERNEL);
1638         BUG_ON(!evtchn_to_irq);
1639
1640         /* No event channels are 'live' right now. */
1641         xen_evtchn_mask_all();
1642
1643         pirq_needs_eoi = pirq_needs_eoi_flag;
1644
1645 #ifdef CONFIG_X86
1646         if (xen_hvm_domain()) {
1647                 xen_callback_vector();
1648                 native_init_IRQ();
1649                 /* pci_xen_hvm_init must be called after native_init_IRQ so that
1650                  * __acpi_register_gsi can point at the right function */
1651                 pci_xen_hvm_init();
1652         } else {
1653                 int rc;
1654                 struct physdev_pirq_eoi_gmfn eoi_gmfn;
1655
1656                 irq_ctx_init(smp_processor_id());
1657                 if (xen_initial_domain())
1658                         pci_xen_initial_domain();
1659
1660                 pirq_eoi_map = (void *)__get_free_page(GFP_KERNEL|__GFP_ZERO);
1661                 eoi_gmfn.gmfn = virt_to_mfn(pirq_eoi_map);
1662                 rc = HYPERVISOR_physdev_op(PHYSDEVOP_pirq_eoi_gmfn_v2, &eoi_gmfn);
1663                 if (rc != 0) {
1664                         free_page((unsigned long) pirq_eoi_map);
1665                         pirq_eoi_map = NULL;
1666                 } else
1667                         pirq_needs_eoi = pirq_check_eoi_map;
1668         }
1669 #endif
1670 }