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/moduleparam.h>
  30 #include <linux/string.h>
  31 #include <linux/bootmem.h>
  32 #include <linux/slab.h>
  33 #include <linux/irqnr.h>
  34 #include <linux/pci.h>
  35
  36 #ifdef CONFIG_X86
  37 #include <asm/desc.h>
  38 #include <asm/ptrace.h>
  39 #include <asm/irq.h>
  40 #include <asm/idle.h>
  41 #include <asm/io_apic.h>
  42 #include <asm/i8259.h>
  43 #include <asm/xen/pci.h>
  44 #endif
  45 #include <asm/sync_bitops.h>
  46 #include <asm/xen/hypercall.h>
  47 #include <asm/xen/hypervisor.h>
  48 #include <xen/page.h>
  49
  50 #include <xen/xen.h>
  51 #include <xen/hvm.h>
  52 #include <xen/xen-ops.h>
  53 #include <xen/events.h>
  54 #include <xen/interface/xen.h>
  55 #include <xen/interface/event_channel.h>
  56 #include <xen/interface/hvm/hvm_op.h>
  57 #include <xen/interface/hvm/params.h>
  58 #include <xen/interface/physdev.h>
  59 #include <xen/interface/sched.h>
  60 #include <xen/interface/vcpu.h>
  61 #include <asm/hw_irq.h>
  62
  63 #include "events_internal.h"
  64
  65 const struct evtchn_ops *evtchn_ops;
  66
  67 /*
  68  * This lock protects updates to the following mapping and reference-count
  69  * arrays. The lock does not need to be acquired to read the mapping tables.
  70  */
  71 static DEFINE_MUTEX(irq_mapping_update_lock);
  72
  73 static LIST_HEAD(xen_irq_list_head);
  74
  75 /* IRQ <-> VIRQ mapping. */
  76 static DEFINE_PER_CPU(int [NR_VIRQS], virq_to_irq) = {[0 ... NR_VIRQS-1] = -1};
  77
  78 /* IRQ <-> IPI mapping */
  79 static DEFINE_PER_CPU(int [XEN_NR_IPIS], ipi_to_irq) = {[0 ... XEN_NR_IPIS-1] = -1};
  80
  81 int **evtchn_to_irq;
  82 #ifdef CONFIG_X86
  83 static unsigned long *pirq_eoi_map;
  84 #endif
  85 static bool (*pirq_needs_eoi)(unsigned irq);
  86
  87 #define EVTCHN_ROW(e)  (e / (PAGE_SIZE/sizeof(**evtchn_to_irq)))
  88 #define EVTCHN_COL(e)  (e % (PAGE_SIZE/sizeof(**evtchn_to_irq)))
  89 #define EVTCHN_PER_ROW (PAGE_SIZE / sizeof(**evtchn_to_irq))
  90
  91 /* Xen will never allocate port zero for any purpose. */
  92 #define VALID_EVTCHN(chn)       ((chn) != 0)
  93
  94 static struct irq_chip xen_dynamic_chip;
  95 static struct irq_chip xen_percpu_chip;
  96 static struct irq_chip xen_pirq_chip;
  97 static void enable_dynirq(struct irq_data *data);
  98 static void disable_dynirq(struct irq_data *data);
  99
 100 static void clear_evtchn_to_irq_row(unsigned row)
 101 {
 102         unsigned col;
 103
 104         for (col = 0; col < EVTCHN_PER_ROW; col++)
 105                 evtchn_to_irq[row][col] = -1;
 106 }
 107
 108 static void clear_evtchn_to_irq_all(void)
 109 {
 110         unsigned row;
 111
 112         for (row = 0; row < EVTCHN_ROW(xen_evtchn_max_channels()); row++) {
 113                 if (evtchn_to_irq[row] == NULL)
 114                         continue;
 115                 clear_evtchn_to_irq_row(row);
 116         }
 117 }
 118
 119 static int set_evtchn_to_irq(unsigned evtchn, unsigned irq)
 120 {
 121         unsigned row;
 122         unsigned col;
 123
 124         if (evtchn >= xen_evtchn_max_channels())
 125                 return -EINVAL;
 126
 127         row = EVTCHN_ROW(evtchn);
 128         col = EVTCHN_COL(evtchn);
 129
 130         if (evtchn_to_irq[row] == NULL) {
 131                 /* Unallocated irq entries return -1 anyway */
 132                 if (irq == -1)
 133                         return 0;
 134
 135                 evtchn_to_irq[row] = (int *)get_zeroed_page(GFP_KERNEL);
 136                 if (evtchn_to_irq[row] == NULL)
 137                         return -ENOMEM;
 138
 139                 clear_evtchn_to_irq_row(row);
 140         }
 141
 142         evtchn_to_irq[EVTCHN_ROW(evtchn)][EVTCHN_COL(evtchn)] = irq;
 143         return 0;
 144 }
 145
 146 int get_evtchn_to_irq(unsigned evtchn)
 147 {
 148         if (evtchn >= xen_evtchn_max_channels())
 149                 return -1;
 150         if (evtchn_to_irq[EVTCHN_ROW(evtchn)] == NULL)
 151                 return -1;
 152         return evtchn_to_irq[EVTCHN_ROW(evtchn)][EVTCHN_COL(evtchn)];
 153 }
 154
 155 /* Get info for IRQ */
 156 struct irq_info *info_for_irq(unsigned irq)
 157 {
 158         return irq_get_handler_data(irq);
 159 }
 160
 161 /* Constructors for packed IRQ information. */
 162 static int xen_irq_info_common_setup(struct irq_info *info,
 163                                      unsigned irq,
 164                                      enum xen_irq_type type,
 165                                      unsigned evtchn,
 166                                      unsigned short cpu)
 167 {
 168         int ret;
 169
 170         BUG_ON(info->type != IRQT_UNBOUND && info->type != type);
 171
 172         info->type = type;
 173         info->irq = irq;
 174         info->evtchn = evtchn;
 175         info->cpu = cpu;
 176
 177         ret = set_evtchn_to_irq(evtchn, irq);
 178         if (ret < 0)
 179                 return ret;
 180
 181         irq_clear_status_flags(irq, IRQ_NOREQUEST|IRQ_NOAUTOEN);
 182
 183         return xen_evtchn_port_setup(info);
 184 }
 185
 186 static int xen_irq_info_evtchn_setup(unsigned irq,
 187                                      unsigned evtchn)
 188 {
 189         struct irq_info *info = info_for_irq(irq);
 190
 191         return xen_irq_info_common_setup(info, irq, IRQT_EVTCHN, evtchn, 0);
 192 }
 193
 194 static int xen_irq_info_ipi_setup(unsigned cpu,
 195                                   unsigned irq,
 196                                   unsigned evtchn,
 197                                   enum ipi_vector ipi)
 198 {
 199         struct irq_info *info = info_for_irq(irq);
 200
 201         info->u.ipi = ipi;
 202
 203         per_cpu(ipi_to_irq, cpu)[ipi] = irq;
 204
 205         return xen_irq_info_common_setup(info, irq, IRQT_IPI, evtchn, 0);
 206 }
 207
 208 static int xen_irq_info_virq_setup(unsigned cpu,
 209                                    unsigned irq,
 210                                    unsigned evtchn,
 211                                    unsigned virq)
 212 {
 213         struct irq_info *info = info_for_irq(irq);
 214
 215         info->u.virq = virq;
 216
 217         per_cpu(virq_to_irq, cpu)[virq] = irq;
 218
 219         return xen_irq_info_common_setup(info, irq, IRQT_VIRQ, evtchn, 0);
 220 }
 221
 222 static int xen_irq_info_pirq_setup(unsigned irq,
 223                                    unsigned evtchn,
 224                                    unsigned pirq,
 225                                    unsigned gsi,
 226                                    uint16_t domid,
 227                                    unsigned char flags)
 228 {
 229         struct irq_info *info = info_for_irq(irq);
 230
 231         info->u.pirq.pirq = pirq;
 232         info->u.pirq.gsi = gsi;
 233         info->u.pirq.domid = domid;
 234         info->u.pirq.flags = flags;
 235
 236         return xen_irq_info_common_setup(info, irq, IRQT_PIRQ, evtchn, 0);
 237 }
 238
 239 static void xen_irq_info_cleanup(struct irq_info *info)
 240 {
 241         set_evtchn_to_irq(info->evtchn, -1);
 242         info->evtchn = 0;
 243 }
 244
 245 /*
 246  * Accessors for packed IRQ information.
 247  */
 248 unsigned int evtchn_from_irq(unsigned irq)
 249 {
 250         if (unlikely(WARN(irq >= nr_irqs, "Invalid irq %d!\n", irq)))
 251                 return 0;
 252
 253         return info_for_irq(irq)->evtchn;
 254 }
 255
 256 unsigned irq_from_evtchn(unsigned int evtchn)
 257 {
 258         return get_evtchn_to_irq(evtchn);
 259 }
 260 EXPORT_SYMBOL_GPL(irq_from_evtchn);
 261
 262 int irq_from_virq(unsigned int cpu, unsigned int virq)
 263 {
 264         return per_cpu(virq_to_irq, cpu)[virq];
 265 }
 266
 267 static enum ipi_vector ipi_from_irq(unsigned irq)
 268 {
 269         struct irq_info *info = info_for_irq(irq);
 270
 271         BUG_ON(info == NULL);
 272         BUG_ON(info->type != IRQT_IPI);
 273
 274         return info->u.ipi;
 275 }
 276
 277 static unsigned virq_from_irq(unsigned irq)
 278 {
 279         struct irq_info *info = info_for_irq(irq);
 280
 281         BUG_ON(info == NULL);
 282         BUG_ON(info->type != IRQT_VIRQ);
 283
 284         return info->u.virq;
 285 }
 286
 287 static unsigned pirq_from_irq(unsigned irq)
 288 {
 289         struct irq_info *info = info_for_irq(irq);
 290
 291         BUG_ON(info == NULL);
 292         BUG_ON(info->type != IRQT_PIRQ);
 293
 294         return info->u.pirq.pirq;
 295 }
 296
 297 static enum xen_irq_type type_from_irq(unsigned irq)
 298 {
 299         return info_for_irq(irq)->type;
 300 }
 301
 302 unsigned cpu_from_irq(unsigned irq)
 303 {
 304         return info_for_irq(irq)->cpu;
 305 }
 306
 307 unsigned int cpu_from_evtchn(unsigned int evtchn)
 308 {
 309         int irq = get_evtchn_to_irq(evtchn);
 310         unsigned ret = 0;
 311
 312         if (irq != -1)
 313                 ret = cpu_from_irq(irq);
 314
 315         return ret;
 316 }
 317
 318 #ifdef CONFIG_X86
 319 static bool pirq_check_eoi_map(unsigned irq)
 320 {
 321         return test_bit(pirq_from_irq(irq), pirq_eoi_map);
 322 }
 323 #endif
 324
 325 static bool pirq_needs_eoi_flag(unsigned irq)
 326 {
 327         struct irq_info *info = info_for_irq(irq);
 328         BUG_ON(info->type != IRQT_PIRQ);
 329
 330         return info->u.pirq.flags & PIRQ_NEEDS_EOI;
 331 }
 332
 333 static void bind_evtchn_to_cpu(unsigned int chn, unsigned int cpu)
 334 {
 335         int irq = get_evtchn_to_irq(chn);
 336         struct irq_info *info = info_for_irq(irq);
 337
 338         BUG_ON(irq == -1);
 339 #ifdef CONFIG_SMP
 340         cpumask_copy(irq_get_affinity_mask(irq), cpumask_of(cpu));
 341 #endif
 342         xen_evtchn_port_bind_to_cpu(info, cpu);
 343
 344         info->cpu = cpu;
 345 }
 346
 347 static void xen_evtchn_mask_all(void)
 348 {
 349         unsigned int evtchn;
 350
 351         for (evtchn = 0; evtchn < xen_evtchn_nr_channels(); evtchn++)
 352                 mask_evtchn(evtchn);
 353 }
 354
 355 /**
 356  * notify_remote_via_irq - send event to remote end of event channel via irq
 357  * @irq: irq of event channel to send event to
 358  *
 359  * Unlike notify_remote_via_evtchn(), this is safe to use across
 360  * save/restore. Notifications on a broken connection are silently
 361  * dropped.
 362  */
 363 void notify_remote_via_irq(int irq)
 364 {
 365         int evtchn = evtchn_from_irq(irq);
 366
 367         if (VALID_EVTCHN(evtchn))
 368                 notify_remote_via_evtchn(evtchn);
 369 }
 370 EXPORT_SYMBOL_GPL(notify_remote_via_irq);
 371
 372 static void xen_irq_init(unsigned irq)
 373 {
 374         struct irq_info *info;
 375 #ifdef CONFIG_SMP
 376         /* By default all event channels notify CPU#0. */
 377         cpumask_copy(irq_get_affinity_mask(irq), cpumask_of(0));
 378 #endif
 379
 380         info = kzalloc(sizeof(*info), GFP_KERNEL);
 381         if (info == NULL)
 382                 panic("Unable to allocate metadata for IRQ%d\n", irq);
 383
 384         info->type = IRQT_UNBOUND;
 385         info->refcnt = -1;
 386
 387         irq_set_handler_data(irq, info);
 388
 389         list_add_tail(&info->list, &xen_irq_list_head);
 390 }
 391
 392 static int __must_check xen_allocate_irqs_dynamic(int nvec)
 393 {
 394         int i, irq = irq_alloc_descs(-1, 0, nvec, -1);
 395
 396         if (irq >= 0) {
 397                 for (i = 0; i < nvec; i++)
 398                         xen_irq_init(irq + i);
 399         }
 400
 401         return irq;
 402 }
 403
 404 static inline int __must_check xen_allocate_irq_dynamic(void)
 405 {
 406
 407         return xen_allocate_irqs_dynamic(1);
 408 }
 409
 410 static int __must_check xen_allocate_irq_gsi(unsigned gsi)
 411 {
 412         int irq;
 413
 414         /*
 415          * A PV guest has no concept of a GSI (since it has no ACPI
 416          * nor access to/knowledge of the physical APICs). Therefore
 417          * all IRQs are dynamically allocated from the entire IRQ
 418          * space.
 419          */
 420         if (xen_pv_domain() && !xen_initial_domain())
 421                 return xen_allocate_irq_dynamic();
 422
 423         /* Legacy IRQ descriptors are already allocated by the arch. */
 424         if (gsi < nr_legacy_irqs())
 425                 irq = gsi;
 426         else
 427                 irq = irq_alloc_desc_at(gsi, -1);
 428
 429         xen_irq_init(irq);
 430
 431         return irq;
 432 }
 433
 434 static void xen_free_irq(unsigned irq)
 435 {
 436         struct irq_info *info = irq_get_handler_data(irq);
 437
 438         if (WARN_ON(!info))
 439                 return;
 440
 441         list_del(&info->list);
 442
 443         irq_set_handler_data(irq, NULL);
 444
 445         WARN_ON(info->refcnt > 0);
 446
 447         kfree(info);
 448
 449         /* Legacy IRQ descriptors are managed by the arch. */
 450         if (irq < nr_legacy_irqs())
 451                 return;
 452
 453         irq_free_desc(irq);
 454 }
 455
 456 static void xen_evtchn_close(unsigned int port)
 457 {
 458         struct evtchn_close close;
 459
 460         close.port = port;
 461         if (HYPERVISOR_event_channel_op(EVTCHNOP_close, &close) != 0)
 462                 BUG();
 463 }
 464
 465 static void pirq_query_unmask(int irq)
 466 {
 467         struct physdev_irq_status_query irq_status;
 468         struct irq_info *info = info_for_irq(irq);
 469
 470         BUG_ON(info->type != IRQT_PIRQ);
 471
 472         irq_status.irq = pirq_from_irq(irq);
 473         if (HYPERVISOR_physdev_op(PHYSDEVOP_irq_status_query, &irq_status))
 474                 irq_status.flags = 0;
 475
 476         info->u.pirq.flags &= ~PIRQ_NEEDS_EOI;
 477         if (irq_status.flags & XENIRQSTAT_needs_eoi)
 478                 info->u.pirq.flags |= PIRQ_NEEDS_EOI;
 479 }
 480
 481 static void eoi_pirq(struct irq_data *data)
 482 {
 483         int evtchn = evtchn_from_irq(data->irq);
 484         struct physdev_eoi eoi = { .irq = pirq_from_irq(data->irq) };
 485         int rc = 0;
 486
 487         if (!VALID_EVTCHN(evtchn))
 488                 return;
 489
 490         if (unlikely(irqd_is_setaffinity_pending(data)) &&
 491             likely(!irqd_irq_disabled(data))) {
 492                 int masked = test_and_set_mask(evtchn);
 493
 494                 clear_evtchn(evtchn);
 495
 496                 irq_move_masked_irq(data);
 497
 498                 if (!masked)
 499                         unmask_evtchn(evtchn);
 500         } else
 501                 clear_evtchn(evtchn);
 502
 503         if (pirq_needs_eoi(data->irq)) {
 504                 rc = HYPERVISOR_physdev_op(PHYSDEVOP_eoi, &eoi);
 505                 WARN_ON(rc);
 506         }
 507 }
 508
 509 static void mask_ack_pirq(struct irq_data *data)
 510 {
 511         disable_dynirq(data);
 512         eoi_pirq(data);
 513 }
 514
 515 static unsigned int __startup_pirq(unsigned int irq)
 516 {
 517         struct evtchn_bind_pirq bind_pirq;
 518         struct irq_info *info = info_for_irq(irq);
 519         int evtchn = evtchn_from_irq(irq);
 520         int rc;
 521
 522         BUG_ON(info->type != IRQT_PIRQ);
 523
 524         if (VALID_EVTCHN(evtchn))
 525                 goto out;
 526
 527         bind_pirq.pirq = pirq_from_irq(irq);
 528         /* NB. We are happy to share unless we are probing. */
 529         bind_pirq.flags = info->u.pirq.flags & PIRQ_SHAREABLE ?
 530                                         BIND_PIRQ__WILL_SHARE : 0;
 531         rc = HYPERVISOR_event_channel_op(EVTCHNOP_bind_pirq, &bind_pirq);
 532         if (rc != 0) {
 533                 pr_warn("Failed to obtain physical IRQ %d\n", irq);
 534                 return 0;
 535         }
 536         evtchn = bind_pirq.port;
 537
 538         pirq_query_unmask(irq);
 539
 540         rc = set_evtchn_to_irq(evtchn, irq);
 541         if (rc)
 542                 goto err;
 543
 544         info->evtchn = evtchn;
 545         bind_evtchn_to_cpu(evtchn, 0);
 546
 547         rc = xen_evtchn_port_setup(info);
 548         if (rc)
 549                 goto err;
 550
 551 out:
 552         unmask_evtchn(evtchn);
 553         eoi_pirq(irq_get_irq_data(irq));
 554
 555         return 0;
 556
 557 err:
 558         pr_err("irq%d: Failed to set port to irq mapping (%d)\n", irq, rc);
 559         xen_evtchn_close(evtchn);
 560         return 0;
 561 }
 562
 563 static unsigned int startup_pirq(struct irq_data *data)
 564 {
 565         return __startup_pirq(data->irq);
 566 }
 567
 568 static void shutdown_pirq(struct irq_data *data)
 569 {
 570         unsigned int irq = data->irq;
 571         struct irq_info *info = info_for_irq(irq);
 572         unsigned evtchn = evtchn_from_irq(irq);
 573
 574         BUG_ON(info->type != IRQT_PIRQ);
 575
 576         if (!VALID_EVTCHN(evtchn))
 577                 return;
 578
 579         mask_evtchn(evtchn);
 580         xen_evtchn_close(evtchn);
 581         xen_irq_info_cleanup(info);
 582 }
 583
 584 static void enable_pirq(struct irq_data *data)
 585 {
 586         startup_pirq(data);
 587 }
 588
 589 static void disable_pirq(struct irq_data *data)
 590 {
 591         disable_dynirq(data);
 592 }
 593
 594 int xen_irq_from_gsi(unsigned gsi)
 595 {
 596         struct irq_info *info;
 597
 598         list_for_each_entry(info, &xen_irq_list_head, list) {
 599                 if (info->type != IRQT_PIRQ)
 600                         continue;
 601
 602                 if (info->u.pirq.gsi == gsi)
 603                         return info->irq;
 604         }
 605
 606         return -1;
 607 }
 608 EXPORT_SYMBOL_GPL(xen_irq_from_gsi);
 609
 610 static void __unbind_from_irq(unsigned int irq)
 611 {
 612         int evtchn = evtchn_from_irq(irq);
 613         struct irq_info *info = irq_get_handler_data(irq);
 614
 615         if (info->refcnt > 0) {
 616                 info->refcnt--;
 617                 if (info->refcnt != 0)
 618                         return;
 619         }
 620
 621         if (VALID_EVTCHN(evtchn)) {
 622                 unsigned int cpu = cpu_from_irq(irq);
 623
 624                 xen_evtchn_close(evtchn);
 625
 626                 switch (type_from_irq(irq)) {
 627                 case IRQT_VIRQ:
 628                         per_cpu(virq_to_irq, cpu)[virq_from_irq(irq)] = -1;
 629                         break;
 630                 case IRQT_IPI:
 631                         per_cpu(ipi_to_irq, cpu)[ipi_from_irq(irq)] = -1;
 632                         break;
 633                 default:
 634                         break;
 635                 }
 636
 637                 xen_irq_info_cleanup(info);
 638         }
 639
 640         BUG_ON(info_for_irq(irq)->type == IRQT_UNBOUND);
 641
 642         xen_free_irq(irq);
 643 }
 644
 645 /*
 646  * Do not make any assumptions regarding the relationship between the
 647  * IRQ number returned here and the Xen pirq argument.
 648  *
 649  * Note: We don't assign an event channel until the irq actually started
 650  * up.  Return an existing irq if we've already got one for the gsi.
 651  *
 652  * Shareable implies level triggered, not shareable implies edge
 653  * triggered here.
 654  */
 655 int xen_bind_pirq_gsi_to_irq(unsigned gsi,
 656                              unsigned pirq, int shareable, char *name)
 657 {
 658         int irq = -1;
 659         struct physdev_irq irq_op;
 660         int ret;
 661
 662         mutex_lock(&irq_mapping_update_lock);
 663
 664         irq = xen_irq_from_gsi(gsi);
 665         if (irq != -1) {
 666                 pr_info("%s: returning irq %d for gsi %u\n",
 667                         __func__, irq, gsi);
 668                 goto out;
 669         }
 670
 671         irq = xen_allocate_irq_gsi(gsi);
 672         if (irq < 0)
 673                 goto out;
 674
 675         irq_op.irq = irq;
 676         irq_op.vector = 0;
 677
 678         /* Only the privileged domain can do this. For non-priv, the pcifront
 679          * driver provides a PCI bus that does the call to do exactly
 680          * this in the priv domain. */
 681         if (xen_initial_domain() &&
 682             HYPERVISOR_physdev_op(PHYSDEVOP_alloc_irq_vector, &irq_op)) {
 683                 xen_free_irq(irq);
 684                 irq = -ENOSPC;
 685                 goto out;
 686         }
 687
 688         ret = xen_irq_info_pirq_setup(irq, 0, pirq, gsi, DOMID_SELF,
 689                                shareable ? PIRQ_SHAREABLE : 0);
 690         if (ret < 0) {
 691                 __unbind_from_irq(irq);
 692                 irq = ret;
 693                 goto out;
 694         }
 695
 696         pirq_query_unmask(irq);
 697         /* We try to use the handler with the appropriate semantic for the
 698          * type of interrupt: if the interrupt is an edge triggered
 699          * interrupt we use handle_edge_irq.
 700          *
 701          * On the other hand if the interrupt is level triggered we use
 702          * handle_fasteoi_irq like the native code does for this kind of
 703          * interrupts.
 704          *
 705          * Depending on the Xen version, pirq_needs_eoi might return true
 706          * not only for level triggered interrupts but for edge triggered
 707          * interrupts too. In any case Xen always honors the eoi mechanism,
 708          * not injecting any more pirqs of the same kind if the first one
 709          * hasn't received an eoi yet. Therefore using the fasteoi handler
 710          * is the right choice either way.
 711          */
 712         if (shareable)
 713                 irq_set_chip_and_handler_name(irq, &xen_pirq_chip,
 714                                 handle_fasteoi_irq, name);
 715         else
 716                 irq_set_chip_and_handler_name(irq, &xen_pirq_chip,
 717                                 handle_edge_irq, name);
 718
 719 out:
 720         mutex_unlock(&irq_mapping_update_lock);
 721
 722         return irq;
 723 }
 724
 725 #ifdef CONFIG_PCI_MSI
 726 int xen_allocate_pirq_msi(struct pci_dev *dev, struct msi_desc *msidesc)
 727 {
 728         int rc;
 729         struct physdev_get_free_pirq op_get_free_pirq;
 730
 731         op_get_free_pirq.type = MAP_PIRQ_TYPE_MSI;
 732         rc = HYPERVISOR_physdev_op(PHYSDEVOP_get_free_pirq, &op_get_free_pirq);
 733
 734         WARN_ONCE(rc == -ENOSYS,
 735                   "hypervisor does not support the PHYSDEVOP_get_free_pirq interface\n");
 736
 737         return rc ? -1 : op_get_free_pirq.pirq;
 738 }
 739
 740 int xen_bind_pirq_msi_to_irq(struct pci_dev *dev, struct msi_desc *msidesc,
 741                              int pirq, int nvec, const char *name, domid_t domid)
 742 {
 743         int i, irq, ret;
 744
 745         mutex_lock(&irq_mapping_update_lock);
 746
 747         irq = xen_allocate_irqs_dynamic(nvec);
 748         if (irq < 0)
 749                 goto out;
 750
 751         for (i = 0; i < nvec; i++) {
 752                 irq_set_chip_and_handler_name(irq + i, &xen_pirq_chip, handle_edge_irq, name);
 753
 754                 ret = xen_irq_info_pirq_setup(irq + i, 0, pirq + i, 0, domid,
 755                                               i == 0 ? 0 : PIRQ_MSI_GROUP);
 756                 if (ret < 0)
 757                         goto error_irq;
 758         }
 759
 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         for (; i >= 0; i--)
 768                 __unbind_from_irq(irq + i);
 769         mutex_unlock(&irq_mapping_update_lock);
 770         return ret;
 771 }
 772 #endif
 773
 774 int xen_destroy_irq(int irq)
 775 {
 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         /*
 783          * If trying to remove a vector in a MSI group different
 784          * than the first one skip the PIRQ unmap unless this vector
 785          * is the first one in the group.
 786          */
 787         if (xen_initial_domain() && !(info->u.pirq.flags & PIRQ_MSI_GROUP)) {
 788                 unmap_irq.pirq = info->u.pirq.pirq;
 789                 unmap_irq.domid = info->u.pirq.domid;
 790                 rc = HYPERVISOR_physdev_op(PHYSDEVOP_unmap_pirq, &unmap_irq);
 791                 /* If another domain quits without making the pci_disable_msix
 792                  * call, the Xen hypervisor takes care of freeing the PIRQs
 793                  * (free_domain_pirqs).
 794                  */
 795                 if ((rc == -ESRCH && info->u.pirq.domid != DOMID_SELF))
 796                         pr_info("domain %d does not have %d anymore\n",
 797                                 info->u.pirq.domid, info->u.pirq.pirq);
 798                 else if (rc) {
 799                         pr_warn("unmap irq failed %d\n", rc);
 800                         goto out;
 801                 }
 802         }
 803
 804         xen_free_irq(irq);
 805
 806 out:
 807         mutex_unlock(&irq_mapping_update_lock);
 808         return rc;
 809 }
 810
 811 int xen_irq_from_pirq(unsigned pirq)
 812 {
 813         int irq;
 814
 815         struct irq_info *info;
 816
 817         mutex_lock(&irq_mapping_update_lock);
 818
 819         list_for_each_entry(info, &xen_irq_list_head, list) {
 820                 if (info->type != IRQT_PIRQ)
 821                         continue;
 822                 irq = info->irq;
 823                 if (info->u.pirq.pirq == pirq)
 824                         goto out;
 825         }
 826         irq = -1;
 827 out:
 828         mutex_unlock(&irq_mapping_update_lock);
 829
 830         return irq;
 831 }
 832
 833
 834 int xen_pirq_from_irq(unsigned irq)
 835 {
 836         return pirq_from_irq(irq);
 837 }
 838 EXPORT_SYMBOL_GPL(xen_pirq_from_irq);
 839
 840 int bind_evtchn_to_irq(unsigned int evtchn)
 841 {
 842         int irq;
 843         int ret;
 844
 845         if (evtchn >= xen_evtchn_max_channels())
 846                 return -ENOMEM;
 847
 848         mutex_lock(&irq_mapping_update_lock);
 849
 850         irq = get_evtchn_to_irq(evtchn);
 851
 852         if (irq == -1) {
 853                 irq = xen_allocate_irq_dynamic();
 854                 if (irq < 0)
 855                         goto out;
 856
 857                 irq_set_chip_and_handler_name(irq, &xen_dynamic_chip,
 858                                               handle_edge_irq, "event");
 859
 860                 ret = xen_irq_info_evtchn_setup(irq, evtchn);
 861                 if (ret < 0) {
 862                         __unbind_from_irq(irq);
 863                         irq = ret;
 864                         goto out;
 865                 }
 866                 /* New interdomain events are bound to VCPU 0. */
 867                 bind_evtchn_to_cpu(evtchn, 0);
 868         } else {
 869                 struct irq_info *info = info_for_irq(irq);
 870                 WARN_ON(info == NULL || info->type != IRQT_EVTCHN);
 871         }
 872
 873 out:
 874         mutex_unlock(&irq_mapping_update_lock);
 875
 876         return irq;
 877 }
 878 EXPORT_SYMBOL_GPL(bind_evtchn_to_irq);
 879
 880 static int bind_ipi_to_irq(unsigned int ipi, unsigned int cpu)
 881 {
 882         struct evtchn_bind_ipi bind_ipi;
 883         int evtchn, irq;
 884         int ret;
 885
 886         mutex_lock(&irq_mapping_update_lock);
 887
 888         irq = per_cpu(ipi_to_irq, cpu)[ipi];
 889
 890         if (irq == -1) {
 891                 irq = xen_allocate_irq_dynamic();
 892                 if (irq < 0)
 893                         goto out;
 894
 895                 irq_set_chip_and_handler_name(irq, &xen_percpu_chip,
 896                                               handle_percpu_irq, "ipi");
 897
 898                 bind_ipi.vcpu = xen_vcpu_nr(cpu);
 899                 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_ipi,
 900                                                 &bind_ipi) != 0)
 901                         BUG();
 902                 evtchn = bind_ipi.port;
 903
 904                 ret = xen_irq_info_ipi_setup(cpu, irq, evtchn, ipi);
 905                 if (ret < 0) {
 906                         __unbind_from_irq(irq);
 907                         irq = ret;
 908                         goto out;
 909                 }
 910                 bind_evtchn_to_cpu(evtchn, cpu);
 911         } else {
 912                 struct irq_info *info = info_for_irq(irq);
 913                 WARN_ON(info == NULL || info->type != IRQT_IPI);
 914         }
 915
 916  out:
 917         mutex_unlock(&irq_mapping_update_lock);
 918         return irq;
 919 }
 920
 921 int bind_interdomain_evtchn_to_irq(unsigned int remote_domain,
 922                                    unsigned int remote_port)
 923 {
 924         struct evtchn_bind_interdomain bind_interdomain;
 925         int err;
 926
 927         bind_interdomain.remote_dom  = remote_domain;
 928         bind_interdomain.remote_port = remote_port;
 929
 930         err = HYPERVISOR_event_channel_op(EVTCHNOP_bind_interdomain,
 931                                           &bind_interdomain);
 932
 933         return err ? : bind_evtchn_to_irq(bind_interdomain.local_port);
 934 }
 935 EXPORT_SYMBOL_GPL(bind_interdomain_evtchn_to_irq);
 936
 937 static int find_virq(unsigned int virq, unsigned int cpu)
 938 {
 939         struct evtchn_status status;
 940         int port, rc = -ENOENT;
 941
 942         memset(&status, 0, sizeof(status));
 943         for (port = 0; port < xen_evtchn_max_channels(); port++) {
 944                 status.dom = DOMID_SELF;
 945                 status.port = port;
 946                 rc = HYPERVISOR_event_channel_op(EVTCHNOP_status, &status);
 947                 if (rc < 0)
 948                         continue;
 949                 if (status.status != EVTCHNSTAT_virq)
 950                         continue;
 951                 if (status.u.virq == virq && status.vcpu == cpu) {
 952                         rc = port;
 953                         break;
 954                 }
 955         }
 956         return rc;
 957 }
 958
 959 /**
 960  * xen_evtchn_nr_channels - number of usable event channel ports
 961  *
 962  * This may be less than the maximum supported by the current
 963  * hypervisor ABI. Use xen_evtchn_max_channels() for the maximum
 964  * supported.
 965  */
 966 unsigned xen_evtchn_nr_channels(void)
 967 {
 968         return evtchn_ops->nr_channels();
 969 }
 970 EXPORT_SYMBOL_GPL(xen_evtchn_nr_channels);
 971
 972 int bind_virq_to_irq(unsigned int virq, unsigned int cpu, bool percpu)
 973 {
 974         struct evtchn_bind_virq bind_virq;
 975         int evtchn, irq, ret;
 976
 977         mutex_lock(&irq_mapping_update_lock);
 978
 979         irq = per_cpu(virq_to_irq, cpu)[virq];
 980
 981         if (irq == -1) {
 982                 irq = xen_allocate_irq_dynamic();
 983                 if (irq < 0)
 984                         goto out;
 985
 986                 if (percpu)
 987                         irq_set_chip_and_handler_name(irq, &xen_percpu_chip,
 988                                                       handle_percpu_irq, "virq");
 989                 else
 990                         irq_set_chip_and_handler_name(irq, &xen_dynamic_chip,
 991                                                       handle_edge_irq, "virq");
 992
 993                 bind_virq.virq = virq;
 994                 bind_virq.vcpu = xen_vcpu_nr(cpu);
 995                 ret = HYPERVISOR_event_channel_op(EVTCHNOP_bind_virq,
 996                                                 &bind_virq);
 997                 if (ret == 0)
 998                         evtchn = bind_virq.port;
 999                 else {
1000                         if (ret == -EEXIST)
1001                                 ret = find_virq(virq, cpu);
1002                         BUG_ON(ret < 0);
1003                         evtchn = ret;
1004                 }
1005
1006                 ret = xen_irq_info_virq_setup(cpu, irq, evtchn, virq);
1007                 if (ret < 0) {
1008                         __unbind_from_irq(irq);
1009                         irq = ret;
1010                         goto out;
1011                 }
1012
1013                 bind_evtchn_to_cpu(evtchn, cpu);
1014         } else {
1015                 struct irq_info *info = info_for_irq(irq);
1016                 WARN_ON(info == NULL || info->type != IRQT_VIRQ);
1017         }
1018
1019 out:
1020         mutex_unlock(&irq_mapping_update_lock);
1021
1022         return irq;
1023 }
1024
1025 static void unbind_from_irq(unsigned int irq)
1026 {
1027         mutex_lock(&irq_mapping_update_lock);
1028         __unbind_from_irq(irq);
1029         mutex_unlock(&irq_mapping_update_lock);
1030 }
1031
1032 int bind_evtchn_to_irqhandler(unsigned int evtchn,
1033                               irq_handler_t handler,
1034                               unsigned long irqflags,
1035                               const char *devname, void *dev_id)
1036 {
1037         int irq, retval;
1038
1039         irq = bind_evtchn_to_irq(evtchn);
1040         if (irq < 0)
1041                 return irq;
1042         retval = request_irq(irq, handler, irqflags, devname, dev_id);
1043         if (retval != 0) {
1044                 unbind_from_irq(irq);
1045                 return retval;
1046         }
1047
1048         return irq;
1049 }
1050 EXPORT_SYMBOL_GPL(bind_evtchn_to_irqhandler);
1051
1052 int bind_interdomain_evtchn_to_irqhandler(unsigned int remote_domain,
1053                                           unsigned int remote_port,
1054                                           irq_handler_t handler,
1055                                           unsigned long irqflags,
1056                                           const char *devname,
1057                                           void *dev_id)
1058 {
1059         int irq, retval;
1060
1061         irq = bind_interdomain_evtchn_to_irq(remote_domain, remote_port);
1062         if (irq < 0)
1063                 return irq;
1064
1065         retval = request_irq(irq, handler, irqflags, devname, dev_id);
1066         if (retval != 0) {
1067                 unbind_from_irq(irq);
1068                 return retval;
1069         }
1070
1071         return irq;
1072 }
1073 EXPORT_SYMBOL_GPL(bind_interdomain_evtchn_to_irqhandler);
1074
1075 int bind_virq_to_irqhandler(unsigned int virq, unsigned int cpu,
1076                             irq_handler_t handler,
1077                             unsigned long irqflags, const char *devname, void *dev_id)
1078 {
1079         int irq, retval;
1080
1081         irq = bind_virq_to_irq(virq, cpu, irqflags & IRQF_PERCPU);
1082         if (irq < 0)
1083                 return irq;
1084         retval = request_irq(irq, handler, irqflags, devname, dev_id);
1085         if (retval != 0) {
1086                 unbind_from_irq(irq);
1087                 return retval;
1088         }
1089
1090         return irq;
1091 }
1092 EXPORT_SYMBOL_GPL(bind_virq_to_irqhandler);
1093
1094 int bind_ipi_to_irqhandler(enum ipi_vector ipi,
1095                            unsigned int cpu,
1096                            irq_handler_t handler,
1097                            unsigned long irqflags,
1098                            const char *devname,
1099                            void *dev_id)
1100 {
1101         int irq, retval;
1102
1103         irq = bind_ipi_to_irq(ipi, cpu);
1104         if (irq < 0)
1105                 return irq;
1106
1107         irqflags |= IRQF_NO_SUSPEND | IRQF_FORCE_RESUME | IRQF_EARLY_RESUME;
1108         retval = request_irq(irq, handler, irqflags, devname, dev_id);
1109         if (retval != 0) {
1110                 unbind_from_irq(irq);
1111                 return retval;
1112         }
1113
1114         return irq;
1115 }
1116
1117 void unbind_from_irqhandler(unsigned int irq, void *dev_id)
1118 {
1119         struct irq_info *info = irq_get_handler_data(irq);
1120
1121         if (WARN_ON(!info))
1122                 return;
1123         free_irq(irq, dev_id);
1124         unbind_from_irq(irq);
1125 }
1126 EXPORT_SYMBOL_GPL(unbind_from_irqhandler);
1127
1128 /**
1129  * xen_set_irq_priority() - set an event channel priority.
1130  * @irq:irq bound to an event channel.
1131  * @priority: priority between XEN_IRQ_PRIORITY_MAX and XEN_IRQ_PRIORITY_MIN.
1132  */
1133 int xen_set_irq_priority(unsigned irq, unsigned priority)
1134 {
1135         struct evtchn_set_priority set_priority;
1136
1137         set_priority.port = evtchn_from_irq(irq);
1138         set_priority.priority = priority;
1139
1140         return HYPERVISOR_event_channel_op(EVTCHNOP_set_priority,
1141                                            &set_priority);
1142 }
1143 EXPORT_SYMBOL_GPL(xen_set_irq_priority);
1144
1145 int evtchn_make_refcounted(unsigned int evtchn)
1146 {
1147         int irq = get_evtchn_to_irq(evtchn);
1148         struct irq_info *info;
1149
1150         if (irq == -1)
1151                 return -ENOENT;
1152
1153         info = irq_get_handler_data(irq);
1154
1155         if (!info)
1156                 return -ENOENT;
1157
1158         WARN_ON(info->refcnt != -1);
1159
1160         info->refcnt = 1;
1161
1162         return 0;
1163 }
1164 EXPORT_SYMBOL_GPL(evtchn_make_refcounted);
1165
1166 int evtchn_get(unsigned int evtchn)
1167 {
1168         int irq;
1169         struct irq_info *info;
1170         int err = -ENOENT;
1171
1172         if (evtchn >= xen_evtchn_max_channels())
1173                 return -EINVAL;
1174
1175         mutex_lock(&irq_mapping_update_lock);
1176
1177         irq = get_evtchn_to_irq(evtchn);
1178         if (irq == -1)
1179                 goto done;
1180
1181         info = irq_get_handler_data(irq);
1182
1183         if (!info)
1184                 goto done;
1185
1186         err = -EINVAL;
1187         if (info->refcnt <= 0)
1188                 goto done;
1189
1190         info->refcnt++;
1191         err = 0;
1192  done:
1193         mutex_unlock(&irq_mapping_update_lock);
1194
1195         return err;
1196 }
1197 EXPORT_SYMBOL_GPL(evtchn_get);
1198
1199 void evtchn_put(unsigned int evtchn)
1200 {
1201         int irq = get_evtchn_to_irq(evtchn);
1202         if (WARN_ON(irq == -1))
1203                 return;
1204         unbind_from_irq(irq);
1205 }
1206 EXPORT_SYMBOL_GPL(evtchn_put);
1207
1208 void xen_send_IPI_one(unsigned int cpu, enum ipi_vector vector)
1209 {
1210         int irq;
1211
1212 #ifdef CONFIG_X86
1213         if (unlikely(vector == XEN_NMI_VECTOR)) {
1214                 int rc =  HYPERVISOR_vcpu_op(VCPUOP_send_nmi, xen_vcpu_nr(cpu),
1215                                              NULL);
1216                 if (rc < 0)
1217                         printk(KERN_WARNING "Sending nmi to CPU%d failed (rc:%d)\n", cpu, rc);
1218                 return;
1219         }
1220 #endif
1221         irq = per_cpu(ipi_to_irq, cpu)[vector];
1222         BUG_ON(irq < 0);
1223         notify_remote_via_irq(irq);
1224 }
1225
1226 static DEFINE_PER_CPU(unsigned, xed_nesting_count);
1227
1228 static void __xen_evtchn_do_upcall(void)
1229 {
1230         struct vcpu_info *vcpu_info = __this_cpu_read(xen_vcpu);
1231         int cpu = get_cpu();
1232         unsigned count;
1233
1234         do {
1235                 vcpu_info->evtchn_upcall_pending = 0;
1236
1237                 if (__this_cpu_inc_return(xed_nesting_count) - 1)
1238                         goto out;
1239
1240                 xen_evtchn_handle_events(cpu);
1241
1242                 BUG_ON(!irqs_disabled());
1243
1244                 count = __this_cpu_read(xed_nesting_count);
1245                 __this_cpu_write(xed_nesting_count, 0);
1246         } while (count != 1 || vcpu_info->evtchn_upcall_pending);
1247
1248 out:
1249
1250         put_cpu();
1251 }
1252
1253 void xen_evtchn_do_upcall(struct pt_regs *regs)
1254 {
1255         struct pt_regs *old_regs = set_irq_regs(regs);
1256
1257         irq_enter();
1258 #ifdef CONFIG_X86
1259         exit_idle();
1260         inc_irq_stat(irq_hv_callback_count);
1261 #endif
1262
1263         __xen_evtchn_do_upcall();
1264
1265         irq_exit();
1266         set_irq_regs(old_regs);
1267 }
1268
1269 void xen_hvm_evtchn_do_upcall(void)
1270 {
1271         __xen_evtchn_do_upcall();
1272 }
1273 EXPORT_SYMBOL_GPL(xen_hvm_evtchn_do_upcall);
1274
1275 /* Rebind a new event channel to an existing irq. */
1276 void rebind_evtchn_irq(int evtchn, int irq)
1277 {
1278         struct irq_info *info = info_for_irq(irq);
1279
1280         if (WARN_ON(!info))
1281                 return;
1282
1283         /* Make sure the irq is masked, since the new event channel
1284            will also be masked. */
1285         disable_irq(irq);
1286
1287         mutex_lock(&irq_mapping_update_lock);
1288
1289         /* After resume the irq<->evtchn mappings are all cleared out */
1290         BUG_ON(get_evtchn_to_irq(evtchn) != -1);
1291         /* Expect irq to have been bound before,
1292            so there should be a proper type */
1293         BUG_ON(info->type == IRQT_UNBOUND);
1294
1295         (void)xen_irq_info_evtchn_setup(irq, evtchn);
1296
1297         mutex_unlock(&irq_mapping_update_lock);
1298
1299         bind_evtchn_to_cpu(evtchn, info->cpu);
1300         /* This will be deferred until interrupt is processed */
1301         irq_set_affinity(irq, cpumask_of(info->cpu));
1302
1303         /* Unmask the event channel. */
1304         enable_irq(irq);
1305 }
1306
1307 /* Rebind an evtchn so that it gets delivered to a specific cpu */
1308 static int rebind_irq_to_cpu(unsigned irq, unsigned tcpu)
1309 {
1310         struct evtchn_bind_vcpu bind_vcpu;
1311         int evtchn = evtchn_from_irq(irq);
1312         int masked;
1313
1314         if (!VALID_EVTCHN(evtchn))
1315                 return -1;
1316
1317         /* Send future instances of this interrupt to other vcpu. */
1318         bind_vcpu.port = evtchn;
1319         bind_vcpu.vcpu = xen_vcpu_nr(tcpu);
1320
1321         /*
1322          * Mask the event while changing the VCPU binding to prevent
1323          * it being delivered on an unexpected VCPU.
1324          */
1325         masked = test_and_set_mask(evtchn);
1326
1327         /*
1328          * If this fails, it usually just indicates that we're dealing with a
1329          * virq or IPI channel, which don't actually need to be rebound. Ignore
1330          * it, but don't do the xenlinux-level rebind in that case.
1331          */
1332         if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_vcpu, &bind_vcpu) >= 0)
1333                 bind_evtchn_to_cpu(evtchn, tcpu);
1334
1335         if (!masked)
1336                 unmask_evtchn(evtchn);
1337
1338         return 0;
1339 }
1340
1341 static int set_affinity_irq(struct irq_data *data, const struct cpumask *dest,
1342                             bool force)
1343 {
1344         unsigned tcpu = cpumask_first_and(dest, cpu_online_mask);
1345
1346         return rebind_irq_to_cpu(data->irq, tcpu);
1347 }
1348
1349 static void enable_dynirq(struct irq_data *data)
1350 {
1351         int evtchn = evtchn_from_irq(data->irq);
1352
1353         if (VALID_EVTCHN(evtchn))
1354                 unmask_evtchn(evtchn);
1355 }
1356
1357 static void disable_dynirq(struct irq_data *data)
1358 {
1359         int evtchn = evtchn_from_irq(data->irq);
1360
1361         if (VALID_EVTCHN(evtchn))
1362                 mask_evtchn(evtchn);
1363 }
1364
1365 static void ack_dynirq(struct irq_data *data)
1366 {
1367         int evtchn = evtchn_from_irq(data->irq);
1368
1369         if (!VALID_EVTCHN(evtchn))
1370                 return;
1371
1372         if (unlikely(irqd_is_setaffinity_pending(data)) &&
1373             likely(!irqd_irq_disabled(data))) {
1374                 int masked = test_and_set_mask(evtchn);
1375
1376                 clear_evtchn(evtchn);
1377
1378                 irq_move_masked_irq(data);
1379
1380                 if (!masked)
1381                         unmask_evtchn(evtchn);
1382         } else
1383                 clear_evtchn(evtchn);
1384 }
1385
1386 static void mask_ack_dynirq(struct irq_data *data)
1387 {
1388         disable_dynirq(data);
1389         ack_dynirq(data);
1390 }
1391
1392 static int retrigger_dynirq(struct irq_data *data)
1393 {
1394         unsigned int evtchn = evtchn_from_irq(data->irq);
1395         int masked;
1396
1397         if (!VALID_EVTCHN(evtchn))
1398                 return 0;
1399
1400         masked = test_and_set_mask(evtchn);
1401         set_evtchn(evtchn);
1402         if (!masked)
1403                 unmask_evtchn(evtchn);
1404
1405         return 1;
1406 }
1407
1408 static void restore_pirqs(void)
1409 {
1410         int pirq, rc, irq, gsi;
1411         struct physdev_map_pirq map_irq;
1412         struct irq_info *info;
1413
1414         list_for_each_entry(info, &xen_irq_list_head, list) {
1415                 if (info->type != IRQT_PIRQ)
1416                         continue;
1417
1418                 pirq = info->u.pirq.pirq;
1419                 gsi = info->u.pirq.gsi;
1420                 irq = info->irq;
1421
1422                 /* save/restore of PT devices doesn't work, so at this point the
1423                  * only devices present are GSI based emulated devices */
1424                 if (!gsi)
1425                         continue;
1426
1427                 map_irq.domid = DOMID_SELF;
1428                 map_irq.type = MAP_PIRQ_TYPE_GSI;
1429                 map_irq.index = gsi;
1430                 map_irq.pirq = pirq;
1431
1432                 rc = HYPERVISOR_physdev_op(PHYSDEVOP_map_pirq, &map_irq);
1433                 if (rc) {
1434                         pr_warn("xen map irq failed gsi=%d irq=%d pirq=%d rc=%d\n",
1435                                 gsi, irq, pirq, rc);
1436                         xen_free_irq(irq);
1437                         continue;
1438                 }
1439
1440                 printk(KERN_DEBUG "xen: --> irq=%d, pirq=%d\n", irq, map_irq.pirq);
1441
1442                 __startup_pirq(irq);
1443         }
1444 }
1445
1446 static void restore_cpu_virqs(unsigned int cpu)
1447 {
1448         struct evtchn_bind_virq bind_virq;
1449         int virq, irq, evtchn;
1450
1451         for (virq = 0; virq < NR_VIRQS; virq++) {
1452                 if ((irq = per_cpu(virq_to_irq, cpu)[virq]) == -1)
1453                         continue;
1454
1455                 BUG_ON(virq_from_irq(irq) != virq);
1456
1457                 /* Get a new binding from Xen. */
1458                 bind_virq.virq = virq;
1459                 bind_virq.vcpu = xen_vcpu_nr(cpu);
1460                 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_virq,
1461                                                 &bind_virq) != 0)
1462                         BUG();
1463                 evtchn = bind_virq.port;
1464
1465                 /* Record the new mapping. */
1466                 (void)xen_irq_info_virq_setup(cpu, irq, evtchn, virq);
1467                 bind_evtchn_to_cpu(evtchn, cpu);
1468         }
1469 }
1470
1471 static void restore_cpu_ipis(unsigned int cpu)
1472 {
1473         struct evtchn_bind_ipi bind_ipi;
1474         int ipi, irq, evtchn;
1475
1476         for (ipi = 0; ipi < XEN_NR_IPIS; ipi++) {
1477                 if ((irq = per_cpu(ipi_to_irq, cpu)[ipi]) == -1)
1478                         continue;
1479
1480                 BUG_ON(ipi_from_irq(irq) != ipi);
1481
1482                 /* Get a new binding from Xen. */
1483                 bind_ipi.vcpu = xen_vcpu_nr(cpu);
1484                 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_ipi,
1485                                                 &bind_ipi) != 0)
1486                         BUG();
1487                 evtchn = bind_ipi.port;
1488
1489                 /* Record the new mapping. */
1490                 (void)xen_irq_info_ipi_setup(cpu, irq, evtchn, ipi);
1491                 bind_evtchn_to_cpu(evtchn, cpu);
1492         }
1493 }
1494
1495 /* Clear an irq's pending state, in preparation for polling on it */
1496 void xen_clear_irq_pending(int irq)
1497 {
1498         int evtchn = evtchn_from_irq(irq);
1499
1500         if (VALID_EVTCHN(evtchn))
1501                 clear_evtchn(evtchn);
1502 }
1503 EXPORT_SYMBOL(xen_clear_irq_pending);
1504 void xen_set_irq_pending(int irq)
1505 {
1506         int evtchn = evtchn_from_irq(irq);
1507
1508         if (VALID_EVTCHN(evtchn))
1509                 set_evtchn(evtchn);
1510 }
1511
1512 bool xen_test_irq_pending(int irq)
1513 {
1514         int evtchn = evtchn_from_irq(irq);
1515         bool ret = false;
1516
1517         if (VALID_EVTCHN(evtchn))
1518                 ret = test_evtchn(evtchn);
1519
1520         return ret;
1521 }
1522
1523 /* Poll waiting for an irq to become pending with timeout.  In the usual case,
1524  * the irq will be disabled so it won't deliver an interrupt. */
1525 void xen_poll_irq_timeout(int irq, u64 timeout)
1526 {
1527         evtchn_port_t evtchn = evtchn_from_irq(irq);
1528
1529         if (VALID_EVTCHN(evtchn)) {
1530                 struct sched_poll poll;
1531
1532                 poll.nr_ports = 1;
1533                 poll.timeout = timeout;
1534                 set_xen_guest_handle(poll.ports, &evtchn);
1535
1536                 if (HYPERVISOR_sched_op(SCHEDOP_poll, &poll) != 0)
1537                         BUG();
1538         }
1539 }
1540 EXPORT_SYMBOL(xen_poll_irq_timeout);
1541 /* Poll waiting for an irq to become pending.  In the usual case, the
1542  * irq will be disabled so it won't deliver an interrupt. */
1543 void xen_poll_irq(int irq)
1544 {
1545         xen_poll_irq_timeout(irq, 0 /* no timeout */);
1546 }
1547
1548 /* Check whether the IRQ line is shared with other guests. */
1549 int xen_test_irq_shared(int irq)
1550 {
1551         struct irq_info *info = info_for_irq(irq);
1552         struct physdev_irq_status_query irq_status;
1553
1554         if (WARN_ON(!info))
1555                 return -ENOENT;
1556
1557         irq_status.irq = info->u.pirq.pirq;
1558
1559         if (HYPERVISOR_physdev_op(PHYSDEVOP_irq_status_query, &irq_status))
1560                 return 0;
1561         return !(irq_status.flags & XENIRQSTAT_shared);
1562 }
1563 EXPORT_SYMBOL_GPL(xen_test_irq_shared);
1564
1565 void xen_irq_resume(void)
1566 {
1567         unsigned int cpu;
1568         struct irq_info *info;
1569
1570         /* New event-channel space is not 'live' yet. */
1571         xen_evtchn_mask_all();
1572         xen_evtchn_resume();
1573
1574         /* No IRQ <-> event-channel mappings. */
1575         list_for_each_entry(info, &xen_irq_list_head, list)
1576                 info->evtchn = 0; /* zap event-channel binding */
1577
1578         clear_evtchn_to_irq_all();
1579
1580         for_each_possible_cpu(cpu) {
1581                 restore_cpu_virqs(cpu);
1582                 restore_cpu_ipis(cpu);
1583         }
1584
1585         restore_pirqs();
1586 }
1587
1588 static struct irq_chip xen_dynamic_chip __read_mostly = {
1589         .name                   = "xen-dyn",
1590
1591         .irq_disable            = disable_dynirq,
1592         .irq_mask               = disable_dynirq,
1593         .irq_unmask             = enable_dynirq,
1594
1595         .irq_ack                = ack_dynirq,
1596         .irq_mask_ack           = mask_ack_dynirq,
1597
1598         .irq_set_affinity       = set_affinity_irq,
1599         .irq_retrigger          = retrigger_dynirq,
1600 };
1601
1602 static struct irq_chip xen_pirq_chip __read_mostly = {
1603         .name                   = "xen-pirq",
1604
1605         .irq_startup            = startup_pirq,
1606         .irq_shutdown           = shutdown_pirq,
1607         .irq_enable             = enable_pirq,
1608         .irq_disable            = disable_pirq,
1609
1610         .irq_mask               = disable_dynirq,
1611         .irq_unmask             = enable_dynirq,
1612
1613         .irq_ack                = eoi_pirq,
1614         .irq_eoi                = eoi_pirq,
1615         .irq_mask_ack           = mask_ack_pirq,
1616
1617         .irq_set_affinity       = set_affinity_irq,
1618
1619         .irq_retrigger          = retrigger_dynirq,
1620 };
1621
1622 static struct irq_chip xen_percpu_chip __read_mostly = {
1623         .name                   = "xen-percpu",
1624
1625         .irq_disable            = disable_dynirq,
1626         .irq_mask               = disable_dynirq,
1627         .irq_unmask             = enable_dynirq,
1628
1629         .irq_ack                = ack_dynirq,
1630 };
1631
1632 int xen_set_callback_via(uint64_t via)
1633 {
1634         struct xen_hvm_param a;
1635         a.domid = DOMID_SELF;
1636         a.index = HVM_PARAM_CALLBACK_IRQ;
1637         a.value = via;
1638         return HYPERVISOR_hvm_op(HVMOP_set_param, &a);
1639 }
1640 EXPORT_SYMBOL_GPL(xen_set_callback_via);
1641
1642 #ifdef CONFIG_XEN_PVHVM
1643 /* Vector callbacks are better than PCI interrupts to receive event
1644  * channel notifications because we can receive vector callbacks on any
1645  * vcpu and we don't need PCI support or APIC interactions. */
1646 void xen_callback_vector(void)
1647 {
1648         int rc;
1649         uint64_t callback_via;
1650
1651         callback_via = HVM_CALLBACK_VECTOR(HYPERVISOR_CALLBACK_VECTOR);
1652         rc = xen_set_callback_via(callback_via);
1653         BUG_ON(rc);
1654         pr_info("Xen HVM callback vector for event delivery is enabled\n");
1655         /* in the restore case the vector has already been allocated */
1656         if (!test_bit(HYPERVISOR_CALLBACK_VECTOR, used_vectors))
1657                 alloc_intr_gate(HYPERVISOR_CALLBACK_VECTOR,
1658                                 xen_hvm_callback_vector);
1659 }
1660 #else
1661 void xen_callback_vector(void) {}
1662 #endif
1663
1664 #undef MODULE_PARAM_PREFIX
1665 #define MODULE_PARAM_PREFIX "xen."
1666
1667 static bool fifo_events = true;
1668 module_param(fifo_events, bool, 0);
1669
1670 void __init xen_init_IRQ(void)
1671 {
1672         int ret = -EINVAL;
1673
1674         if (fifo_events)
1675                 ret = xen_evtchn_fifo_init();
1676         if (ret < 0)
1677                 xen_evtchn_2l_init();
1678
1679         evtchn_to_irq = kcalloc(EVTCHN_ROW(xen_evtchn_max_channels()),
1680                                 sizeof(*evtchn_to_irq), GFP_KERNEL);
1681         BUG_ON(!evtchn_to_irq);
1682
1683         /* No event channels are 'live' right now. */
1684         xen_evtchn_mask_all();
1685
1686         pirq_needs_eoi = pirq_needs_eoi_flag;
1687
1688 #ifdef CONFIG_X86
1689         if (xen_pv_domain()) {
1690                 irq_ctx_init(smp_processor_id());
1691                 if (xen_initial_domain())
1692                         pci_xen_initial_domain();
1693         }
1694         if (xen_feature(XENFEAT_hvm_callback_vector))
1695                 xen_callback_vector();
1696
1697         if (xen_hvm_domain()) {
1698                 native_init_IRQ();
1699                 /* pci_xen_hvm_init must be called after native_init_IRQ so that
1700                  * __acpi_register_gsi can point at the right function */
1701                 pci_xen_hvm_init();
1702         } else {
1703                 int rc;
1704                 struct physdev_pirq_eoi_gmfn eoi_gmfn;
1705
1706                 pirq_eoi_map = (void *)__get_free_page(GFP_KERNEL|__GFP_ZERO);
1707                 eoi_gmfn.gmfn = virt_to_gfn(pirq_eoi_map);
1708                 rc = HYPERVISOR_physdev_op(PHYSDEVOP_pirq_eoi_gmfn_v2, &eoi_gmfn);
1709                 /* TODO: No PVH support for PIRQ EOI */
1710                 if (rc != 0) {
1711                         free_page((unsigned long) pirq_eoi_map);
1712                         pirq_eoi_map = NULL;
1713                 } else
1714                         pirq_needs_eoi = pirq_check_eoi_map;
1715         }
1716 #endif
1717 }