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/io_apic.h>
  41 #include <asm/i8259.h>
  42 #include <asm/xen/pci.h>
  43 #endif
  44 #include <asm/sync_bitops.h>
  45 #include <asm/xen/hypercall.h>
  46 #include <asm/xen/hypervisor.h>
  47 #include <xen/page.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[row][col] = irq;
 142         return 0;
 143 }
 144
 145 int get_evtchn_to_irq(unsigned evtchn)
 146 {
 147         if (evtchn >= xen_evtchn_max_channels())
 148                 return -1;
 149         if (evtchn_to_irq[EVTCHN_ROW(evtchn)] == NULL)
 150                 return -1;
 151         return evtchn_to_irq[EVTCHN_ROW(evtchn)][EVTCHN_COL(evtchn)];
 152 }
 153
 154 /* Get info for IRQ */
 155 struct irq_info *info_for_irq(unsigned irq)
 156 {
 157         return irq_get_handler_data(irq);
 158 }
 159
 160 /* Constructors for packed IRQ information. */
 161 static int xen_irq_info_common_setup(struct irq_info *info,
 162                                      unsigned irq,
 163                                      enum xen_irq_type type,
 164                                      unsigned evtchn,
 165                                      unsigned short cpu)
 166 {
 167         int ret;
 168
 169         BUG_ON(info->type != IRQT_UNBOUND && info->type != type);
 170
 171         info->type = type;
 172         info->irq = irq;
 173         info->evtchn = evtchn;
 174         info->cpu = cpu;
 175
 176         ret = set_evtchn_to_irq(evtchn, irq);
 177         if (ret < 0)
 178                 return ret;
 179
 180         irq_clear_status_flags(irq, IRQ_NOREQUEST|IRQ_NOAUTOEN);
 181
 182         return xen_evtchn_port_setup(info);
 183 }
 184
 185 static int xen_irq_info_evtchn_setup(unsigned irq,
 186                                      unsigned evtchn)
 187 {
 188         struct irq_info *info = info_for_irq(irq);
 189
 190         return xen_irq_info_common_setup(info, irq, IRQT_EVTCHN, evtchn, 0);
 191 }
 192
 193 static int xen_irq_info_ipi_setup(unsigned cpu,
 194                                   unsigned irq,
 195                                   unsigned evtchn,
 196                                   enum ipi_vector ipi)
 197 {
 198         struct irq_info *info = info_for_irq(irq);
 199
 200         info->u.ipi = ipi;
 201
 202         per_cpu(ipi_to_irq, cpu)[ipi] = irq;
 203
 204         return xen_irq_info_common_setup(info, irq, IRQT_IPI, evtchn, 0);
 205 }
 206
 207 static int xen_irq_info_virq_setup(unsigned cpu,
 208                                    unsigned irq,
 209                                    unsigned evtchn,
 210                                    unsigned virq)
 211 {
 212         struct irq_info *info = info_for_irq(irq);
 213
 214         info->u.virq = virq;
 215
 216         per_cpu(virq_to_irq, cpu)[virq] = irq;
 217
 218         return xen_irq_info_common_setup(info, irq, IRQT_VIRQ, evtchn, 0);
 219 }
 220
 221 static int xen_irq_info_pirq_setup(unsigned irq,
 222                                    unsigned evtchn,
 223                                    unsigned pirq,
 224                                    unsigned gsi,
 225                                    uint16_t domid,
 226                                    unsigned char flags)
 227 {
 228         struct irq_info *info = info_for_irq(irq);
 229
 230         info->u.pirq.pirq = pirq;
 231         info->u.pirq.gsi = gsi;
 232         info->u.pirq.domid = domid;
 233         info->u.pirq.flags = flags;
 234
 235         return xen_irq_info_common_setup(info, irq, IRQT_PIRQ, evtchn, 0);
 236 }
 237
 238 static void xen_irq_info_cleanup(struct irq_info *info)
 239 {
 240         set_evtchn_to_irq(info->evtchn, -1);
 241         info->evtchn = 0;
 242 }
 243
 244 /*
 245  * Accessors for packed IRQ information.
 246  */
 247 unsigned int evtchn_from_irq(unsigned irq)
 248 {
 249         if (unlikely(WARN(irq >= nr_irqs, "Invalid irq %d!\n", irq)))
 250                 return 0;
 251
 252         return info_for_irq(irq)->evtchn;
 253 }
 254
 255 unsigned irq_from_evtchn(unsigned int evtchn)
 256 {
 257         return get_evtchn_to_irq(evtchn);
 258 }
 259 EXPORT_SYMBOL_GPL(irq_from_evtchn);
 260
 261 int irq_from_virq(unsigned int cpu, unsigned int virq)
 262 {
 263         return per_cpu(virq_to_irq, cpu)[virq];
 264 }
 265
 266 static enum ipi_vector ipi_from_irq(unsigned irq)
 267 {
 268         struct irq_info *info = info_for_irq(irq);
 269
 270         BUG_ON(info == NULL);
 271         BUG_ON(info->type != IRQT_IPI);
 272
 273         return info->u.ipi;
 274 }
 275
 276 static unsigned virq_from_irq(unsigned irq)
 277 {
 278         struct irq_info *info = info_for_irq(irq);
 279
 280         BUG_ON(info == NULL);
 281         BUG_ON(info->type != IRQT_VIRQ);
 282
 283         return info->u.virq;
 284 }
 285
 286 static unsigned pirq_from_irq(unsigned irq)
 287 {
 288         struct irq_info *info = info_for_irq(irq);
 289
 290         BUG_ON(info == NULL);
 291         BUG_ON(info->type != IRQT_PIRQ);
 292
 293         return info->u.pirq.pirq;
 294 }
 295
 296 static enum xen_irq_type type_from_irq(unsigned irq)
 297 {
 298         return info_for_irq(irq)->type;
 299 }
 300
 301 unsigned cpu_from_irq(unsigned irq)
 302 {
 303         return info_for_irq(irq)->cpu;
 304 }
 305
 306 unsigned int cpu_from_evtchn(unsigned int evtchn)
 307 {
 308         int irq = get_evtchn_to_irq(evtchn);
 309         unsigned ret = 0;
 310
 311         if (irq != -1)
 312                 ret = cpu_from_irq(irq);
 313
 314         return ret;
 315 }
 316
 317 #ifdef CONFIG_X86
 318 static bool pirq_check_eoi_map(unsigned irq)
 319 {
 320         return test_bit(pirq_from_irq(irq), pirq_eoi_map);
 321 }
 322 #endif
 323
 324 static bool pirq_needs_eoi_flag(unsigned irq)
 325 {
 326         struct irq_info *info = info_for_irq(irq);
 327         BUG_ON(info->type != IRQT_PIRQ);
 328
 329         return info->u.pirq.flags & PIRQ_NEEDS_EOI;
 330 }
 331
 332 static void bind_evtchn_to_cpu(unsigned int chn, unsigned int cpu)
 333 {
 334         int irq = get_evtchn_to_irq(chn);
 335         struct irq_info *info = info_for_irq(irq);
 336
 337         BUG_ON(irq == -1);
 338 #ifdef CONFIG_SMP
 339         cpumask_copy(irq_get_affinity_mask(irq), cpumask_of(cpu));
 340 #endif
 341         xen_evtchn_port_bind_to_cpu(info, cpu);
 342
 343         info->cpu = cpu;
 344 }
 345
 346 /**
 347  * notify_remote_via_irq - send event to remote end of event channel via irq
 348  * @irq: irq of event channel to send event to
 349  *
 350  * Unlike notify_remote_via_evtchn(), this is safe to use across
 351  * save/restore. Notifications on a broken connection are silently
 352  * dropped.
 353  */
 354 void notify_remote_via_irq(int irq)
 355 {
 356         int evtchn = evtchn_from_irq(irq);
 357
 358         if (VALID_EVTCHN(evtchn))
 359                 notify_remote_via_evtchn(evtchn);
 360 }
 361 EXPORT_SYMBOL_GPL(notify_remote_via_irq);
 362
 363 static void xen_irq_init(unsigned irq)
 364 {
 365         struct irq_info *info;
 366 #ifdef CONFIG_SMP
 367         /* By default all event channels notify CPU#0. */
 368         cpumask_copy(irq_get_affinity_mask(irq), cpumask_of(0));
 369 #endif
 370
 371         info = kzalloc(sizeof(*info), GFP_KERNEL);
 372         if (info == NULL)
 373                 panic("Unable to allocate metadata for IRQ%d\n", irq);
 374
 375         info->type = IRQT_UNBOUND;
 376         info->refcnt = -1;
 377
 378         irq_set_handler_data(irq, info);
 379
 380         list_add_tail(&info->list, &xen_irq_list_head);
 381 }
 382
 383 static int __must_check xen_allocate_irqs_dynamic(int nvec)
 384 {
 385         int i, irq = irq_alloc_descs(-1, 0, nvec, -1);
 386
 387         if (irq >= 0) {
 388                 for (i = 0; i < nvec; i++)
 389                         xen_irq_init(irq + i);
 390         }
 391
 392         return irq;
 393 }
 394
 395 static inline int __must_check xen_allocate_irq_dynamic(void)
 396 {
 397
 398         return xen_allocate_irqs_dynamic(1);
 399 }
 400
 401 static int __must_check xen_allocate_irq_gsi(unsigned gsi)
 402 {
 403         int irq;
 404
 405         /*
 406          * A PV guest has no concept of a GSI (since it has no ACPI
 407          * nor access to/knowledge of the physical APICs). Therefore
 408          * all IRQs are dynamically allocated from the entire IRQ
 409          * space.
 410          */
 411         if (xen_pv_domain() && !xen_initial_domain())
 412                 return xen_allocate_irq_dynamic();
 413
 414         /* Legacy IRQ descriptors are already allocated by the arch. */
 415         if (gsi < nr_legacy_irqs())
 416                 irq = gsi;
 417         else
 418                 irq = irq_alloc_desc_at(gsi, -1);
 419
 420         xen_irq_init(irq);
 421
 422         return irq;
 423 }
 424
 425 static void xen_free_irq(unsigned irq)
 426 {
 427         struct irq_info *info = irq_get_handler_data(irq);
 428
 429         if (WARN_ON(!info))
 430                 return;
 431
 432         list_del(&info->list);
 433
 434         irq_set_handler_data(irq, NULL);
 435
 436         WARN_ON(info->refcnt > 0);
 437
 438         kfree(info);
 439
 440         /* Legacy IRQ descriptors are managed by the arch. */
 441         if (irq < nr_legacy_irqs())
 442                 return;
 443
 444         irq_free_desc(irq);
 445 }
 446
 447 static void xen_evtchn_close(unsigned int port)
 448 {
 449         struct evtchn_close close;
 450
 451         close.port = port;
 452         if (HYPERVISOR_event_channel_op(EVTCHNOP_close, &close) != 0)
 453                 BUG();
 454 }
 455
 456 static void pirq_query_unmask(int irq)
 457 {
 458         struct physdev_irq_status_query irq_status;
 459         struct irq_info *info = info_for_irq(irq);
 460
 461         BUG_ON(info->type != IRQT_PIRQ);
 462
 463         irq_status.irq = pirq_from_irq(irq);
 464         if (HYPERVISOR_physdev_op(PHYSDEVOP_irq_status_query, &irq_status))
 465                 irq_status.flags = 0;
 466
 467         info->u.pirq.flags &= ~PIRQ_NEEDS_EOI;
 468         if (irq_status.flags & XENIRQSTAT_needs_eoi)
 469                 info->u.pirq.flags |= PIRQ_NEEDS_EOI;
 470 }
 471
 472 static void eoi_pirq(struct irq_data *data)
 473 {
 474         int evtchn = evtchn_from_irq(data->irq);
 475         struct physdev_eoi eoi = { .irq = pirq_from_irq(data->irq) };
 476         int rc = 0;
 477
 478         if (!VALID_EVTCHN(evtchn))
 479                 return;
 480
 481         if (unlikely(irqd_is_setaffinity_pending(data)) &&
 482             likely(!irqd_irq_disabled(data))) {
 483                 int masked = test_and_set_mask(evtchn);
 484
 485                 clear_evtchn(evtchn);
 486
 487                 irq_move_masked_irq(data);
 488
 489                 if (!masked)
 490                         unmask_evtchn(evtchn);
 491         } else
 492                 clear_evtchn(evtchn);
 493
 494         if (pirq_needs_eoi(data->irq)) {
 495                 rc = HYPERVISOR_physdev_op(PHYSDEVOP_eoi, &eoi);
 496                 WARN_ON(rc);
 497         }
 498 }
 499
 500 static void mask_ack_pirq(struct irq_data *data)
 501 {
 502         disable_dynirq(data);
 503         eoi_pirq(data);
 504 }
 505
 506 static unsigned int __startup_pirq(unsigned int irq)
 507 {
 508         struct evtchn_bind_pirq bind_pirq;
 509         struct irq_info *info = info_for_irq(irq);
 510         int evtchn = evtchn_from_irq(irq);
 511         int rc;
 512
 513         BUG_ON(info->type != IRQT_PIRQ);
 514
 515         if (VALID_EVTCHN(evtchn))
 516                 goto out;
 517
 518         bind_pirq.pirq = pirq_from_irq(irq);
 519         /* NB. We are happy to share unless we are probing. */
 520         bind_pirq.flags = info->u.pirq.flags & PIRQ_SHAREABLE ?
 521                                         BIND_PIRQ__WILL_SHARE : 0;
 522         rc = HYPERVISOR_event_channel_op(EVTCHNOP_bind_pirq, &bind_pirq);
 523         if (rc != 0) {
 524                 pr_warn("Failed to obtain physical IRQ %d\n", irq);
 525                 return 0;
 526         }
 527         evtchn = bind_pirq.port;
 528
 529         pirq_query_unmask(irq);
 530
 531         rc = set_evtchn_to_irq(evtchn, irq);
 532         if (rc)
 533                 goto err;
 534
 535         info->evtchn = evtchn;
 536         bind_evtchn_to_cpu(evtchn, 0);
 537
 538         rc = xen_evtchn_port_setup(info);
 539         if (rc)
 540                 goto err;
 541
 542 out:
 543         unmask_evtchn(evtchn);
 544         eoi_pirq(irq_get_irq_data(irq));
 545
 546         return 0;
 547
 548 err:
 549         pr_err("irq%d: Failed to set port to irq mapping (%d)\n", irq, rc);
 550         xen_evtchn_close(evtchn);
 551         return 0;
 552 }
 553
 554 static unsigned int startup_pirq(struct irq_data *data)
 555 {
 556         return __startup_pirq(data->irq);
 557 }
 558
 559 static void shutdown_pirq(struct irq_data *data)
 560 {
 561         unsigned int irq = data->irq;
 562         struct irq_info *info = info_for_irq(irq);
 563         unsigned evtchn = evtchn_from_irq(irq);
 564
 565         BUG_ON(info->type != IRQT_PIRQ);
 566
 567         if (!VALID_EVTCHN(evtchn))
 568                 return;
 569
 570         mask_evtchn(evtchn);
 571         xen_evtchn_close(evtchn);
 572         xen_irq_info_cleanup(info);
 573 }
 574
 575 static void enable_pirq(struct irq_data *data)
 576 {
 577         enable_dynirq(data);
 578 }
 579
 580 static void disable_pirq(struct irq_data *data)
 581 {
 582         disable_dynirq(data);
 583 }
 584
 585 int xen_irq_from_gsi(unsigned gsi)
 586 {
 587         struct irq_info *info;
 588
 589         list_for_each_entry(info, &xen_irq_list_head, list) {
 590                 if (info->type != IRQT_PIRQ)
 591                         continue;
 592
 593                 if (info->u.pirq.gsi == gsi)
 594                         return info->irq;
 595         }
 596
 597         return -1;
 598 }
 599 EXPORT_SYMBOL_GPL(xen_irq_from_gsi);
 600
 601 static void __unbind_from_irq(unsigned int irq)
 602 {
 603         int evtchn = evtchn_from_irq(irq);
 604         struct irq_info *info = irq_get_handler_data(irq);
 605
 606         if (info->refcnt > 0) {
 607                 info->refcnt--;
 608                 if (info->refcnt != 0)
 609                         return;
 610         }
 611
 612         if (VALID_EVTCHN(evtchn)) {
 613                 unsigned int cpu = cpu_from_irq(irq);
 614
 615                 xen_evtchn_close(evtchn);
 616
 617                 switch (type_from_irq(irq)) {
 618                 case IRQT_VIRQ:
 619                         per_cpu(virq_to_irq, cpu)[virq_from_irq(irq)] = -1;
 620                         break;
 621                 case IRQT_IPI:
 622                         per_cpu(ipi_to_irq, cpu)[ipi_from_irq(irq)] = -1;
 623                         break;
 624                 default:
 625                         break;
 626                 }
 627
 628                 xen_irq_info_cleanup(info);
 629         }
 630
 631         xen_free_irq(irq);
 632 }
 633
 634 /*
 635  * Do not make any assumptions regarding the relationship between the
 636  * IRQ number returned here and the Xen pirq argument.
 637  *
 638  * Note: We don't assign an event channel until the irq actually started
 639  * up.  Return an existing irq if we've already got one for the gsi.
 640  *
 641  * Shareable implies level triggered, not shareable implies edge
 642  * triggered here.
 643  */
 644 int xen_bind_pirq_gsi_to_irq(unsigned gsi,
 645                              unsigned pirq, int shareable, char *name)
 646 {
 647         int irq = -1;
 648         struct physdev_irq irq_op;
 649         int ret;
 650
 651         mutex_lock(&irq_mapping_update_lock);
 652
 653         irq = xen_irq_from_gsi(gsi);
 654         if (irq != -1) {
 655                 pr_info("%s: returning irq %d for gsi %u\n",
 656                         __func__, irq, gsi);
 657                 goto out;
 658         }
 659
 660         irq = xen_allocate_irq_gsi(gsi);
 661         if (irq < 0)
 662                 goto out;
 663
 664         irq_op.irq = irq;
 665         irq_op.vector = 0;
 666
 667         /* Only the privileged domain can do this. For non-priv, the pcifront
 668          * driver provides a PCI bus that does the call to do exactly
 669          * this in the priv domain. */
 670         if (xen_initial_domain() &&
 671             HYPERVISOR_physdev_op(PHYSDEVOP_alloc_irq_vector, &irq_op)) {
 672                 xen_free_irq(irq);
 673                 irq = -ENOSPC;
 674                 goto out;
 675         }
 676
 677         ret = xen_irq_info_pirq_setup(irq, 0, pirq, gsi, DOMID_SELF,
 678                                shareable ? PIRQ_SHAREABLE : 0);
 679         if (ret < 0) {
 680                 __unbind_from_irq(irq);
 681                 irq = ret;
 682                 goto out;
 683         }
 684
 685         pirq_query_unmask(irq);
 686         /* We try to use the handler with the appropriate semantic for the
 687          * type of interrupt: if the interrupt is an edge triggered
 688          * interrupt we use handle_edge_irq.
 689          *
 690          * On the other hand if the interrupt is level triggered we use
 691          * handle_fasteoi_irq like the native code does for this kind of
 692          * interrupts.
 693          *
 694          * Depending on the Xen version, pirq_needs_eoi might return true
 695          * not only for level triggered interrupts but for edge triggered
 696          * interrupts too. In any case Xen always honors the eoi mechanism,
 697          * not injecting any more pirqs of the same kind if the first one
 698          * hasn't received an eoi yet. Therefore using the fasteoi handler
 699          * is the right choice either way.
 700          */
 701         if (shareable)
 702                 irq_set_chip_and_handler_name(irq, &xen_pirq_chip,
 703                                 handle_fasteoi_irq, name);
 704         else
 705                 irq_set_chip_and_handler_name(irq, &xen_pirq_chip,
 706                                 handle_edge_irq, name);
 707
 708 out:
 709         mutex_unlock(&irq_mapping_update_lock);
 710
 711         return irq;
 712 }
 713
 714 #ifdef CONFIG_PCI_MSI
 715 int xen_allocate_pirq_msi(struct pci_dev *dev, struct msi_desc *msidesc)
 716 {
 717         int rc;
 718         struct physdev_get_free_pirq op_get_free_pirq;
 719
 720         op_get_free_pirq.type = MAP_PIRQ_TYPE_MSI;
 721         rc = HYPERVISOR_physdev_op(PHYSDEVOP_get_free_pirq, &op_get_free_pirq);
 722
 723         WARN_ONCE(rc == -ENOSYS,
 724                   "hypervisor does not support the PHYSDEVOP_get_free_pirq interface\n");
 725
 726         return rc ? -1 : op_get_free_pirq.pirq;
 727 }
 728
 729 int xen_bind_pirq_msi_to_irq(struct pci_dev *dev, struct msi_desc *msidesc,
 730                              int pirq, int nvec, const char *name, domid_t domid)
 731 {
 732         int i, irq, ret;
 733
 734         mutex_lock(&irq_mapping_update_lock);
 735
 736         irq = xen_allocate_irqs_dynamic(nvec);
 737         if (irq < 0)
 738                 goto out;
 739
 740         for (i = 0; i < nvec; i++) {
 741                 irq_set_chip_and_handler_name(irq + i, &xen_pirq_chip, handle_edge_irq, name);
 742
 743                 ret = xen_irq_info_pirq_setup(irq + i, 0, pirq + i, 0, domid,
 744                                               i == 0 ? 0 : PIRQ_MSI_GROUP);
 745                 if (ret < 0)
 746                         goto error_irq;
 747         }
 748
 749         ret = irq_set_msi_desc(irq, msidesc);
 750         if (ret < 0)
 751                 goto error_irq;
 752 out:
 753         mutex_unlock(&irq_mapping_update_lock);
 754         return irq;
 755 error_irq:
 756         while (nvec--)
 757                 __unbind_from_irq(irq + nvec);
 758         mutex_unlock(&irq_mapping_update_lock);
 759         return ret;
 760 }
 761 #endif
 762
 763 int xen_destroy_irq(int irq)
 764 {
 765         struct physdev_unmap_pirq unmap_irq;
 766         struct irq_info *info = info_for_irq(irq);
 767         int rc = -ENOENT;
 768
 769         mutex_lock(&irq_mapping_update_lock);
 770
 771         /*
 772          * If trying to remove a vector in a MSI group different
 773          * than the first one skip the PIRQ unmap unless this vector
 774          * is the first one in the group.
 775          */
 776         if (xen_initial_domain() && !(info->u.pirq.flags & PIRQ_MSI_GROUP)) {
 777                 unmap_irq.pirq = info->u.pirq.pirq;
 778                 unmap_irq.domid = info->u.pirq.domid;
 779                 rc = HYPERVISOR_physdev_op(PHYSDEVOP_unmap_pirq, &unmap_irq);
 780                 /* If another domain quits without making the pci_disable_msix
 781                  * call, the Xen hypervisor takes care of freeing the PIRQs
 782                  * (free_domain_pirqs).
 783                  */
 784                 if ((rc == -ESRCH && info->u.pirq.domid != DOMID_SELF))
 785                         pr_info("domain %d does not have %d anymore\n",
 786                                 info->u.pirq.domid, info->u.pirq.pirq);
 787                 else if (rc) {
 788                         pr_warn("unmap irq failed %d\n", rc);
 789                         goto out;
 790                 }
 791         }
 792
 793         xen_free_irq(irq);
 794
 795 out:
 796         mutex_unlock(&irq_mapping_update_lock);
 797         return rc;
 798 }
 799
 800 int xen_irq_from_pirq(unsigned pirq)
 801 {
 802         int irq;
 803
 804         struct irq_info *info;
 805
 806         mutex_lock(&irq_mapping_update_lock);
 807
 808         list_for_each_entry(info, &xen_irq_list_head, list) {
 809                 if (info->type != IRQT_PIRQ)
 810                         continue;
 811                 irq = info->irq;
 812                 if (info->u.pirq.pirq == pirq)
 813                         goto out;
 814         }
 815         irq = -1;
 816 out:
 817         mutex_unlock(&irq_mapping_update_lock);
 818
 819         return irq;
 820 }
 821
 822
 823 int xen_pirq_from_irq(unsigned irq)
 824 {
 825         return pirq_from_irq(irq);
 826 }
 827 EXPORT_SYMBOL_GPL(xen_pirq_from_irq);
 828
 829 int bind_evtchn_to_irq(unsigned int evtchn)
 830 {
 831         int irq;
 832         int ret;
 833
 834         if (evtchn >= xen_evtchn_max_channels())
 835                 return -ENOMEM;
 836
 837         mutex_lock(&irq_mapping_update_lock);
 838
 839         irq = get_evtchn_to_irq(evtchn);
 840
 841         if (irq == -1) {
 842                 irq = xen_allocate_irq_dynamic();
 843                 if (irq < 0)
 844                         goto out;
 845
 846                 irq_set_chip_and_handler_name(irq, &xen_dynamic_chip,
 847                                               handle_edge_irq, "event");
 848
 849                 ret = xen_irq_info_evtchn_setup(irq, evtchn);
 850                 if (ret < 0) {
 851                         __unbind_from_irq(irq);
 852                         irq = ret;
 853                         goto out;
 854                 }
 855                 /* New interdomain events are bound to VCPU 0. */
 856                 bind_evtchn_to_cpu(evtchn, 0);
 857         } else {
 858                 struct irq_info *info = info_for_irq(irq);
 859                 WARN_ON(info == NULL || info->type != IRQT_EVTCHN);
 860         }
 861
 862 out:
 863         mutex_unlock(&irq_mapping_update_lock);
 864
 865         return irq;
 866 }
 867 EXPORT_SYMBOL_GPL(bind_evtchn_to_irq);
 868
 869 static int bind_ipi_to_irq(unsigned int ipi, unsigned int cpu)
 870 {
 871         struct evtchn_bind_ipi bind_ipi;
 872         int evtchn, irq;
 873         int ret;
 874
 875         mutex_lock(&irq_mapping_update_lock);
 876
 877         irq = per_cpu(ipi_to_irq, cpu)[ipi];
 878
 879         if (irq == -1) {
 880                 irq = xen_allocate_irq_dynamic();
 881                 if (irq < 0)
 882                         goto out;
 883
 884                 irq_set_chip_and_handler_name(irq, &xen_percpu_chip,
 885                                               handle_percpu_irq, "ipi");
 886
 887                 bind_ipi.vcpu = xen_vcpu_nr(cpu);
 888                 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_ipi,
 889                                                 &bind_ipi) != 0)
 890                         BUG();
 891                 evtchn = bind_ipi.port;
 892
 893                 ret = xen_irq_info_ipi_setup(cpu, irq, evtchn, ipi);
 894                 if (ret < 0) {
 895                         __unbind_from_irq(irq);
 896                         irq = ret;
 897                         goto out;
 898                 }
 899                 bind_evtchn_to_cpu(evtchn, cpu);
 900         } else {
 901                 struct irq_info *info = info_for_irq(irq);
 902                 WARN_ON(info == NULL || info->type != IRQT_IPI);
 903         }
 904
 905  out:
 906         mutex_unlock(&irq_mapping_update_lock);
 907         return irq;
 908 }
 909
 910 int bind_interdomain_evtchn_to_irq(unsigned int remote_domain,
 911                                    unsigned int remote_port)
 912 {
 913         struct evtchn_bind_interdomain bind_interdomain;
 914         int err;
 915
 916         bind_interdomain.remote_dom  = remote_domain;
 917         bind_interdomain.remote_port = remote_port;
 918
 919         err = HYPERVISOR_event_channel_op(EVTCHNOP_bind_interdomain,
 920                                           &bind_interdomain);
 921
 922         return err ? : bind_evtchn_to_irq(bind_interdomain.local_port);
 923 }
 924 EXPORT_SYMBOL_GPL(bind_interdomain_evtchn_to_irq);
 925
 926 static int find_virq(unsigned int virq, unsigned int cpu)
 927 {
 928         struct evtchn_status status;
 929         int port, rc = -ENOENT;
 930
 931         memset(&status, 0, sizeof(status));
 932         for (port = 0; port < xen_evtchn_max_channels(); port++) {
 933                 status.dom = DOMID_SELF;
 934                 status.port = port;
 935                 rc = HYPERVISOR_event_channel_op(EVTCHNOP_status, &status);
 936                 if (rc < 0)
 937                         continue;
 938                 if (status.status != EVTCHNSTAT_virq)
 939                         continue;
 940                 if (status.u.virq == virq && status.vcpu == xen_vcpu_nr(cpu)) {
 941                         rc = port;
 942                         break;
 943                 }
 944         }
 945         return rc;
 946 }
 947
 948 /**
 949  * xen_evtchn_nr_channels - number of usable event channel ports
 950  *
 951  * This may be less than the maximum supported by the current
 952  * hypervisor ABI. Use xen_evtchn_max_channels() for the maximum
 953  * supported.
 954  */
 955 unsigned xen_evtchn_nr_channels(void)
 956 {
 957         return evtchn_ops->nr_channels();
 958 }
 959 EXPORT_SYMBOL_GPL(xen_evtchn_nr_channels);
 960
 961 int bind_virq_to_irq(unsigned int virq, unsigned int cpu, bool percpu)
 962 {
 963         struct evtchn_bind_virq bind_virq;
 964         int evtchn, irq, ret;
 965
 966         mutex_lock(&irq_mapping_update_lock);
 967
 968         irq = per_cpu(virq_to_irq, cpu)[virq];
 969
 970         if (irq == -1) {
 971                 irq = xen_allocate_irq_dynamic();
 972                 if (irq < 0)
 973                         goto out;
 974
 975                 if (percpu)
 976                         irq_set_chip_and_handler_name(irq, &xen_percpu_chip,
 977                                                       handle_percpu_irq, "virq");
 978                 else
 979                         irq_set_chip_and_handler_name(irq, &xen_dynamic_chip,
 980                                                       handle_edge_irq, "virq");
 981
 982                 bind_virq.virq = virq;
 983                 bind_virq.vcpu = xen_vcpu_nr(cpu);
 984                 ret = HYPERVISOR_event_channel_op(EVTCHNOP_bind_virq,
 985                                                 &bind_virq);
 986                 if (ret == 0)
 987                         evtchn = bind_virq.port;
 988                 else {
 989                         if (ret == -EEXIST)
 990                                 ret = find_virq(virq, cpu);
 991                         BUG_ON(ret < 0);
 992                         evtchn = ret;
 993                 }
 994
 995                 ret = xen_irq_info_virq_setup(cpu, irq, evtchn, virq);
 996                 if (ret < 0) {
 997                         __unbind_from_irq(irq);
 998                         irq = ret;
 999                         goto out;
1000                 }
1001
1002                 bind_evtchn_to_cpu(evtchn, cpu);
1003         } else {
1004                 struct irq_info *info = info_for_irq(irq);
1005                 WARN_ON(info == NULL || info->type != IRQT_VIRQ);
1006         }
1007
1008 out:
1009         mutex_unlock(&irq_mapping_update_lock);
1010
1011         return irq;
1012 }
1013
1014 static void unbind_from_irq(unsigned int irq)
1015 {
1016         mutex_lock(&irq_mapping_update_lock);
1017         __unbind_from_irq(irq);
1018         mutex_unlock(&irq_mapping_update_lock);
1019 }
1020
1021 int bind_evtchn_to_irqhandler(unsigned int evtchn,
1022                               irq_handler_t handler,
1023                               unsigned long irqflags,
1024                               const char *devname, void *dev_id)
1025 {
1026         int irq, retval;
1027
1028         irq = bind_evtchn_to_irq(evtchn);
1029         if (irq < 0)
1030                 return irq;
1031         retval = request_irq(irq, handler, irqflags, devname, dev_id);
1032         if (retval != 0) {
1033                 unbind_from_irq(irq);
1034                 return retval;
1035         }
1036
1037         return irq;
1038 }
1039 EXPORT_SYMBOL_GPL(bind_evtchn_to_irqhandler);
1040
1041 int bind_interdomain_evtchn_to_irqhandler(unsigned int remote_domain,
1042                                           unsigned int remote_port,
1043                                           irq_handler_t handler,
1044                                           unsigned long irqflags,
1045                                           const char *devname,
1046                                           void *dev_id)
1047 {
1048         int irq, retval;
1049
1050         irq = bind_interdomain_evtchn_to_irq(remote_domain, remote_port);
1051         if (irq < 0)
1052                 return irq;
1053
1054         retval = request_irq(irq, handler, irqflags, devname, dev_id);
1055         if (retval != 0) {
1056                 unbind_from_irq(irq);
1057                 return retval;
1058         }
1059
1060         return irq;
1061 }
1062 EXPORT_SYMBOL_GPL(bind_interdomain_evtchn_to_irqhandler);
1063
1064 int bind_virq_to_irqhandler(unsigned int virq, unsigned int cpu,
1065                             irq_handler_t handler,
1066                             unsigned long irqflags, const char *devname, void *dev_id)
1067 {
1068         int irq, retval;
1069
1070         irq = bind_virq_to_irq(virq, cpu, irqflags & IRQF_PERCPU);
1071         if (irq < 0)
1072                 return irq;
1073         retval = request_irq(irq, handler, irqflags, devname, dev_id);
1074         if (retval != 0) {
1075                 unbind_from_irq(irq);
1076                 return retval;
1077         }
1078
1079         return irq;
1080 }
1081 EXPORT_SYMBOL_GPL(bind_virq_to_irqhandler);
1082
1083 int bind_ipi_to_irqhandler(enum ipi_vector ipi,
1084                            unsigned int cpu,
1085                            irq_handler_t handler,
1086                            unsigned long irqflags,
1087                            const char *devname,
1088                            void *dev_id)
1089 {
1090         int irq, retval;
1091
1092         irq = bind_ipi_to_irq(ipi, cpu);
1093         if (irq < 0)
1094                 return irq;
1095
1096         irqflags |= IRQF_NO_SUSPEND | IRQF_FORCE_RESUME | IRQF_EARLY_RESUME;
1097         retval = request_irq(irq, handler, irqflags, devname, dev_id);
1098         if (retval != 0) {
1099                 unbind_from_irq(irq);
1100                 return retval;
1101         }
1102
1103         return irq;
1104 }
1105
1106 void unbind_from_irqhandler(unsigned int irq, void *dev_id)
1107 {
1108         struct irq_info *info = irq_get_handler_data(irq);
1109
1110         if (WARN_ON(!info))
1111                 return;
1112         free_irq(irq, dev_id);
1113         unbind_from_irq(irq);
1114 }
1115 EXPORT_SYMBOL_GPL(unbind_from_irqhandler);
1116
1117 /**
1118  * xen_set_irq_priority() - set an event channel priority.
1119  * @irq:irq bound to an event channel.
1120  * @priority: priority between XEN_IRQ_PRIORITY_MAX and XEN_IRQ_PRIORITY_MIN.
1121  */
1122 int xen_set_irq_priority(unsigned irq, unsigned priority)
1123 {
1124         struct evtchn_set_priority set_priority;
1125
1126         set_priority.port = evtchn_from_irq(irq);
1127         set_priority.priority = priority;
1128
1129         return HYPERVISOR_event_channel_op(EVTCHNOP_set_priority,
1130                                            &set_priority);
1131 }
1132 EXPORT_SYMBOL_GPL(xen_set_irq_priority);
1133
1134 int evtchn_make_refcounted(unsigned int evtchn)
1135 {
1136         int irq = get_evtchn_to_irq(evtchn);
1137         struct irq_info *info;
1138
1139         if (irq == -1)
1140                 return -ENOENT;
1141
1142         info = irq_get_handler_data(irq);
1143
1144         if (!info)
1145                 return -ENOENT;
1146
1147         WARN_ON(info->refcnt != -1);
1148
1149         info->refcnt = 1;
1150
1151         return 0;
1152 }
1153 EXPORT_SYMBOL_GPL(evtchn_make_refcounted);
1154
1155 int evtchn_get(unsigned int evtchn)
1156 {
1157         int irq;
1158         struct irq_info *info;
1159         int err = -ENOENT;
1160
1161         if (evtchn >= xen_evtchn_max_channels())
1162                 return -EINVAL;
1163
1164         mutex_lock(&irq_mapping_update_lock);
1165
1166         irq = get_evtchn_to_irq(evtchn);
1167         if (irq == -1)
1168                 goto done;
1169
1170         info = irq_get_handler_data(irq);
1171
1172         if (!info)
1173                 goto done;
1174
1175         err = -EINVAL;
1176         if (info->refcnt <= 0)
1177                 goto done;
1178
1179         info->refcnt++;
1180         err = 0;
1181  done:
1182         mutex_unlock(&irq_mapping_update_lock);
1183
1184         return err;
1185 }
1186 EXPORT_SYMBOL_GPL(evtchn_get);
1187
1188 void evtchn_put(unsigned int evtchn)
1189 {
1190         int irq = get_evtchn_to_irq(evtchn);
1191         if (WARN_ON(irq == -1))
1192                 return;
1193         unbind_from_irq(irq);
1194 }
1195 EXPORT_SYMBOL_GPL(evtchn_put);
1196
1197 void xen_send_IPI_one(unsigned int cpu, enum ipi_vector vector)
1198 {
1199         int irq;
1200
1201 #ifdef CONFIG_X86
1202         if (unlikely(vector == XEN_NMI_VECTOR)) {
1203                 int rc =  HYPERVISOR_vcpu_op(VCPUOP_send_nmi, xen_vcpu_nr(cpu),
1204                                              NULL);
1205                 if (rc < 0)
1206                         printk(KERN_WARNING "Sending nmi to CPU%d failed (rc:%d)\n", cpu, rc);
1207                 return;
1208         }
1209 #endif
1210         irq = per_cpu(ipi_to_irq, cpu)[vector];
1211         BUG_ON(irq < 0);
1212         notify_remote_via_irq(irq);
1213 }
1214
1215 static DEFINE_PER_CPU(unsigned, xed_nesting_count);
1216
1217 static void __xen_evtchn_do_upcall(void)
1218 {
1219         struct vcpu_info *vcpu_info = __this_cpu_read(xen_vcpu);
1220         int cpu = get_cpu();
1221         unsigned count;
1222
1223         do {
1224                 vcpu_info->evtchn_upcall_pending = 0;
1225
1226                 if (__this_cpu_inc_return(xed_nesting_count) - 1)
1227                         goto out;
1228
1229                 xen_evtchn_handle_events(cpu);
1230
1231                 BUG_ON(!irqs_disabled());
1232
1233                 count = __this_cpu_read(xed_nesting_count);
1234                 __this_cpu_write(xed_nesting_count, 0);
1235         } while (count != 1 || vcpu_info->evtchn_upcall_pending);
1236
1237 out:
1238
1239         put_cpu();
1240 }
1241
1242 void xen_evtchn_do_upcall(struct pt_regs *regs)
1243 {
1244         struct pt_regs *old_regs = set_irq_regs(regs);
1245
1246         irq_enter();
1247 #ifdef CONFIG_X86
1248         inc_irq_stat(irq_hv_callback_count);
1249 #endif
1250
1251         __xen_evtchn_do_upcall();
1252
1253         irq_exit();
1254         set_irq_regs(old_regs);
1255 }
1256
1257 void xen_hvm_evtchn_do_upcall(void)
1258 {
1259         __xen_evtchn_do_upcall();
1260 }
1261 EXPORT_SYMBOL_GPL(xen_hvm_evtchn_do_upcall);
1262
1263 /* Rebind a new event channel to an existing irq. */
1264 void rebind_evtchn_irq(int evtchn, int irq)
1265 {
1266         struct irq_info *info = info_for_irq(irq);
1267
1268         if (WARN_ON(!info))
1269                 return;
1270
1271         /* Make sure the irq is masked, since the new event channel
1272            will also be masked. */
1273         disable_irq(irq);
1274
1275         mutex_lock(&irq_mapping_update_lock);
1276
1277         /* After resume the irq<->evtchn mappings are all cleared out */
1278         BUG_ON(get_evtchn_to_irq(evtchn) != -1);
1279         /* Expect irq to have been bound before,
1280            so there should be a proper type */
1281         BUG_ON(info->type == IRQT_UNBOUND);
1282
1283         (void)xen_irq_info_evtchn_setup(irq, evtchn);
1284
1285         mutex_unlock(&irq_mapping_update_lock);
1286
1287         bind_evtchn_to_cpu(evtchn, info->cpu);
1288         /* This will be deferred until interrupt is processed */
1289         irq_set_affinity(irq, cpumask_of(info->cpu));
1290
1291         /* Unmask the event channel. */
1292         enable_irq(irq);
1293 }
1294
1295 /* Rebind an evtchn so that it gets delivered to a specific cpu */
1296 int xen_rebind_evtchn_to_cpu(int evtchn, unsigned tcpu)
1297 {
1298         struct evtchn_bind_vcpu bind_vcpu;
1299         int masked;
1300
1301         if (!VALID_EVTCHN(evtchn))
1302                 return -1;
1303
1304         if (!xen_support_evtchn_rebind())
1305                 return -1;
1306
1307         /* Send future instances of this interrupt to other vcpu. */
1308         bind_vcpu.port = evtchn;
1309         bind_vcpu.vcpu = xen_vcpu_nr(tcpu);
1310
1311         /*
1312          * Mask the event while changing the VCPU binding to prevent
1313          * it being delivered on an unexpected VCPU.
1314          */
1315         masked = test_and_set_mask(evtchn);
1316
1317         /*
1318          * If this fails, it usually just indicates that we're dealing with a
1319          * virq or IPI channel, which don't actually need to be rebound. Ignore
1320          * it, but don't do the xenlinux-level rebind in that case.
1321          */
1322         if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_vcpu, &bind_vcpu) >= 0)
1323                 bind_evtchn_to_cpu(evtchn, tcpu);
1324
1325         if (!masked)
1326                 unmask_evtchn(evtchn);
1327
1328         return 0;
1329 }
1330 EXPORT_SYMBOL_GPL(xen_rebind_evtchn_to_cpu);
1331
1332 static int set_affinity_irq(struct irq_data *data, const struct cpumask *dest,
1333                             bool force)
1334 {
1335         unsigned tcpu = cpumask_first_and(dest, cpu_online_mask);
1336         int ret = xen_rebind_evtchn_to_cpu(evtchn_from_irq(data->irq), tcpu);
1337
1338         if (!ret)
1339                 irq_data_update_effective_affinity(data, cpumask_of(tcpu));
1340
1341         return ret;
1342 }
1343
1344 static void enable_dynirq(struct irq_data *data)
1345 {
1346         int evtchn = evtchn_from_irq(data->irq);
1347
1348         if (VALID_EVTCHN(evtchn))
1349                 unmask_evtchn(evtchn);
1350 }
1351
1352 static void disable_dynirq(struct irq_data *data)
1353 {
1354         int evtchn = evtchn_from_irq(data->irq);
1355
1356         if (VALID_EVTCHN(evtchn))
1357                 mask_evtchn(evtchn);
1358 }
1359
1360 static void ack_dynirq(struct irq_data *data)
1361 {
1362         int evtchn = evtchn_from_irq(data->irq);
1363
1364         if (!VALID_EVTCHN(evtchn))
1365                 return;
1366
1367         if (unlikely(irqd_is_setaffinity_pending(data)) &&
1368             likely(!irqd_irq_disabled(data))) {
1369                 int masked = test_and_set_mask(evtchn);
1370
1371                 clear_evtchn(evtchn);
1372
1373                 irq_move_masked_irq(data);
1374
1375                 if (!masked)
1376                         unmask_evtchn(evtchn);
1377         } else
1378                 clear_evtchn(evtchn);
1379 }
1380
1381 static void mask_ack_dynirq(struct irq_data *data)
1382 {
1383         disable_dynirq(data);
1384         ack_dynirq(data);
1385 }
1386
1387 static int retrigger_dynirq(struct irq_data *data)
1388 {
1389         unsigned int evtchn = evtchn_from_irq(data->irq);
1390         int masked;
1391
1392         if (!VALID_EVTCHN(evtchn))
1393                 return 0;
1394
1395         masked = test_and_set_mask(evtchn);
1396         set_evtchn(evtchn);
1397         if (!masked)
1398                 unmask_evtchn(evtchn);
1399
1400         return 1;
1401 }
1402
1403 static void restore_pirqs(void)
1404 {
1405         int pirq, rc, irq, gsi;
1406         struct physdev_map_pirq map_irq;
1407         struct irq_info *info;
1408
1409         list_for_each_entry(info, &xen_irq_list_head, list) {
1410                 if (info->type != IRQT_PIRQ)
1411                         continue;
1412
1413                 pirq = info->u.pirq.pirq;
1414                 gsi = info->u.pirq.gsi;
1415                 irq = info->irq;
1416
1417                 /* save/restore of PT devices doesn't work, so at this point the
1418                  * only devices present are GSI based emulated devices */
1419                 if (!gsi)
1420                         continue;
1421
1422                 map_irq.domid = DOMID_SELF;
1423                 map_irq.type = MAP_PIRQ_TYPE_GSI;
1424                 map_irq.index = gsi;
1425                 map_irq.pirq = pirq;
1426
1427                 rc = HYPERVISOR_physdev_op(PHYSDEVOP_map_pirq, &map_irq);
1428                 if (rc) {
1429                         pr_warn("xen map irq failed gsi=%d irq=%d pirq=%d rc=%d\n",
1430                                 gsi, irq, pirq, rc);
1431                         xen_free_irq(irq);
1432                         continue;
1433                 }
1434
1435                 printk(KERN_DEBUG "xen: --> irq=%d, pirq=%d\n", irq, map_irq.pirq);
1436
1437                 __startup_pirq(irq);
1438         }
1439 }
1440
1441 static void restore_cpu_virqs(unsigned int cpu)
1442 {
1443         struct evtchn_bind_virq bind_virq;
1444         int virq, irq, evtchn;
1445
1446         for (virq = 0; virq < NR_VIRQS; virq++) {
1447                 if ((irq = per_cpu(virq_to_irq, cpu)[virq]) == -1)
1448                         continue;
1449
1450                 BUG_ON(virq_from_irq(irq) != virq);
1451
1452                 /* Get a new binding from Xen. */
1453                 bind_virq.virq = virq;
1454                 bind_virq.vcpu = xen_vcpu_nr(cpu);
1455                 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_virq,
1456                                                 &bind_virq) != 0)
1457                         BUG();
1458                 evtchn = bind_virq.port;
1459
1460                 /* Record the new mapping. */
1461                 (void)xen_irq_info_virq_setup(cpu, irq, evtchn, virq);
1462                 bind_evtchn_to_cpu(evtchn, cpu);
1463         }
1464 }
1465
1466 static void restore_cpu_ipis(unsigned int cpu)
1467 {
1468         struct evtchn_bind_ipi bind_ipi;
1469         int ipi, irq, evtchn;
1470
1471         for (ipi = 0; ipi < XEN_NR_IPIS; ipi++) {
1472                 if ((irq = per_cpu(ipi_to_irq, cpu)[ipi]) == -1)
1473                         continue;
1474
1475                 BUG_ON(ipi_from_irq(irq) != ipi);
1476
1477                 /* Get a new binding from Xen. */
1478                 bind_ipi.vcpu = xen_vcpu_nr(cpu);
1479                 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_ipi,
1480                                                 &bind_ipi) != 0)
1481                         BUG();
1482                 evtchn = bind_ipi.port;
1483
1484                 /* Record the new mapping. */
1485                 (void)xen_irq_info_ipi_setup(cpu, irq, evtchn, ipi);
1486                 bind_evtchn_to_cpu(evtchn, cpu);
1487         }
1488 }
1489
1490 /* Clear an irq's pending state, in preparation for polling on it */
1491 void xen_clear_irq_pending(int irq)
1492 {
1493         int evtchn = evtchn_from_irq(irq);
1494
1495         if (VALID_EVTCHN(evtchn))
1496                 clear_evtchn(evtchn);
1497 }
1498 EXPORT_SYMBOL(xen_clear_irq_pending);
1499 void xen_set_irq_pending(int irq)
1500 {
1501         int evtchn = evtchn_from_irq(irq);
1502
1503         if (VALID_EVTCHN(evtchn))
1504                 set_evtchn(evtchn);
1505 }
1506
1507 bool xen_test_irq_pending(int irq)
1508 {
1509         int evtchn = evtchn_from_irq(irq);
1510         bool ret = false;
1511
1512         if (VALID_EVTCHN(evtchn))
1513                 ret = test_evtchn(evtchn);
1514
1515         return ret;
1516 }
1517
1518 /* Poll waiting for an irq to become pending with timeout.  In the usual case,
1519  * the irq will be disabled so it won't deliver an interrupt. */
1520 void xen_poll_irq_timeout(int irq, u64 timeout)
1521 {
1522         evtchn_port_t evtchn = evtchn_from_irq(irq);
1523
1524         if (VALID_EVTCHN(evtchn)) {
1525                 struct sched_poll poll;
1526
1527                 poll.nr_ports = 1;
1528                 poll.timeout = timeout;
1529                 set_xen_guest_handle(poll.ports, &evtchn);
1530
1531                 if (HYPERVISOR_sched_op(SCHEDOP_poll, &poll) != 0)
1532                         BUG();
1533         }
1534 }
1535 EXPORT_SYMBOL(xen_poll_irq_timeout);
1536 /* Poll waiting for an irq to become pending.  In the usual case, the
1537  * irq will be disabled so it won't deliver an interrupt. */
1538 void xen_poll_irq(int irq)
1539 {
1540         xen_poll_irq_timeout(irq, 0 /* no timeout */);
1541 }
1542
1543 /* Check whether the IRQ line is shared with other guests. */
1544 int xen_test_irq_shared(int irq)
1545 {
1546         struct irq_info *info = info_for_irq(irq);
1547         struct physdev_irq_status_query irq_status;
1548
1549         if (WARN_ON(!info))
1550                 return -ENOENT;
1551
1552         irq_status.irq = info->u.pirq.pirq;
1553
1554         if (HYPERVISOR_physdev_op(PHYSDEVOP_irq_status_query, &irq_status))
1555                 return 0;
1556         return !(irq_status.flags & XENIRQSTAT_shared);
1557 }
1558 EXPORT_SYMBOL_GPL(xen_test_irq_shared);
1559
1560 void xen_irq_resume(void)
1561 {
1562         unsigned int cpu;
1563         struct irq_info *info;
1564
1565         /* New event-channel space is not 'live' yet. */
1566         xen_evtchn_resume();
1567
1568         /* No IRQ <-> event-channel mappings. */
1569         list_for_each_entry(info, &xen_irq_list_head, list)
1570                 info->evtchn = 0; /* zap event-channel binding */
1571
1572         clear_evtchn_to_irq_all();
1573
1574         for_each_possible_cpu(cpu) {
1575                 restore_cpu_virqs(cpu);
1576                 restore_cpu_ipis(cpu);
1577         }
1578
1579         restore_pirqs();
1580 }
1581
1582 static struct irq_chip xen_dynamic_chip __read_mostly = {
1583         .name                   = "xen-dyn",
1584
1585         .irq_disable            = disable_dynirq,
1586         .irq_mask               = disable_dynirq,
1587         .irq_unmask             = enable_dynirq,
1588
1589         .irq_ack                = ack_dynirq,
1590         .irq_mask_ack           = mask_ack_dynirq,
1591
1592         .irq_set_affinity       = set_affinity_irq,
1593         .irq_retrigger          = retrigger_dynirq,
1594 };
1595
1596 static struct irq_chip xen_pirq_chip __read_mostly = {
1597         .name                   = "xen-pirq",
1598
1599         .irq_startup            = startup_pirq,
1600         .irq_shutdown           = shutdown_pirq,
1601         .irq_enable             = enable_pirq,
1602         .irq_disable            = disable_pirq,
1603
1604         .irq_mask               = disable_dynirq,
1605         .irq_unmask             = enable_dynirq,
1606
1607         .irq_ack                = eoi_pirq,
1608         .irq_eoi                = eoi_pirq,
1609         .irq_mask_ack           = mask_ack_pirq,
1610
1611         .irq_set_affinity       = set_affinity_irq,
1612
1613         .irq_retrigger          = retrigger_dynirq,
1614 };
1615
1616 static struct irq_chip xen_percpu_chip __read_mostly = {
1617         .name                   = "xen-percpu",
1618
1619         .irq_disable            = disable_dynirq,
1620         .irq_mask               = disable_dynirq,
1621         .irq_unmask             = enable_dynirq,
1622
1623         .irq_ack                = ack_dynirq,
1624 };
1625
1626 int xen_set_callback_via(uint64_t via)
1627 {
1628         struct xen_hvm_param a;
1629         a.domid = DOMID_SELF;
1630         a.index = HVM_PARAM_CALLBACK_IRQ;
1631         a.value = via;
1632         return HYPERVISOR_hvm_op(HVMOP_set_param, &a);
1633 }
1634 EXPORT_SYMBOL_GPL(xen_set_callback_via);
1635
1636 #ifdef CONFIG_XEN_PVHVM
1637 /* Vector callbacks are better than PCI interrupts to receive event
1638  * channel notifications because we can receive vector callbacks on any
1639  * vcpu and we don't need PCI support or APIC interactions. */
1640 void xen_callback_vector(void)
1641 {
1642         int rc;
1643         uint64_t callback_via;
1644
1645         if (xen_have_vector_callback) {
1646                 callback_via = HVM_CALLBACK_VECTOR(HYPERVISOR_CALLBACK_VECTOR);
1647                 rc = xen_set_callback_via(callback_via);
1648                 if (rc) {
1649                         pr_err("Request for Xen HVM callback vector failed\n");
1650                         xen_have_vector_callback = 0;
1651                         return;
1652                 }
1653                 pr_info("Xen HVM callback vector for event delivery is enabled\n");
1654                 alloc_intr_gate(HYPERVISOR_CALLBACK_VECTOR,
1655                                 xen_hvm_callback_vector);
1656         }
1657 }
1658 #else
1659 void xen_callback_vector(void) {}
1660 #endif
1661
1662 #undef MODULE_PARAM_PREFIX
1663 #define MODULE_PARAM_PREFIX "xen."
1664
1665 static bool fifo_events = true;
1666 module_param(fifo_events, bool, 0);
1667
1668 void __init xen_init_IRQ(void)
1669 {
1670         int ret = -EINVAL;
1671         unsigned int evtchn;
1672
1673         if (fifo_events)
1674                 ret = xen_evtchn_fifo_init();
1675         if (ret < 0)
1676                 xen_evtchn_2l_init();
1677
1678         evtchn_to_irq = kcalloc(EVTCHN_ROW(xen_evtchn_max_channels()),
1679                                 sizeof(*evtchn_to_irq), GFP_KERNEL);
1680         BUG_ON(!evtchn_to_irq);
1681
1682         /* No event channels are 'live' right now. */
1683         for (evtchn = 0; evtchn < xen_evtchn_nr_channels(); evtchn++)
1684                 mask_evtchn(evtchn);
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                 if (rc != 0) {
1710                         free_page((unsigned long) pirq_eoi_map);
1711                         pirq_eoi_map = NULL;
1712                 } else
1713                         pirq_needs_eoi = pirq_check_eoi_map;
1714         }
1715 #endif
1716 }