drivers/xen/events/events_base.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * Xen event channels
   4  *
   5  * Xen models interrupts with abstract event channels.  Because each
   6  * domain gets 1024 event channels, but NR_IRQ is not that large, we
   7  * must dynamically map irqs<->event channels.  The event channels
   8  * interface with the rest of the kernel by defining a xen interrupt
   9  * chip.  When an event is received, it is mapped to an irq and sent
  10  * through the normal interrupt processing path.
  11  *
  12  * There are four kinds of events which can be mapped to an event
  13  * channel:
  14  *
  15  * 1. Inter-domain notifications.  This includes all the virtual
  16  *    device events, since they're driven by front-ends in another domain
  17  *    (typically dom0).
  18  * 2. VIRQs, typically used for timers.  These are per-cpu events.
  19  * 3. IPIs.
  20  * 4. PIRQs - Hardware interrupts.
  21  *
  22  * Jeremy Fitzhardinge <jeremy@xensource.com>, XenSource Inc, 2007
  23  */
  24
  25 #define pr_fmt(fmt) "xen:" KBUILD_MODNAME ": " fmt
  26
  27 #include <linux/linkage.h>
  28 #include <linux/interrupt.h>
  29 #include <linux/irq.h>
  30 #include <linux/moduleparam.h>
  31 #include <linux/string.h>
  32 #include <linux/memblock.h>
  33 #include <linux/slab.h>
  34 #include <linux/irqnr.h>
  35 #include <linux/pci.h>
  36
  37 #ifdef CONFIG_X86
  38 #include <asm/desc.h>
  39 #include <asm/ptrace.h>
  40 #include <asm/irq.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[row][col] = 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 /**
 348  * notify_remote_via_irq - send event to remote end of event channel via irq
 349  * @irq: irq of event channel to send event to
 350  *
 351  * Unlike notify_remote_via_evtchn(), this is safe to use across
 352  * save/restore. Notifications on a broken connection are silently
 353  * dropped.
 354  */
 355 void notify_remote_via_irq(int irq)
 356 {
 357         int evtchn = evtchn_from_irq(irq);
 358
 359         if (VALID_EVTCHN(evtchn))
 360                 notify_remote_via_evtchn(evtchn);
 361 }
 362 EXPORT_SYMBOL_GPL(notify_remote_via_irq);
 363
 364 static void xen_irq_init(unsigned irq)
 365 {
 366         struct irq_info *info;
 367 #ifdef CONFIG_SMP
 368         /* By default all event channels notify CPU#0. */
 369         cpumask_copy(irq_get_affinity_mask(irq), cpumask_of(0));
 370 #endif
 371
 372         info = kzalloc(sizeof(*info), GFP_KERNEL);
 373         if (info == NULL)
 374                 panic("Unable to allocate metadata for IRQ%d\n", irq);
 375
 376         info->type = IRQT_UNBOUND;
 377         info->refcnt = -1;
 378
 379         irq_set_handler_data(irq, info);
 380
 381         list_add_tail(&info->list, &xen_irq_list_head);
 382 }
 383
 384 static int __must_check xen_allocate_irqs_dynamic(int nvec)
 385 {
 386         int i, irq = irq_alloc_descs(-1, 0, nvec, -1);
 387
 388         if (irq >= 0) {
 389                 for (i = 0; i < nvec; i++)
 390                         xen_irq_init(irq + i);
 391         }
 392
 393         return irq;
 394 }
 395
 396 static inline int __must_check xen_allocate_irq_dynamic(void)
 397 {
 398
 399         return xen_allocate_irqs_dynamic(1);
 400 }
 401
 402 static int __must_check xen_allocate_irq_gsi(unsigned gsi)
 403 {
 404         int irq;
 405
 406         /*
 407          * A PV guest has no concept of a GSI (since it has no ACPI
 408          * nor access to/knowledge of the physical APICs). Therefore
 409          * all IRQs are dynamically allocated from the entire IRQ
 410          * space.
 411          */
 412         if (xen_pv_domain() && !xen_initial_domain())
 413                 return xen_allocate_irq_dynamic();
 414
 415         /* Legacy IRQ descriptors are already allocated by the arch. */
 416         if (gsi < nr_legacy_irqs())
 417                 irq = gsi;
 418         else
 419                 irq = irq_alloc_desc_at(gsi, -1);
 420
 421         xen_irq_init(irq);
 422
 423         return irq;
 424 }
 425
 426 static void xen_free_irq(unsigned irq)
 427 {
 428         struct irq_info *info = irq_get_handler_data(irq);
 429
 430         if (WARN_ON(!info))
 431                 return;
 432
 433         list_del(&info->list);
 434
 435         irq_set_handler_data(irq, NULL);
 436
 437         WARN_ON(info->refcnt > 0);
 438
 439         kfree(info);
 440
 441         /* Legacy IRQ descriptors are managed by the arch. */
 442         if (irq < nr_legacy_irqs())
 443                 return;
 444
 445         irq_free_desc(irq);
 446 }
 447
 448 static void xen_evtchn_close(unsigned int port)
 449 {
 450         struct evtchn_close close;
 451
 452         close.port = port;
 453         if (HYPERVISOR_event_channel_op(EVTCHNOP_close, &close) != 0)
 454                 BUG();
 455 }
 456
 457 static void pirq_query_unmask(int irq)
 458 {
 459         struct physdev_irq_status_query irq_status;
 460         struct irq_info *info = info_for_irq(irq);
 461
 462         BUG_ON(info->type != IRQT_PIRQ);
 463
 464         irq_status.irq = pirq_from_irq(irq);
 465         if (HYPERVISOR_physdev_op(PHYSDEVOP_irq_status_query, &irq_status))
 466                 irq_status.flags = 0;
 467
 468         info->u.pirq.flags &= ~PIRQ_NEEDS_EOI;
 469         if (irq_status.flags & XENIRQSTAT_needs_eoi)
 470                 info->u.pirq.flags |= PIRQ_NEEDS_EOI;
 471 }
 472
 473 static void eoi_pirq(struct irq_data *data)
 474 {
 475         int evtchn = evtchn_from_irq(data->irq);
 476         struct physdev_eoi eoi = { .irq = pirq_from_irq(data->irq) };
 477         int rc = 0;
 478
 479         if (!VALID_EVTCHN(evtchn))
 480                 return;
 481
 482         if (unlikely(irqd_is_setaffinity_pending(data)) &&
 483             likely(!irqd_irq_disabled(data))) {
 484                 int masked = test_and_set_mask(evtchn);
 485
 486                 clear_evtchn(evtchn);
 487
 488                 irq_move_masked_irq(data);
 489
 490                 if (!masked)
 491                         unmask_evtchn(evtchn);
 492         } else
 493                 clear_evtchn(evtchn);
 494
 495         if (pirq_needs_eoi(data->irq)) {
 496                 rc = HYPERVISOR_physdev_op(PHYSDEVOP_eoi, &eoi);
 497                 WARN_ON(rc);
 498         }
 499 }
 500
 501 static void mask_ack_pirq(struct irq_data *data)
 502 {
 503         disable_dynirq(data);
 504         eoi_pirq(data);
 505 }
 506
 507 static unsigned int __startup_pirq(unsigned int irq)
 508 {
 509         struct evtchn_bind_pirq bind_pirq;
 510         struct irq_info *info = info_for_irq(irq);
 511         int evtchn = evtchn_from_irq(irq);
 512         int rc;
 513
 514         BUG_ON(info->type != IRQT_PIRQ);
 515
 516         if (VALID_EVTCHN(evtchn))
 517                 goto out;
 518
 519         bind_pirq.pirq = pirq_from_irq(irq);
 520         /* NB. We are happy to share unless we are probing. */
 521         bind_pirq.flags = info->u.pirq.flags & PIRQ_SHAREABLE ?
 522                                         BIND_PIRQ__WILL_SHARE : 0;
 523         rc = HYPERVISOR_event_channel_op(EVTCHNOP_bind_pirq, &bind_pirq);
 524         if (rc != 0) {
 525                 pr_warn("Failed to obtain physical IRQ %d\n", irq);
 526                 return 0;
 527         }
 528         evtchn = bind_pirq.port;
 529
 530         pirq_query_unmask(irq);
 531
 532         rc = set_evtchn_to_irq(evtchn, irq);
 533         if (rc)
 534                 goto err;
 535
 536         info->evtchn = evtchn;
 537         bind_evtchn_to_cpu(evtchn, 0);
 538
 539         rc = xen_evtchn_port_setup(info);
 540         if (rc)
 541                 goto err;
 542
 543 out:
 544         unmask_evtchn(evtchn);
 545         eoi_pirq(irq_get_irq_data(irq));
 546
 547         return 0;
 548
 549 err:
 550         pr_err("irq%d: Failed to set port to irq mapping (%d)\n", irq, rc);
 551         xen_evtchn_close(evtchn);
 552         return 0;
 553 }
 554
 555 static unsigned int startup_pirq(struct irq_data *data)
 556 {
 557         return __startup_pirq(data->irq);
 558 }
 559
 560 static void shutdown_pirq(struct irq_data *data)
 561 {
 562         unsigned int irq = data->irq;
 563         struct irq_info *info = info_for_irq(irq);
 564         unsigned evtchn = evtchn_from_irq(irq);
 565
 566         BUG_ON(info->type != IRQT_PIRQ);
 567
 568         if (!VALID_EVTCHN(evtchn))
 569                 return;
 570
 571         mask_evtchn(evtchn);
 572         xen_evtchn_close(evtchn);
 573         xen_irq_info_cleanup(info);
 574 }
 575
 576 static void enable_pirq(struct irq_data *data)
 577 {
 578         enable_dynirq(data);
 579 }
 580
 581 static void disable_pirq(struct irq_data *data)
 582 {
 583         disable_dynirq(data);
 584 }
 585
 586 int xen_irq_from_gsi(unsigned gsi)
 587 {
 588         struct irq_info *info;
 589
 590         list_for_each_entry(info, &xen_irq_list_head, list) {
 591                 if (info->type != IRQT_PIRQ)
 592                         continue;
 593
 594                 if (info->u.pirq.gsi == gsi)
 595                         return info->irq;
 596         }
 597
 598         return -1;
 599 }
 600 EXPORT_SYMBOL_GPL(xen_irq_from_gsi);
 601
 602 static void __unbind_from_irq(unsigned int irq)
 603 {
 604         int evtchn = evtchn_from_irq(irq);
 605         struct irq_info *info = irq_get_handler_data(irq);
 606
 607         if (info->refcnt > 0) {
 608                 info->refcnt--;
 609                 if (info->refcnt != 0)
 610                         return;
 611         }
 612
 613         if (VALID_EVTCHN(evtchn)) {
 614                 unsigned int cpu = cpu_from_irq(irq);
 615
 616                 xen_evtchn_close(evtchn);
 617
 618                 switch (type_from_irq(irq)) {
 619                 case IRQT_VIRQ:
 620                         per_cpu(virq_to_irq, cpu)[virq_from_irq(irq)] = -1;
 621                         break;
 622                 case IRQT_IPI:
 623                         per_cpu(ipi_to_irq, cpu)[ipi_from_irq(irq)] = -1;
 624                         break;
 625                 default:
 626                         break;
 627                 }
 628
 629                 xen_irq_info_cleanup(info);
 630         }
 631
 632         xen_free_irq(irq);
 633 }
 634
 635 /*
 636  * Do not make any assumptions regarding the relationship between the
 637  * IRQ number returned here and the Xen pirq argument.
 638  *
 639  * Note: We don't assign an event channel until the irq actually started
 640  * up.  Return an existing irq if we've already got one for the gsi.
 641  *
 642  * Shareable implies level triggered, not shareable implies edge
 643  * triggered here.
 644  */
 645 int xen_bind_pirq_gsi_to_irq(unsigned gsi,
 646                              unsigned pirq, int shareable, char *name)
 647 {
 648         int irq = -1;
 649         struct physdev_irq irq_op;
 650         int ret;
 651
 652         mutex_lock(&irq_mapping_update_lock);
 653
 654         irq = xen_irq_from_gsi(gsi);
 655         if (irq != -1) {
 656                 pr_info("%s: returning irq %d for gsi %u\n",
 657                         __func__, irq, gsi);
 658                 goto out;
 659         }
 660
 661         irq = xen_allocate_irq_gsi(gsi);
 662         if (irq < 0)
 663                 goto out;
 664
 665         irq_op.irq = irq;
 666         irq_op.vector = 0;
 667
 668         /* Only the privileged domain can do this. For non-priv, the pcifront
 669          * driver provides a PCI bus that does the call to do exactly
 670          * this in the priv domain. */
 671         if (xen_initial_domain() &&
 672             HYPERVISOR_physdev_op(PHYSDEVOP_alloc_irq_vector, &irq_op)) {
 673                 xen_free_irq(irq);
 674                 irq = -ENOSPC;
 675                 goto out;
 676         }
 677
 678         ret = xen_irq_info_pirq_setup(irq, 0, pirq, gsi, DOMID_SELF,
 679                                shareable ? PIRQ_SHAREABLE : 0);
 680         if (ret < 0) {
 681                 __unbind_from_irq(irq);
 682                 irq = ret;
 683                 goto out;
 684         }
 685
 686         pirq_query_unmask(irq);
 687         /* We try to use the handler with the appropriate semantic for the
 688          * type of interrupt: if the interrupt is an edge triggered
 689          * interrupt we use handle_edge_irq.
 690          *
 691          * On the other hand if the interrupt is level triggered we use
 692          * handle_fasteoi_irq like the native code does for this kind of
 693          * interrupts.
 694          *
 695          * Depending on the Xen version, pirq_needs_eoi might return true
 696          * not only for level triggered interrupts but for edge triggered
 697          * interrupts too. In any case Xen always honors the eoi mechanism,
 698          * not injecting any more pirqs of the same kind if the first one
 699          * hasn't received an eoi yet. Therefore using the fasteoi handler
 700          * is the right choice either way.
 701          */
 702         if (shareable)
 703                 irq_set_chip_and_handler_name(irq, &xen_pirq_chip,
 704                                 handle_fasteoi_irq, name);
 705         else
 706                 irq_set_chip_and_handler_name(irq, &xen_pirq_chip,
 707                                 handle_edge_irq, name);
 708
 709 out:
 710         mutex_unlock(&irq_mapping_update_lock);
 711
 712         return irq;
 713 }
 714
 715 #ifdef CONFIG_PCI_MSI
 716 int xen_allocate_pirq_msi(struct pci_dev *dev, struct msi_desc *msidesc)
 717 {
 718         int rc;
 719         struct physdev_get_free_pirq op_get_free_pirq;
 720
 721         op_get_free_pirq.type = MAP_PIRQ_TYPE_MSI;
 722         rc = HYPERVISOR_physdev_op(PHYSDEVOP_get_free_pirq, &op_get_free_pirq);
 723
 724         WARN_ONCE(rc == -ENOSYS,
 725                   "hypervisor does not support the PHYSDEVOP_get_free_pirq interface\n");
 726
 727         return rc ? -1 : op_get_free_pirq.pirq;
 728 }
 729
 730 int xen_bind_pirq_msi_to_irq(struct pci_dev *dev, struct msi_desc *msidesc,
 731                              int pirq, int nvec, const char *name, domid_t domid)
 732 {
 733         int i, irq, ret;
 734
 735         mutex_lock(&irq_mapping_update_lock);
 736
 737         irq = xen_allocate_irqs_dynamic(nvec);
 738         if (irq < 0)
 739                 goto out;
 740
 741         for (i = 0; i < nvec; i++) {
 742                 irq_set_chip_and_handler_name(irq + i, &xen_pirq_chip, handle_edge_irq, name);
 743
 744                 ret = xen_irq_info_pirq_setup(irq + i, 0, pirq + i, 0, domid,
 745                                               i == 0 ? 0 : PIRQ_MSI_GROUP);
 746                 if (ret < 0)
 747                         goto error_irq;
 748         }
 749
 750         ret = irq_set_msi_desc(irq, msidesc);
 751         if (ret < 0)
 752                 goto error_irq;
 753 out:
 754         mutex_unlock(&irq_mapping_update_lock);
 755         return irq;
 756 error_irq:
 757         while (nvec--)
 758                 __unbind_from_irq(irq + nvec);
 759         mutex_unlock(&irq_mapping_update_lock);
 760         return ret;
 761 }
 762 #endif
 763
 764 int xen_destroy_irq(int irq)
 765 {
 766         struct physdev_unmap_pirq unmap_irq;
 767         struct irq_info *info = info_for_irq(irq);
 768         int rc = -ENOENT;
 769
 770         mutex_lock(&irq_mapping_update_lock);
 771
 772         /*
 773          * If trying to remove a vector in a MSI group different
 774          * than the first one skip the PIRQ unmap unless this vector
 775          * is the first one in the group.
 776          */
 777         if (xen_initial_domain() && !(info->u.pirq.flags & PIRQ_MSI_GROUP)) {
 778                 unmap_irq.pirq = info->u.pirq.pirq;
 779                 unmap_irq.domid = info->u.pirq.domid;
 780                 rc = HYPERVISOR_physdev_op(PHYSDEVOP_unmap_pirq, &unmap_irq);
 781                 /* If another domain quits without making the pci_disable_msix
 782                  * call, the Xen hypervisor takes care of freeing the PIRQs
 783                  * (free_domain_pirqs).
 784                  */
 785                 if ((rc == -ESRCH && info->u.pirq.domid != DOMID_SELF))
 786                         pr_info("domain %d does not have %d anymore\n",
 787                                 info->u.pirq.domid, info->u.pirq.pirq);
 788                 else if (rc) {
 789                         pr_warn("unmap irq failed %d\n", rc);
 790                         goto out;
 791                 }
 792         }
 793
 794         xen_free_irq(irq);
 795
 796 out:
 797         mutex_unlock(&irq_mapping_update_lock);
 798         return rc;
 799 }
 800
 801 int xen_irq_from_pirq(unsigned pirq)
 802 {
 803         int irq;
 804
 805         struct irq_info *info;
 806
 807         mutex_lock(&irq_mapping_update_lock);
 808
 809         list_for_each_entry(info, &xen_irq_list_head, list) {
 810                 if (info->type != IRQT_PIRQ)
 811                         continue;
 812                 irq = info->irq;
 813                 if (info->u.pirq.pirq == pirq)
 814                         goto out;
 815         }
 816         irq = -1;
 817 out:
 818         mutex_unlock(&irq_mapping_update_lock);
 819
 820         return irq;
 821 }
 822
 823
 824 int xen_pirq_from_irq(unsigned irq)
 825 {
 826         return pirq_from_irq(irq);
 827 }
 828 EXPORT_SYMBOL_GPL(xen_pirq_from_irq);
 829
 830 int bind_evtchn_to_irq(unsigned int evtchn)
 831 {
 832         int irq;
 833         int ret;
 834
 835         if (evtchn >= xen_evtchn_max_channels())
 836                 return -ENOMEM;
 837
 838         mutex_lock(&irq_mapping_update_lock);
 839
 840         irq = get_evtchn_to_irq(evtchn);
 841
 842         if (irq == -1) {
 843                 irq = xen_allocate_irq_dynamic();
 844                 if (irq < 0)
 845                         goto out;
 846
 847                 irq_set_chip_and_handler_name(irq, &xen_dynamic_chip,
 848                                               handle_edge_irq, "event");
 849
 850                 ret = xen_irq_info_evtchn_setup(irq, evtchn);
 851                 if (ret < 0) {
 852                         __unbind_from_irq(irq);
 853                         irq = ret;
 854                         goto out;
 855                 }
 856                 /* New interdomain events are bound to VCPU 0. */
 857                 bind_evtchn_to_cpu(evtchn, 0);
 858         } else {
 859                 struct irq_info *info = info_for_irq(irq);
 860                 WARN_ON(info == NULL || info->type != IRQT_EVTCHN);
 861         }
 862
 863 out:
 864         mutex_unlock(&irq_mapping_update_lock);
 865
 866         return irq;
 867 }
 868 EXPORT_SYMBOL_GPL(bind_evtchn_to_irq);
 869
 870 static int bind_ipi_to_irq(unsigned int ipi, unsigned int cpu)
 871 {
 872         struct evtchn_bind_ipi bind_ipi;
 873         int evtchn, irq;
 874         int ret;
 875
 876         mutex_lock(&irq_mapping_update_lock);
 877
 878         irq = per_cpu(ipi_to_irq, cpu)[ipi];
 879
 880         if (irq == -1) {
 881                 irq = xen_allocate_irq_dynamic();
 882                 if (irq < 0)
 883                         goto out;
 884
 885                 irq_set_chip_and_handler_name(irq, &xen_percpu_chip,
 886                                               handle_percpu_irq, "ipi");
 887
 888                 bind_ipi.vcpu = xen_vcpu_nr(cpu);
 889                 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_ipi,
 890                                                 &bind_ipi) != 0)
 891                         BUG();
 892                 evtchn = bind_ipi.port;
 893
 894                 ret = xen_irq_info_ipi_setup(cpu, irq, evtchn, ipi);
 895                 if (ret < 0) {
 896                         __unbind_from_irq(irq);
 897                         irq = ret;
 898                         goto out;
 899                 }
 900                 bind_evtchn_to_cpu(evtchn, cpu);
 901         } else {
 902                 struct irq_info *info = info_for_irq(irq);
 903                 WARN_ON(info == NULL || info->type != IRQT_IPI);
 904         }
 905
 906  out:
 907         mutex_unlock(&irq_mapping_update_lock);
 908         return irq;
 909 }
 910
 911 int bind_interdomain_evtchn_to_irq(unsigned int remote_domain,
 912                                    unsigned int remote_port)
 913 {
 914         struct evtchn_bind_interdomain bind_interdomain;
 915         int err;
 916
 917         bind_interdomain.remote_dom  = remote_domain;
 918         bind_interdomain.remote_port = remote_port;
 919
 920         err = HYPERVISOR_event_channel_op(EVTCHNOP_bind_interdomain,
 921                                           &bind_interdomain);
 922
 923         return err ? : bind_evtchn_to_irq(bind_interdomain.local_port);
 924 }
 925 EXPORT_SYMBOL_GPL(bind_interdomain_evtchn_to_irq);
 926
 927 static int find_virq(unsigned int virq, unsigned int cpu)
 928 {
 929         struct evtchn_status status;
 930         int port, rc = -ENOENT;
 931
 932         memset(&status, 0, sizeof(status));
 933         for (port = 0; port < xen_evtchn_max_channels(); port++) {
 934                 status.dom = DOMID_SELF;
 935                 status.port = port;
 936                 rc = HYPERVISOR_event_channel_op(EVTCHNOP_status, &status);
 937                 if (rc < 0)
 938                         continue;
 939                 if (status.status != EVTCHNSTAT_virq)
 940                         continue;
 941                 if (status.u.virq == virq && status.vcpu == xen_vcpu_nr(cpu)) {
 942                         rc = port;
 943                         break;
 944                 }
 945         }
 946         return rc;
 947 }
 948
 949 /**
 950  * xen_evtchn_nr_channels - number of usable event channel ports
 951  *
 952  * This may be less than the maximum supported by the current
 953  * hypervisor ABI. Use xen_evtchn_max_channels() for the maximum
 954  * supported.
 955  */
 956 unsigned xen_evtchn_nr_channels(void)
 957 {
 958         return evtchn_ops->nr_channels();
 959 }
 960 EXPORT_SYMBOL_GPL(xen_evtchn_nr_channels);
 961
 962 int bind_virq_to_irq(unsigned int virq, unsigned int cpu, bool percpu)
 963 {
 964         struct evtchn_bind_virq bind_virq;
 965         int evtchn, irq, ret;
 966
 967         mutex_lock(&irq_mapping_update_lock);
 968
 969         irq = per_cpu(virq_to_irq, cpu)[virq];
 970
 971         if (irq == -1) {
 972                 irq = xen_allocate_irq_dynamic();
 973                 if (irq < 0)
 974                         goto out;
 975
 976                 if (percpu)
 977                         irq_set_chip_and_handler_name(irq, &xen_percpu_chip,
 978                                                       handle_percpu_irq, "virq");
 979                 else
 980                         irq_set_chip_and_handler_name(irq, &xen_dynamic_chip,
 981                                                       handle_edge_irq, "virq");
 982
 983                 bind_virq.virq = virq;
 984                 bind_virq.vcpu = xen_vcpu_nr(cpu);
 985                 ret = HYPERVISOR_event_channel_op(EVTCHNOP_bind_virq,
 986                                                 &bind_virq);
 987                 if (ret == 0)
 988                         evtchn = bind_virq.port;
 989                 else {
 990                         if (ret == -EEXIST)
 991                                 ret = find_virq(virq, cpu);
 992                         BUG_ON(ret < 0);
 993                         evtchn = ret;
 994                 }
 995
 996                 ret = xen_irq_info_virq_setup(cpu, irq, evtchn, virq);
 997                 if (ret < 0) {
 998                         __unbind_from_irq(irq);
 999                         irq = ret;
1000                         goto out;
1001                 }
1002
1003                 bind_evtchn_to_cpu(evtchn, cpu);
1004         } else {
1005                 struct irq_info *info = info_for_irq(irq);
1006                 WARN_ON(info == NULL || info->type != IRQT_VIRQ);
1007         }
1008
1009 out:
1010         mutex_unlock(&irq_mapping_update_lock);
1011
1012         return irq;
1013 }
1014
1015 static void unbind_from_irq(unsigned int irq)
1016 {
1017         mutex_lock(&irq_mapping_update_lock);
1018         __unbind_from_irq(irq);
1019         mutex_unlock(&irq_mapping_update_lock);
1020 }
1021
1022 int bind_evtchn_to_irqhandler(unsigned int evtchn,
1023                               irq_handler_t handler,
1024                               unsigned long irqflags,
1025                               const char *devname, void *dev_id)
1026 {
1027         int irq, retval;
1028
1029         irq = bind_evtchn_to_irq(evtchn);
1030         if (irq < 0)
1031                 return irq;
1032         retval = request_irq(irq, handler, irqflags, devname, dev_id);
1033         if (retval != 0) {
1034                 unbind_from_irq(irq);
1035                 return retval;
1036         }
1037
1038         return irq;
1039 }
1040 EXPORT_SYMBOL_GPL(bind_evtchn_to_irqhandler);
1041
1042 int bind_interdomain_evtchn_to_irqhandler(unsigned int remote_domain,
1043                                           unsigned int remote_port,
1044                                           irq_handler_t handler,
1045                                           unsigned long irqflags,
1046                                           const char *devname,
1047                                           void *dev_id)
1048 {
1049         int irq, retval;
1050
1051         irq = bind_interdomain_evtchn_to_irq(remote_domain, remote_port);
1052         if (irq < 0)
1053                 return irq;
1054
1055         retval = request_irq(irq, handler, irqflags, devname, dev_id);
1056         if (retval != 0) {
1057                 unbind_from_irq(irq);
1058                 return retval;
1059         }
1060
1061         return irq;
1062 }
1063 EXPORT_SYMBOL_GPL(bind_interdomain_evtchn_to_irqhandler);
1064
1065 int bind_virq_to_irqhandler(unsigned int virq, unsigned int cpu,
1066                             irq_handler_t handler,
1067                             unsigned long irqflags, const char *devname, void *dev_id)
1068 {
1069         int irq, retval;
1070
1071         irq = bind_virq_to_irq(virq, cpu, irqflags & IRQF_PERCPU);
1072         if (irq < 0)
1073                 return irq;
1074         retval = request_irq(irq, handler, irqflags, devname, dev_id);
1075         if (retval != 0) {
1076                 unbind_from_irq(irq);
1077                 return retval;
1078         }
1079
1080         return irq;
1081 }
1082 EXPORT_SYMBOL_GPL(bind_virq_to_irqhandler);
1083
1084 int bind_ipi_to_irqhandler(enum ipi_vector ipi,
1085                            unsigned int cpu,
1086                            irq_handler_t handler,
1087                            unsigned long irqflags,
1088                            const char *devname,
1089                            void *dev_id)
1090 {
1091         int irq, retval;
1092
1093         irq = bind_ipi_to_irq(ipi, cpu);
1094         if (irq < 0)
1095                 return irq;
1096
1097         irqflags |= IRQF_NO_SUSPEND | IRQF_FORCE_RESUME | IRQF_EARLY_RESUME;
1098         retval = request_irq(irq, handler, irqflags, devname, dev_id);
1099         if (retval != 0) {
1100                 unbind_from_irq(irq);
1101                 return retval;
1102         }
1103
1104         return irq;
1105 }
1106
1107 void unbind_from_irqhandler(unsigned int irq, void *dev_id)
1108 {
1109         struct irq_info *info = irq_get_handler_data(irq);
1110
1111         if (WARN_ON(!info))
1112                 return;
1113         free_irq(irq, dev_id);
1114         unbind_from_irq(irq);
1115 }
1116 EXPORT_SYMBOL_GPL(unbind_from_irqhandler);
1117
1118 /**
1119  * xen_set_irq_priority() - set an event channel priority.
1120  * @irq:irq bound to an event channel.
1121  * @priority: priority between XEN_IRQ_PRIORITY_MAX and XEN_IRQ_PRIORITY_MIN.
1122  */
1123 int xen_set_irq_priority(unsigned irq, unsigned priority)
1124 {
1125         struct evtchn_set_priority set_priority;
1126
1127         set_priority.port = evtchn_from_irq(irq);
1128         set_priority.priority = priority;
1129
1130         return HYPERVISOR_event_channel_op(EVTCHNOP_set_priority,
1131                                            &set_priority);
1132 }
1133 EXPORT_SYMBOL_GPL(xen_set_irq_priority);
1134
1135 int evtchn_make_refcounted(unsigned int evtchn)
1136 {
1137         int irq = get_evtchn_to_irq(evtchn);
1138         struct irq_info *info;
1139
1140         if (irq == -1)
1141                 return -ENOENT;
1142
1143         info = irq_get_handler_data(irq);
1144
1145         if (!info)
1146                 return -ENOENT;
1147
1148         WARN_ON(info->refcnt != -1);
1149
1150         info->refcnt = 1;
1151
1152         return 0;
1153 }
1154 EXPORT_SYMBOL_GPL(evtchn_make_refcounted);
1155
1156 int evtchn_get(unsigned int evtchn)
1157 {
1158         int irq;
1159         struct irq_info *info;
1160         int err = -ENOENT;
1161
1162         if (evtchn >= xen_evtchn_max_channels())
1163                 return -EINVAL;
1164
1165         mutex_lock(&irq_mapping_update_lock);
1166
1167         irq = get_evtchn_to_irq(evtchn);
1168         if (irq == -1)
1169                 goto done;
1170
1171         info = irq_get_handler_data(irq);
1172
1173         if (!info)
1174                 goto done;
1175
1176         err = -EINVAL;
1177         if (info->refcnt <= 0)
1178                 goto done;
1179
1180         info->refcnt++;
1181         err = 0;
1182  done:
1183         mutex_unlock(&irq_mapping_update_lock);
1184
1185         return err;
1186 }
1187 EXPORT_SYMBOL_GPL(evtchn_get);
1188
1189 void evtchn_put(unsigned int evtchn)
1190 {
1191         int irq = get_evtchn_to_irq(evtchn);
1192         if (WARN_ON(irq == -1))
1193                 return;
1194         unbind_from_irq(irq);
1195 }
1196 EXPORT_SYMBOL_GPL(evtchn_put);
1197
1198 void xen_send_IPI_one(unsigned int cpu, enum ipi_vector vector)
1199 {
1200         int irq;
1201
1202 #ifdef CONFIG_X86
1203         if (unlikely(vector == XEN_NMI_VECTOR)) {
1204                 int rc =  HYPERVISOR_vcpu_op(VCPUOP_send_nmi, xen_vcpu_nr(cpu),
1205                                              NULL);
1206                 if (rc < 0)
1207                         printk(KERN_WARNING "Sending nmi to CPU%d failed (rc:%d)\n", cpu, rc);
1208                 return;
1209         }
1210 #endif
1211         irq = per_cpu(ipi_to_irq, cpu)[vector];
1212         BUG_ON(irq < 0);
1213         notify_remote_via_irq(irq);
1214 }
1215
1216 static DEFINE_PER_CPU(unsigned, xed_nesting_count);
1217
1218 static void __xen_evtchn_do_upcall(void)
1219 {
1220         struct vcpu_info *vcpu_info = __this_cpu_read(xen_vcpu);
1221         int cpu = get_cpu();
1222         unsigned count;
1223
1224         do {
1225                 vcpu_info->evtchn_upcall_pending = 0;
1226
1227                 if (__this_cpu_inc_return(xed_nesting_count) - 1)
1228                         goto out;
1229
1230                 xen_evtchn_handle_events(cpu);
1231
1232                 BUG_ON(!irqs_disabled());
1233
1234                 count = __this_cpu_read(xed_nesting_count);
1235                 __this_cpu_write(xed_nesting_count, 0);
1236         } while (count != 1 || vcpu_info->evtchn_upcall_pending);
1237
1238 out:
1239
1240         put_cpu();
1241 }
1242
1243 void xen_evtchn_do_upcall(struct pt_regs *regs)
1244 {
1245         struct pt_regs *old_regs = set_irq_regs(regs);
1246
1247         irq_enter();
1248 #ifdef CONFIG_X86
1249         inc_irq_stat(irq_hv_callback_count);
1250 #endif
1251
1252         __xen_evtchn_do_upcall();
1253
1254         irq_exit();
1255         set_irq_regs(old_regs);
1256 }
1257
1258 void xen_hvm_evtchn_do_upcall(void)
1259 {
1260         __xen_evtchn_do_upcall();
1261 }
1262 EXPORT_SYMBOL_GPL(xen_hvm_evtchn_do_upcall);
1263
1264 /* Rebind a new event channel to an existing irq. */
1265 void rebind_evtchn_irq(int evtchn, int irq)
1266 {
1267         struct irq_info *info = info_for_irq(irq);
1268
1269         if (WARN_ON(!info))
1270                 return;
1271
1272         /* Make sure the irq is masked, since the new event channel
1273            will also be masked. */
1274         disable_irq(irq);
1275
1276         mutex_lock(&irq_mapping_update_lock);
1277
1278         /* After resume the irq<->evtchn mappings are all cleared out */
1279         BUG_ON(get_evtchn_to_irq(evtchn) != -1);
1280         /* Expect irq to have been bound before,
1281            so there should be a proper type */
1282         BUG_ON(info->type == IRQT_UNBOUND);
1283
1284         (void)xen_irq_info_evtchn_setup(irq, evtchn);
1285
1286         mutex_unlock(&irq_mapping_update_lock);
1287
1288         bind_evtchn_to_cpu(evtchn, info->cpu);
1289         /* This will be deferred until interrupt is processed */
1290         irq_set_affinity(irq, cpumask_of(info->cpu));
1291
1292         /* Unmask the event channel. */
1293         enable_irq(irq);
1294 }
1295
1296 /* Rebind an evtchn so that it gets delivered to a specific cpu */
1297 static int xen_rebind_evtchn_to_cpu(int evtchn, unsigned int tcpu)
1298 {
1299         struct evtchn_bind_vcpu bind_vcpu;
1300         int masked;
1301
1302         if (!VALID_EVTCHN(evtchn))
1303                 return -1;
1304
1305         if (!xen_support_evtchn_rebind())
1306                 return -1;
1307
1308         /* Send future instances of this interrupt to other vcpu. */
1309         bind_vcpu.port = evtchn;
1310         bind_vcpu.vcpu = xen_vcpu_nr(tcpu);
1311
1312         /*
1313          * Mask the event while changing the VCPU binding to prevent
1314          * it being delivered on an unexpected VCPU.
1315          */
1316         masked = test_and_set_mask(evtchn);
1317
1318         /*
1319          * If this fails, it usually just indicates that we're dealing with a
1320          * virq or IPI channel, which don't actually need to be rebound. Ignore
1321          * it, but don't do the xenlinux-level rebind in that case.
1322          */
1323         if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_vcpu, &bind_vcpu) >= 0)
1324                 bind_evtchn_to_cpu(evtchn, tcpu);
1325
1326         if (!masked)
1327                 unmask_evtchn(evtchn);
1328
1329         return 0;
1330 }
1331
1332 static int set_affinity_irq(struct irq_data *data, const struct cpumask *dest,
1333                             bool force)
1334 {
1335         unsigned tcpu = cpumask_first_and(dest, cpu_online_mask);
1336         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 /* To be called with desc->lock held. */
1345 int xen_set_affinity_evtchn(struct irq_desc *desc, unsigned int tcpu)
1346 {
1347         struct irq_data *d = irq_desc_get_irq_data(desc);
1348
1349         return set_affinity_irq(d, cpumask_of(tcpu), false);
1350 }
1351 EXPORT_SYMBOL_GPL(xen_set_affinity_evtchn);
1352
1353 static void enable_dynirq(struct irq_data *data)
1354 {
1355         int evtchn = evtchn_from_irq(data->irq);
1356
1357         if (VALID_EVTCHN(evtchn))
1358                 unmask_evtchn(evtchn);
1359 }
1360
1361 static void disable_dynirq(struct irq_data *data)
1362 {
1363         int evtchn = evtchn_from_irq(data->irq);
1364
1365         if (VALID_EVTCHN(evtchn))
1366                 mask_evtchn(evtchn);
1367 }
1368
1369 static void ack_dynirq(struct irq_data *data)
1370 {
1371         int evtchn = evtchn_from_irq(data->irq);
1372
1373         if (!VALID_EVTCHN(evtchn))
1374                 return;
1375
1376         if (unlikely(irqd_is_setaffinity_pending(data)) &&
1377             likely(!irqd_irq_disabled(data))) {
1378                 int masked = test_and_set_mask(evtchn);
1379
1380                 clear_evtchn(evtchn);
1381
1382                 irq_move_masked_irq(data);
1383
1384                 if (!masked)
1385                         unmask_evtchn(evtchn);
1386         } else
1387                 clear_evtchn(evtchn);
1388 }
1389
1390 static void mask_ack_dynirq(struct irq_data *data)
1391 {
1392         disable_dynirq(data);
1393         ack_dynirq(data);
1394 }
1395
1396 static int retrigger_dynirq(struct irq_data *data)
1397 {
1398         unsigned int evtchn = evtchn_from_irq(data->irq);
1399         int masked;
1400
1401         if (!VALID_EVTCHN(evtchn))
1402                 return 0;
1403
1404         masked = test_and_set_mask(evtchn);
1405         set_evtchn(evtchn);
1406         if (!masked)
1407                 unmask_evtchn(evtchn);
1408
1409         return 1;
1410 }
1411
1412 static void restore_pirqs(void)
1413 {
1414         int pirq, rc, irq, gsi;
1415         struct physdev_map_pirq map_irq;
1416         struct irq_info *info;
1417
1418         list_for_each_entry(info, &xen_irq_list_head, list) {
1419                 if (info->type != IRQT_PIRQ)
1420                         continue;
1421
1422                 pirq = info->u.pirq.pirq;
1423                 gsi = info->u.pirq.gsi;
1424                 irq = info->irq;
1425
1426                 /* save/restore of PT devices doesn't work, so at this point the
1427                  * only devices present are GSI based emulated devices */
1428                 if (!gsi)
1429                         continue;
1430
1431                 map_irq.domid = DOMID_SELF;
1432                 map_irq.type = MAP_PIRQ_TYPE_GSI;
1433                 map_irq.index = gsi;
1434                 map_irq.pirq = pirq;
1435
1436                 rc = HYPERVISOR_physdev_op(PHYSDEVOP_map_pirq, &map_irq);
1437                 if (rc) {
1438                         pr_warn("xen map irq failed gsi=%d irq=%d pirq=%d rc=%d\n",
1439                                 gsi, irq, pirq, rc);
1440                         xen_free_irq(irq);
1441                         continue;
1442                 }
1443
1444                 printk(KERN_DEBUG "xen: --> irq=%d, pirq=%d\n", irq, map_irq.pirq);
1445
1446                 __startup_pirq(irq);
1447         }
1448 }
1449
1450 static void restore_cpu_virqs(unsigned int cpu)
1451 {
1452         struct evtchn_bind_virq bind_virq;
1453         int virq, irq, evtchn;
1454
1455         for (virq = 0; virq < NR_VIRQS; virq++) {
1456                 if ((irq = per_cpu(virq_to_irq, cpu)[virq]) == -1)
1457                         continue;
1458
1459                 BUG_ON(virq_from_irq(irq) != virq);
1460
1461                 /* Get a new binding from Xen. */
1462                 bind_virq.virq = virq;
1463                 bind_virq.vcpu = xen_vcpu_nr(cpu);
1464                 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_virq,
1465                                                 &bind_virq) != 0)
1466                         BUG();
1467                 evtchn = bind_virq.port;
1468
1469                 /* Record the new mapping. */
1470                 (void)xen_irq_info_virq_setup(cpu, irq, evtchn, virq);
1471                 bind_evtchn_to_cpu(evtchn, cpu);
1472         }
1473 }
1474
1475 static void restore_cpu_ipis(unsigned int cpu)
1476 {
1477         struct evtchn_bind_ipi bind_ipi;
1478         int ipi, irq, evtchn;
1479
1480         for (ipi = 0; ipi < XEN_NR_IPIS; ipi++) {
1481                 if ((irq = per_cpu(ipi_to_irq, cpu)[ipi]) == -1)
1482                         continue;
1483
1484                 BUG_ON(ipi_from_irq(irq) != ipi);
1485
1486                 /* Get a new binding from Xen. */
1487                 bind_ipi.vcpu = xen_vcpu_nr(cpu);
1488                 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_ipi,
1489                                                 &bind_ipi) != 0)
1490                         BUG();
1491                 evtchn = bind_ipi.port;
1492
1493                 /* Record the new mapping. */
1494                 (void)xen_irq_info_ipi_setup(cpu, irq, evtchn, ipi);
1495                 bind_evtchn_to_cpu(evtchn, cpu);
1496         }
1497 }
1498
1499 /* Clear an irq's pending state, in preparation for polling on it */
1500 void xen_clear_irq_pending(int irq)
1501 {
1502         int evtchn = evtchn_from_irq(irq);
1503
1504         if (VALID_EVTCHN(evtchn))
1505                 clear_evtchn(evtchn);
1506 }
1507 EXPORT_SYMBOL(xen_clear_irq_pending);
1508 void xen_set_irq_pending(int irq)
1509 {
1510         int evtchn = evtchn_from_irq(irq);
1511
1512         if (VALID_EVTCHN(evtchn))
1513                 set_evtchn(evtchn);
1514 }
1515
1516 bool xen_test_irq_pending(int irq)
1517 {
1518         int evtchn = evtchn_from_irq(irq);
1519         bool ret = false;
1520
1521         if (VALID_EVTCHN(evtchn))
1522                 ret = test_evtchn(evtchn);
1523
1524         return ret;
1525 }
1526
1527 /* Poll waiting for an irq to become pending with timeout.  In the usual case,
1528  * the irq will be disabled so it won't deliver an interrupt. */
1529 void xen_poll_irq_timeout(int irq, u64 timeout)
1530 {
1531         evtchn_port_t evtchn = evtchn_from_irq(irq);
1532
1533         if (VALID_EVTCHN(evtchn)) {
1534                 struct sched_poll poll;
1535
1536                 poll.nr_ports = 1;
1537                 poll.timeout = timeout;
1538                 set_xen_guest_handle(poll.ports, &evtchn);
1539
1540                 if (HYPERVISOR_sched_op(SCHEDOP_poll, &poll) != 0)
1541                         BUG();
1542         }
1543 }
1544 EXPORT_SYMBOL(xen_poll_irq_timeout);
1545 /* Poll waiting for an irq to become pending.  In the usual case, the
1546  * irq will be disabled so it won't deliver an interrupt. */
1547 void xen_poll_irq(int irq)
1548 {
1549         xen_poll_irq_timeout(irq, 0 /* no timeout */);
1550 }
1551
1552 /* Check whether the IRQ line is shared with other guests. */
1553 int xen_test_irq_shared(int irq)
1554 {
1555         struct irq_info *info = info_for_irq(irq);
1556         struct physdev_irq_status_query irq_status;
1557
1558         if (WARN_ON(!info))
1559                 return -ENOENT;
1560
1561         irq_status.irq = info->u.pirq.pirq;
1562
1563         if (HYPERVISOR_physdev_op(PHYSDEVOP_irq_status_query, &irq_status))
1564                 return 0;
1565         return !(irq_status.flags & XENIRQSTAT_shared);
1566 }
1567 EXPORT_SYMBOL_GPL(xen_test_irq_shared);
1568
1569 void xen_irq_resume(void)
1570 {
1571         unsigned int cpu;
1572         struct irq_info *info;
1573
1574         /* New event-channel space is not 'live' yet. */
1575         xen_evtchn_resume();
1576
1577         /* No IRQ <-> event-channel mappings. */
1578         list_for_each_entry(info, &xen_irq_list_head, list)
1579                 info->evtchn = 0; /* zap event-channel binding */
1580
1581         clear_evtchn_to_irq_all();
1582
1583         for_each_possible_cpu(cpu) {
1584                 restore_cpu_virqs(cpu);
1585                 restore_cpu_ipis(cpu);
1586         }
1587
1588         restore_pirqs();
1589 }
1590
1591 static struct irq_chip xen_dynamic_chip __read_mostly = {
1592         .name                   = "xen-dyn",
1593
1594         .irq_disable            = disable_dynirq,
1595         .irq_mask               = disable_dynirq,
1596         .irq_unmask             = enable_dynirq,
1597
1598         .irq_ack                = ack_dynirq,
1599         .irq_mask_ack           = mask_ack_dynirq,
1600
1601         .irq_set_affinity       = set_affinity_irq,
1602         .irq_retrigger          = retrigger_dynirq,
1603 };
1604
1605 static struct irq_chip xen_pirq_chip __read_mostly = {
1606         .name                   = "xen-pirq",
1607
1608         .irq_startup            = startup_pirq,
1609         .irq_shutdown           = shutdown_pirq,
1610         .irq_enable             = enable_pirq,
1611         .irq_disable            = disable_pirq,
1612
1613         .irq_mask               = disable_dynirq,
1614         .irq_unmask             = enable_dynirq,
1615
1616         .irq_ack                = eoi_pirq,
1617         .irq_eoi                = eoi_pirq,
1618         .irq_mask_ack           = mask_ack_pirq,
1619
1620         .irq_set_affinity       = set_affinity_irq,
1621
1622         .irq_retrigger          = retrigger_dynirq,
1623 };
1624
1625 static struct irq_chip xen_percpu_chip __read_mostly = {
1626         .name                   = "xen-percpu",
1627
1628         .irq_disable            = disable_dynirq,
1629         .irq_mask               = disable_dynirq,
1630         .irq_unmask             = enable_dynirq,
1631
1632         .irq_ack                = ack_dynirq,
1633 };
1634
1635 int xen_set_callback_via(uint64_t via)
1636 {
1637         struct xen_hvm_param a;
1638         a.domid = DOMID_SELF;
1639         a.index = HVM_PARAM_CALLBACK_IRQ;
1640         a.value = via;
1641         return HYPERVISOR_hvm_op(HVMOP_set_param, &a);
1642 }
1643 EXPORT_SYMBOL_GPL(xen_set_callback_via);
1644
1645 #ifdef CONFIG_XEN_PVHVM
1646 /* Vector callbacks are better than PCI interrupts to receive event
1647  * channel notifications because we can receive vector callbacks on any
1648  * vcpu and we don't need PCI support or APIC interactions. */
1649 void xen_callback_vector(void)
1650 {
1651         int rc;
1652         uint64_t callback_via;
1653
1654         if (xen_have_vector_callback) {
1655                 callback_via = HVM_CALLBACK_VECTOR(HYPERVISOR_CALLBACK_VECTOR);
1656                 rc = xen_set_callback_via(callback_via);
1657                 if (rc) {
1658                         pr_err("Request for Xen HVM callback vector failed\n");
1659                         xen_have_vector_callback = 0;
1660                         return;
1661                 }
1662                 pr_info_once("Xen HVM callback vector for event delivery is enabled\n");
1663                 alloc_intr_gate(HYPERVISOR_CALLBACK_VECTOR,
1664                                 xen_hvm_callback_vector);
1665         }
1666 }
1667 #else
1668 void xen_callback_vector(void) {}
1669 #endif
1670
1671 #undef MODULE_PARAM_PREFIX
1672 #define MODULE_PARAM_PREFIX "xen."
1673
1674 static bool fifo_events = true;
1675 module_param(fifo_events, bool, 0);
1676
1677 void __init xen_init_IRQ(void)
1678 {
1679         int ret = -EINVAL;
1680         unsigned int evtchn;
1681
1682         if (fifo_events)
1683                 ret = xen_evtchn_fifo_init();
1684         if (ret < 0)
1685                 xen_evtchn_2l_init();
1686
1687         evtchn_to_irq = kcalloc(EVTCHN_ROW(xen_evtchn_max_channels()),
1688                                 sizeof(*evtchn_to_irq), GFP_KERNEL);
1689         BUG_ON(!evtchn_to_irq);
1690
1691         /* No event channels are 'live' right now. */
1692         for (evtchn = 0; evtchn < xen_evtchn_nr_channels(); evtchn++)
1693                 mask_evtchn(evtchn);
1694
1695         pirq_needs_eoi = pirq_needs_eoi_flag;
1696
1697 #ifdef CONFIG_X86
1698         if (xen_pv_domain()) {
1699                 if (xen_initial_domain())
1700                         pci_xen_initial_domain();
1701         }
1702         if (xen_feature(XENFEAT_hvm_callback_vector))
1703                 xen_callback_vector();
1704
1705         if (xen_hvm_domain()) {
1706                 native_init_IRQ();
1707                 /* pci_xen_hvm_init must be called after native_init_IRQ so that
1708                  * __acpi_register_gsi can point at the right function */
1709                 pci_xen_hvm_init();
1710         } else {
1711                 int rc;
1712                 struct physdev_pirq_eoi_gmfn eoi_gmfn;
1713
1714                 pirq_eoi_map = (void *)__get_free_page(GFP_KERNEL|__GFP_ZERO);
1715                 eoi_gmfn.gmfn = virt_to_gfn(pirq_eoi_map);
1716                 rc = HYPERVISOR_physdev_op(PHYSDEVOP_pirq_eoi_gmfn_v2, &eoi_gmfn);
1717                 if (rc != 0) {
1718                         free_page((unsigned long) pirq_eoi_map);
1719                         pirq_eoi_map = NULL;
1720                 } else
1721                         pirq_needs_eoi = pirq_check_eoi_map;
1722         }
1723 #endif
1724 }