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 #include <linux/spinlock.h>
  36 #include <linux/cpuhotplug.h>
  37 #include <linux/atomic.h>
  38 #include <linux/ktime.h>
  39
  40 #ifdef CONFIG_X86
  41 #include <asm/desc.h>
  42 #include <asm/ptrace.h>
  43 #include <asm/irq.h>
  44 #include <asm/io_apic.h>
  45 #include <asm/i8259.h>
  46 #include <asm/xen/pci.h>
  47 #endif
  48 #include <asm/sync_bitops.h>
  49 #include <asm/xen/hypercall.h>
  50 #include <asm/xen/hypervisor.h>
  51 #include <xen/page.h>
  52
  53 #include <xen/xen.h>
  54 #include <xen/hvm.h>
  55 #include <xen/xen-ops.h>
  56 #include <xen/events.h>
  57 #include <xen/interface/xen.h>
  58 #include <xen/interface/event_channel.h>
  59 #include <xen/interface/hvm/hvm_op.h>
  60 #include <xen/interface/hvm/params.h>
  61 #include <xen/interface/physdev.h>
  62 #include <xen/interface/sched.h>
  63 #include <xen/interface/vcpu.h>
  64 #include <asm/hw_irq.h>
  65
  66 #include "events_internal.h"
  67
  68 #undef MODULE_PARAM_PREFIX
  69 #define MODULE_PARAM_PREFIX "xen."
  70
  71 static uint __read_mostly event_loop_timeout = 2;
  72 module_param(event_loop_timeout, uint, 0644);
  73
  74 static uint __read_mostly event_eoi_delay = 10;
  75 module_param(event_eoi_delay, uint, 0644);
  76
  77 const struct evtchn_ops *evtchn_ops;
  78
  79 /*
  80  * This lock protects updates to the following mapping and reference-count
  81  * arrays. The lock does not need to be acquired to read the mapping tables.
  82  */
  83 static DEFINE_MUTEX(irq_mapping_update_lock);
  84
  85 /*
  86  * Lock protecting event handling loop against removing event channels.
  87  * Adding of event channels is no issue as the associated IRQ becomes active
  88  * only after everything is setup (before request_[threaded_]irq() the handler
  89  * can't be entered for an event, as the event channel will be unmasked only
  90  * then).
  91  */
  92 static DEFINE_RWLOCK(evtchn_rwlock);
  93
  94 /*
  95  * Lock hierarchy:
  96  *
  97  * irq_mapping_update_lock
  98  *   evtchn_rwlock
  99  *     IRQ-desc lock
 100  *       percpu eoi_list_lock
 101  *         irq_info->lock
 102  */
 103
 104 static LIST_HEAD(xen_irq_list_head);
 105
 106 /* IRQ <-> VIRQ mapping. */
 107 static DEFINE_PER_CPU(int [NR_VIRQS], virq_to_irq) = {[0 ... NR_VIRQS-1] = -1};
 108
 109 /* IRQ <-> IPI mapping */
 110 static DEFINE_PER_CPU(int [XEN_NR_IPIS], ipi_to_irq) = {[0 ... XEN_NR_IPIS-1] = -1};
 111
 112 int **evtchn_to_irq;
 113 #ifdef CONFIG_X86
 114 static unsigned long *pirq_eoi_map;
 115 #endif
 116 static bool (*pirq_needs_eoi)(unsigned irq);
 117
 118 #define EVTCHN_ROW(e)  (e / (PAGE_SIZE/sizeof(**evtchn_to_irq)))
 119 #define EVTCHN_COL(e)  (e % (PAGE_SIZE/sizeof(**evtchn_to_irq)))
 120 #define EVTCHN_PER_ROW (PAGE_SIZE / sizeof(**evtchn_to_irq))
 121
 122 /* Xen will never allocate port zero for any purpose. */
 123 #define VALID_EVTCHN(chn)       ((chn) != 0)
 124
 125 static struct irq_info *legacy_info_ptrs[NR_IRQS_LEGACY];
 126
 127 static struct irq_chip xen_dynamic_chip;
 128 static struct irq_chip xen_lateeoi_chip;
 129 static struct irq_chip xen_percpu_chip;
 130 static struct irq_chip xen_pirq_chip;
 131 static void enable_dynirq(struct irq_data *data);
 132 static void disable_dynirq(struct irq_data *data);
 133
 134 static DEFINE_PER_CPU(unsigned int, irq_epoch);
 135
 136 static void clear_evtchn_to_irq_row(int *evtchn_row)
 137 {
 138         unsigned col;
 139
 140         for (col = 0; col < EVTCHN_PER_ROW; col++)
 141                 WRITE_ONCE(evtchn_row[col], -1);
 142 }
 143
 144 static void clear_evtchn_to_irq_all(void)
 145 {
 146         unsigned row;
 147
 148         for (row = 0; row < EVTCHN_ROW(xen_evtchn_max_channels()); row++) {
 149                 if (evtchn_to_irq[row] == NULL)
 150                         continue;
 151                 clear_evtchn_to_irq_row(evtchn_to_irq[row]);
 152         }
 153 }
 154
 155 static int set_evtchn_to_irq(unsigned evtchn, unsigned irq)
 156 {
 157         unsigned row;
 158         unsigned col;
 159         int *evtchn_row;
 160
 161         if (evtchn >= xen_evtchn_max_channels())
 162                 return -EINVAL;
 163
 164         row = EVTCHN_ROW(evtchn);
 165         col = EVTCHN_COL(evtchn);
 166
 167         if (evtchn_to_irq[row] == NULL) {
 168                 /* Unallocated irq entries return -1 anyway */
 169                 if (irq == -1)
 170                         return 0;
 171
 172                 evtchn_row = (int *) __get_free_pages(GFP_KERNEL, 0);
 173                 if (evtchn_row == NULL)
 174                         return -ENOMEM;
 175
 176                 clear_evtchn_to_irq_row(evtchn_row);
 177
 178                 /*
 179                  * We've prepared an empty row for the mapping. If a different
 180                  * thread was faster inserting it, we can drop ours.
 181                  */
 182                 if (cmpxchg(&evtchn_to_irq[row], NULL, evtchn_row) != NULL)
 183                         free_page((unsigned long) evtchn_row);
 184         }
 185
 186         WRITE_ONCE(evtchn_to_irq[row][col], irq);
 187         return 0;
 188 }
 189
 190 int get_evtchn_to_irq(unsigned evtchn)
 191 {
 192         if (evtchn >= xen_evtchn_max_channels())
 193                 return -1;
 194         if (evtchn_to_irq[EVTCHN_ROW(evtchn)] == NULL)
 195                 return -1;
 196         return READ_ONCE(evtchn_to_irq[EVTCHN_ROW(evtchn)][EVTCHN_COL(evtchn)]);
 197 }
 198
 199 /* Get info for IRQ */
 200 struct irq_info *info_for_irq(unsigned irq)
 201 {
 202         if (irq < nr_legacy_irqs())
 203                 return legacy_info_ptrs[irq];
 204         else
 205                 return irq_get_chip_data(irq);
 206 }
 207
 208 static void set_info_for_irq(unsigned int irq, struct irq_info *info)
 209 {
 210         if (irq < nr_legacy_irqs())
 211                 legacy_info_ptrs[irq] = info;
 212         else
 213                 irq_set_chip_data(irq, info);
 214 }
 215
 216 /* Constructors for packed IRQ information. */
 217 static int xen_irq_info_common_setup(struct irq_info *info,
 218                                      unsigned irq,
 219                                      enum xen_irq_type type,
 220                                      unsigned evtchn,
 221                                      unsigned short cpu)
 222 {
 223         int ret;
 224
 225         BUG_ON(info->type != IRQT_UNBOUND && info->type != type);
 226
 227         info->type = type;
 228         info->irq = irq;
 229         info->evtchn = evtchn;
 230         info->cpu = cpu;
 231         info->mask_reason = EVT_MASK_REASON_EXPLICIT;
 232         raw_spin_lock_init(&info->lock);
 233
 234         ret = set_evtchn_to_irq(evtchn, irq);
 235         if (ret < 0)
 236                 return ret;
 237
 238         irq_clear_status_flags(irq, IRQ_NOREQUEST|IRQ_NOAUTOEN);
 239
 240         return xen_evtchn_port_setup(info);
 241 }
 242
 243 static int xen_irq_info_evtchn_setup(unsigned irq,
 244                                      unsigned evtchn)
 245 {
 246         struct irq_info *info = info_for_irq(irq);
 247
 248         return xen_irq_info_common_setup(info, irq, IRQT_EVTCHN, evtchn, 0);
 249 }
 250
 251 static int xen_irq_info_ipi_setup(unsigned cpu,
 252                                   unsigned irq,
 253                                   unsigned evtchn,
 254                                   enum ipi_vector ipi)
 255 {
 256         struct irq_info *info = info_for_irq(irq);
 257
 258         info->u.ipi = ipi;
 259
 260         per_cpu(ipi_to_irq, cpu)[ipi] = irq;
 261
 262         return xen_irq_info_common_setup(info, irq, IRQT_IPI, evtchn, 0);
 263 }
 264
 265 static int xen_irq_info_virq_setup(unsigned cpu,
 266                                    unsigned irq,
 267                                    unsigned evtchn,
 268                                    unsigned virq)
 269 {
 270         struct irq_info *info = info_for_irq(irq);
 271
 272         info->u.virq = virq;
 273
 274         per_cpu(virq_to_irq, cpu)[virq] = irq;
 275
 276         return xen_irq_info_common_setup(info, irq, IRQT_VIRQ, evtchn, 0);
 277 }
 278
 279 static int xen_irq_info_pirq_setup(unsigned irq,
 280                                    unsigned evtchn,
 281                                    unsigned pirq,
 282                                    unsigned gsi,
 283                                    uint16_t domid,
 284                                    unsigned char flags)
 285 {
 286         struct irq_info *info = info_for_irq(irq);
 287
 288         info->u.pirq.pirq = pirq;
 289         info->u.pirq.gsi = gsi;
 290         info->u.pirq.domid = domid;
 291         info->u.pirq.flags = flags;
 292
 293         return xen_irq_info_common_setup(info, irq, IRQT_PIRQ, evtchn, 0);
 294 }
 295
 296 static void xen_irq_info_cleanup(struct irq_info *info)
 297 {
 298         set_evtchn_to_irq(info->evtchn, -1);
 299         xen_evtchn_port_remove(info->evtchn, info->cpu);
 300         info->evtchn = 0;
 301 }
 302
 303 /*
 304  * Accessors for packed IRQ information.
 305  */
 306 unsigned int evtchn_from_irq(unsigned irq)
 307 {
 308         const struct irq_info *info = NULL;
 309
 310         if (likely(irq < nr_irqs))
 311                 info = info_for_irq(irq);
 312         if (!info)
 313                 return 0;
 314
 315         return info->evtchn;
 316 }
 317
 318 unsigned irq_from_evtchn(unsigned int evtchn)
 319 {
 320         return get_evtchn_to_irq(evtchn);
 321 }
 322 EXPORT_SYMBOL_GPL(irq_from_evtchn);
 323
 324 int irq_from_virq(unsigned int cpu, unsigned int virq)
 325 {
 326         return per_cpu(virq_to_irq, cpu)[virq];
 327 }
 328
 329 static enum ipi_vector ipi_from_irq(unsigned irq)
 330 {
 331         struct irq_info *info = info_for_irq(irq);
 332
 333         BUG_ON(info == NULL);
 334         BUG_ON(info->type != IRQT_IPI);
 335
 336         return info->u.ipi;
 337 }
 338
 339 static unsigned virq_from_irq(unsigned irq)
 340 {
 341         struct irq_info *info = info_for_irq(irq);
 342
 343         BUG_ON(info == NULL);
 344         BUG_ON(info->type != IRQT_VIRQ);
 345
 346         return info->u.virq;
 347 }
 348
 349 static unsigned pirq_from_irq(unsigned irq)
 350 {
 351         struct irq_info *info = info_for_irq(irq);
 352
 353         BUG_ON(info == NULL);
 354         BUG_ON(info->type != IRQT_PIRQ);
 355
 356         return info->u.pirq.pirq;
 357 }
 358
 359 static enum xen_irq_type type_from_irq(unsigned irq)
 360 {
 361         return info_for_irq(irq)->type;
 362 }
 363
 364 unsigned cpu_from_irq(unsigned irq)
 365 {
 366         return info_for_irq(irq)->cpu;
 367 }
 368
 369 unsigned int cpu_from_evtchn(unsigned int evtchn)
 370 {
 371         int irq = get_evtchn_to_irq(evtchn);
 372         unsigned ret = 0;
 373
 374         if (irq != -1)
 375                 ret = cpu_from_irq(irq);
 376
 377         return ret;
 378 }
 379
 380 static void do_mask(struct irq_info *info, u8 reason)
 381 {
 382         unsigned long flags;
 383
 384         raw_spin_lock_irqsave(&info->lock, flags);
 385
 386         if (!info->mask_reason)
 387                 mask_evtchn(info->evtchn);
 388
 389         info->mask_reason |= reason;
 390
 391         raw_spin_unlock_irqrestore(&info->lock, flags);
 392 }
 393
 394 static void do_unmask(struct irq_info *info, u8 reason)
 395 {
 396         unsigned long flags;
 397
 398         raw_spin_lock_irqsave(&info->lock, flags);
 399
 400         info->mask_reason &= ~reason;
 401
 402         if (!info->mask_reason)
 403                 unmask_evtchn(info->evtchn);
 404
 405         raw_spin_unlock_irqrestore(&info->lock, flags);
 406 }
 407
 408 #ifdef CONFIG_X86
 409 static bool pirq_check_eoi_map(unsigned irq)
 410 {
 411         return test_bit(pirq_from_irq(irq), pirq_eoi_map);
 412 }
 413 #endif
 414
 415 static bool pirq_needs_eoi_flag(unsigned irq)
 416 {
 417         struct irq_info *info = info_for_irq(irq);
 418         BUG_ON(info->type != IRQT_PIRQ);
 419
 420         return info->u.pirq.flags & PIRQ_NEEDS_EOI;
 421 }
 422
 423 static void bind_evtchn_to_cpu(unsigned int chn, unsigned int cpu)
 424 {
 425         int irq = get_evtchn_to_irq(chn);
 426         struct irq_info *info = info_for_irq(irq);
 427
 428         BUG_ON(irq == -1);
 429 #ifdef CONFIG_SMP
 430         cpumask_copy(irq_get_affinity_mask(irq), cpumask_of(cpu));
 431 #endif
 432         xen_evtchn_port_bind_to_cpu(info, cpu);
 433
 434         info->cpu = cpu;
 435 }
 436
 437 /**
 438  * notify_remote_via_irq - send event to remote end of event channel via irq
 439  * @irq: irq of event channel to send event to
 440  *
 441  * Unlike notify_remote_via_evtchn(), this is safe to use across
 442  * save/restore. Notifications on a broken connection are silently
 443  * dropped.
 444  */
 445 void notify_remote_via_irq(int irq)
 446 {
 447         int evtchn = evtchn_from_irq(irq);
 448
 449         if (VALID_EVTCHN(evtchn))
 450                 notify_remote_via_evtchn(evtchn);
 451 }
 452 EXPORT_SYMBOL_GPL(notify_remote_via_irq);
 453
 454 struct lateeoi_work {
 455         struct delayed_work delayed;
 456         spinlock_t eoi_list_lock;
 457         struct list_head eoi_list;
 458 };
 459
 460 static DEFINE_PER_CPU(struct lateeoi_work, lateeoi);
 461
 462 static void lateeoi_list_del(struct irq_info *info)
 463 {
 464         struct lateeoi_work *eoi = &per_cpu(lateeoi, info->eoi_cpu);
 465         unsigned long flags;
 466
 467         spin_lock_irqsave(&eoi->eoi_list_lock, flags);
 468         list_del_init(&info->eoi_list);
 469         spin_unlock_irqrestore(&eoi->eoi_list_lock, flags);
 470 }
 471
 472 static void lateeoi_list_add(struct irq_info *info)
 473 {
 474         struct lateeoi_work *eoi = &per_cpu(lateeoi, info->eoi_cpu);
 475         struct irq_info *elem;
 476         u64 now = get_jiffies_64();
 477         unsigned long delay;
 478         unsigned long flags;
 479
 480         if (now < info->eoi_time)
 481                 delay = info->eoi_time - now;
 482         else
 483                 delay = 1;
 484
 485         spin_lock_irqsave(&eoi->eoi_list_lock, flags);
 486
 487         if (list_empty(&eoi->eoi_list)) {
 488                 list_add(&info->eoi_list, &eoi->eoi_list);
 489                 mod_delayed_work_on(info->eoi_cpu, system_wq,
 490                                     &eoi->delayed, delay);
 491         } else {
 492                 list_for_each_entry_reverse(elem, &eoi->eoi_list, eoi_list) {
 493                         if (elem->eoi_time <= info->eoi_time)
 494                                 break;
 495                 }
 496                 list_add(&info->eoi_list, &elem->eoi_list);
 497         }
 498
 499         spin_unlock_irqrestore(&eoi->eoi_list_lock, flags);
 500 }
 501
 502 static void xen_irq_lateeoi_locked(struct irq_info *info, bool spurious)
 503 {
 504         evtchn_port_t evtchn;
 505         unsigned int cpu;
 506         unsigned int delay = 0;
 507
 508         evtchn = info->evtchn;
 509         if (!VALID_EVTCHN(evtchn) || !list_empty(&info->eoi_list))
 510                 return;
 511
 512         if (spurious) {
 513                 if ((1 << info->spurious_cnt) < (HZ << 2))
 514                         info->spurious_cnt++;
 515                 if (info->spurious_cnt > 1) {
 516                         delay = 1 << (info->spurious_cnt - 2);
 517                         if (delay > HZ)
 518                                 delay = HZ;
 519                         if (!info->eoi_time)
 520                                 info->eoi_cpu = smp_processor_id();
 521                         info->eoi_time = get_jiffies_64() + delay;
 522                 }
 523         } else {
 524                 info->spurious_cnt = 0;
 525         }
 526
 527         cpu = info->eoi_cpu;
 528         if (info->eoi_time &&
 529             (info->irq_epoch == per_cpu(irq_epoch, cpu) || delay)) {
 530                 lateeoi_list_add(info);
 531                 return;
 532         }
 533
 534         info->eoi_time = 0;
 535
 536         /* is_active hasn't been reset yet, do it now. */
 537         smp_store_release(&info->is_active, 0);
 538         do_unmask(info, EVT_MASK_REASON_EOI_PENDING);
 539 }
 540
 541 static void xen_irq_lateeoi_worker(struct work_struct *work)
 542 {
 543         struct lateeoi_work *eoi;
 544         struct irq_info *info;
 545         u64 now = get_jiffies_64();
 546         unsigned long flags;
 547
 548         eoi = container_of(to_delayed_work(work), struct lateeoi_work, delayed);
 549
 550         read_lock_irqsave(&evtchn_rwlock, flags);
 551
 552         while (true) {
 553                 spin_lock(&eoi->eoi_list_lock);
 554
 555                 info = list_first_entry_or_null(&eoi->eoi_list, struct irq_info,
 556                                                 eoi_list);
 557
 558                 if (info == NULL || now < info->eoi_time) {
 559                         spin_unlock(&eoi->eoi_list_lock);
 560                         break;
 561                 }
 562
 563                 list_del_init(&info->eoi_list);
 564
 565                 spin_unlock(&eoi->eoi_list_lock);
 566
 567                 info->eoi_time = 0;
 568
 569                 xen_irq_lateeoi_locked(info, false);
 570         }
 571
 572         if (info)
 573                 mod_delayed_work_on(info->eoi_cpu, system_wq,
 574                                     &eoi->delayed, info->eoi_time - now);
 575
 576         read_unlock_irqrestore(&evtchn_rwlock, flags);
 577 }
 578
 579 static void xen_cpu_init_eoi(unsigned int cpu)
 580 {
 581         struct lateeoi_work *eoi = &per_cpu(lateeoi, cpu);
 582
 583         INIT_DELAYED_WORK(&eoi->delayed, xen_irq_lateeoi_worker);
 584         spin_lock_init(&eoi->eoi_list_lock);
 585         INIT_LIST_HEAD(&eoi->eoi_list);
 586 }
 587
 588 void xen_irq_lateeoi(unsigned int irq, unsigned int eoi_flags)
 589 {
 590         struct irq_info *info;
 591         unsigned long flags;
 592
 593         read_lock_irqsave(&evtchn_rwlock, flags);
 594
 595         info = info_for_irq(irq);
 596
 597         if (info)
 598                 xen_irq_lateeoi_locked(info, eoi_flags & XEN_EOI_FLAG_SPURIOUS);
 599
 600         read_unlock_irqrestore(&evtchn_rwlock, flags);
 601 }
 602 EXPORT_SYMBOL_GPL(xen_irq_lateeoi);
 603
 604 static void xen_irq_init(unsigned irq)
 605 {
 606         struct irq_info *info;
 607
 608 #ifdef CONFIG_SMP
 609         /* By default all event channels notify CPU#0. */
 610         cpumask_copy(irq_get_affinity_mask(irq), cpumask_of(0));
 611 #endif
 612
 613         info = kzalloc(sizeof(*info), GFP_KERNEL);
 614         if (info == NULL)
 615                 panic("Unable to allocate metadata for IRQ%d\n", irq);
 616
 617         info->type = IRQT_UNBOUND;
 618         info->refcnt = -1;
 619
 620         set_info_for_irq(irq, info);
 621
 622         INIT_LIST_HEAD(&info->eoi_list);
 623         list_add_tail(&info->list, &xen_irq_list_head);
 624 }
 625
 626 static int __must_check xen_allocate_irqs_dynamic(int nvec)
 627 {
 628         int i, irq = irq_alloc_descs(-1, 0, nvec, -1);
 629
 630         if (irq >= 0) {
 631                 for (i = 0; i < nvec; i++)
 632                         xen_irq_init(irq + i);
 633         }
 634
 635         return irq;
 636 }
 637
 638 static inline int __must_check xen_allocate_irq_dynamic(void)
 639 {
 640
 641         return xen_allocate_irqs_dynamic(1);
 642 }
 643
 644 static int __must_check xen_allocate_irq_gsi(unsigned gsi)
 645 {
 646         int irq;
 647
 648         /*
 649          * A PV guest has no concept of a GSI (since it has no ACPI
 650          * nor access to/knowledge of the physical APICs). Therefore
 651          * all IRQs are dynamically allocated from the entire IRQ
 652          * space.
 653          */
 654         if (xen_pv_domain() && !xen_initial_domain())
 655                 return xen_allocate_irq_dynamic();
 656
 657         /* Legacy IRQ descriptors are already allocated by the arch. */
 658         if (gsi < nr_legacy_irqs())
 659                 irq = gsi;
 660         else
 661                 irq = irq_alloc_desc_at(gsi, -1);
 662
 663         xen_irq_init(irq);
 664
 665         return irq;
 666 }
 667
 668 static void xen_free_irq(unsigned irq)
 669 {
 670         struct irq_info *info = info_for_irq(irq);
 671         unsigned long flags;
 672
 673         if (WARN_ON(!info))
 674                 return;
 675
 676         write_lock_irqsave(&evtchn_rwlock, flags);
 677
 678         if (!list_empty(&info->eoi_list))
 679                 lateeoi_list_del(info);
 680
 681         list_del(&info->list);
 682
 683         set_info_for_irq(irq, NULL);
 684
 685         WARN_ON(info->refcnt > 0);
 686
 687         write_unlock_irqrestore(&evtchn_rwlock, flags);
 688
 689         kfree(info);
 690
 691         /* Legacy IRQ descriptors are managed by the arch. */
 692         if (irq < nr_legacy_irqs())
 693                 return;
 694
 695         irq_free_desc(irq);
 696 }
 697
 698 static void xen_evtchn_close(unsigned int port)
 699 {
 700         struct evtchn_close close;
 701
 702         close.port = port;
 703         if (HYPERVISOR_event_channel_op(EVTCHNOP_close, &close) != 0)
 704                 BUG();
 705 }
 706
 707 static void event_handler_exit(struct irq_info *info)
 708 {
 709         smp_store_release(&info->is_active, 0);
 710         clear_evtchn(info->evtchn);
 711 }
 712
 713 static void pirq_query_unmask(int irq)
 714 {
 715         struct physdev_irq_status_query irq_status;
 716         struct irq_info *info = info_for_irq(irq);
 717
 718         BUG_ON(info->type != IRQT_PIRQ);
 719
 720         irq_status.irq = pirq_from_irq(irq);
 721         if (HYPERVISOR_physdev_op(PHYSDEVOP_irq_status_query, &irq_status))
 722                 irq_status.flags = 0;
 723
 724         info->u.pirq.flags &= ~PIRQ_NEEDS_EOI;
 725         if (irq_status.flags & XENIRQSTAT_needs_eoi)
 726                 info->u.pirq.flags |= PIRQ_NEEDS_EOI;
 727 }
 728
 729 static void eoi_pirq(struct irq_data *data)
 730 {
 731         struct irq_info *info = info_for_irq(data->irq);
 732         int evtchn = info ? info->evtchn : 0;
 733         struct physdev_eoi eoi = { .irq = pirq_from_irq(data->irq) };
 734         int rc = 0;
 735
 736         if (!VALID_EVTCHN(evtchn))
 737                 return;
 738
 739         if (unlikely(irqd_is_setaffinity_pending(data)) &&
 740             likely(!irqd_irq_disabled(data))) {
 741                 do_mask(info, EVT_MASK_REASON_TEMPORARY);
 742
 743                 event_handler_exit(info);
 744
 745                 irq_move_masked_irq(data);
 746
 747                 do_unmask(info, EVT_MASK_REASON_TEMPORARY);
 748         } else
 749                 event_handler_exit(info);
 750
 751         if (pirq_needs_eoi(data->irq)) {
 752                 rc = HYPERVISOR_physdev_op(PHYSDEVOP_eoi, &eoi);
 753                 WARN_ON(rc);
 754         }
 755 }
 756
 757 static void mask_ack_pirq(struct irq_data *data)
 758 {
 759         disable_dynirq(data);
 760         eoi_pirq(data);
 761 }
 762
 763 static unsigned int __startup_pirq(unsigned int irq)
 764 {
 765         struct evtchn_bind_pirq bind_pirq;
 766         struct irq_info *info = info_for_irq(irq);
 767         int evtchn = evtchn_from_irq(irq);
 768         int rc;
 769
 770         BUG_ON(info->type != IRQT_PIRQ);
 771
 772         if (VALID_EVTCHN(evtchn))
 773                 goto out;
 774
 775         bind_pirq.pirq = pirq_from_irq(irq);
 776         /* NB. We are happy to share unless we are probing. */
 777         bind_pirq.flags = info->u.pirq.flags & PIRQ_SHAREABLE ?
 778                                         BIND_PIRQ__WILL_SHARE : 0;
 779         rc = HYPERVISOR_event_channel_op(EVTCHNOP_bind_pirq, &bind_pirq);
 780         if (rc != 0) {
 781                 pr_warn("Failed to obtain physical IRQ %d\n", irq);
 782                 return 0;
 783         }
 784         evtchn = bind_pirq.port;
 785
 786         pirq_query_unmask(irq);
 787
 788         rc = set_evtchn_to_irq(evtchn, irq);
 789         if (rc)
 790                 goto err;
 791
 792         info->evtchn = evtchn;
 793         bind_evtchn_to_cpu(evtchn, 0);
 794
 795         rc = xen_evtchn_port_setup(info);
 796         if (rc)
 797                 goto err;
 798
 799 out:
 800         do_unmask(info, EVT_MASK_REASON_EXPLICIT);
 801
 802         eoi_pirq(irq_get_irq_data(irq));
 803
 804         return 0;
 805
 806 err:
 807         pr_err("irq%d: Failed to set port to irq mapping (%d)\n", irq, rc);
 808         xen_evtchn_close(evtchn);
 809         return 0;
 810 }
 811
 812 static unsigned int startup_pirq(struct irq_data *data)
 813 {
 814         return __startup_pirq(data->irq);
 815 }
 816
 817 static void shutdown_pirq(struct irq_data *data)
 818 {
 819         unsigned int irq = data->irq;
 820         struct irq_info *info = info_for_irq(irq);
 821         unsigned evtchn = evtchn_from_irq(irq);
 822
 823         BUG_ON(info->type != IRQT_PIRQ);
 824
 825         if (!VALID_EVTCHN(evtchn))
 826                 return;
 827
 828         do_mask(info, EVT_MASK_REASON_EXPLICIT);
 829         xen_evtchn_close(evtchn);
 830         xen_irq_info_cleanup(info);
 831 }
 832
 833 static void enable_pirq(struct irq_data *data)
 834 {
 835         enable_dynirq(data);
 836 }
 837
 838 static void disable_pirq(struct irq_data *data)
 839 {
 840         disable_dynirq(data);
 841 }
 842
 843 int xen_irq_from_gsi(unsigned gsi)
 844 {
 845         struct irq_info *info;
 846
 847         list_for_each_entry(info, &xen_irq_list_head, list) {
 848                 if (info->type != IRQT_PIRQ)
 849                         continue;
 850
 851                 if (info->u.pirq.gsi == gsi)
 852                         return info->irq;
 853         }
 854
 855         return -1;
 856 }
 857 EXPORT_SYMBOL_GPL(xen_irq_from_gsi);
 858
 859 static void __unbind_from_irq(unsigned int irq)
 860 {
 861         int evtchn = evtchn_from_irq(irq);
 862         struct irq_info *info = info_for_irq(irq);
 863
 864         if (info->refcnt > 0) {
 865                 info->refcnt--;
 866                 if (info->refcnt != 0)
 867                         return;
 868         }
 869
 870         if (VALID_EVTCHN(evtchn)) {
 871                 unsigned int cpu = cpu_from_irq(irq);
 872
 873                 xen_evtchn_close(evtchn);
 874
 875                 switch (type_from_irq(irq)) {
 876                 case IRQT_VIRQ:
 877                         per_cpu(virq_to_irq, cpu)[virq_from_irq(irq)] = -1;
 878                         break;
 879                 case IRQT_IPI:
 880                         per_cpu(ipi_to_irq, cpu)[ipi_from_irq(irq)] = -1;
 881                         break;
 882                 default:
 883                         break;
 884                 }
 885
 886                 xen_irq_info_cleanup(info);
 887         }
 888
 889         xen_free_irq(irq);
 890 }
 891
 892 /*
 893  * Do not make any assumptions regarding the relationship between the
 894  * IRQ number returned here and the Xen pirq argument.
 895  *
 896  * Note: We don't assign an event channel until the irq actually started
 897  * up.  Return an existing irq if we've already got one for the gsi.
 898  *
 899  * Shareable implies level triggered, not shareable implies edge
 900  * triggered here.
 901  */
 902 int xen_bind_pirq_gsi_to_irq(unsigned gsi,
 903                              unsigned pirq, int shareable, char *name)
 904 {
 905         int irq = -1;
 906         struct physdev_irq irq_op;
 907         int ret;
 908
 909         mutex_lock(&irq_mapping_update_lock);
 910
 911         irq = xen_irq_from_gsi(gsi);
 912         if (irq != -1) {
 913                 pr_info("%s: returning irq %d for gsi %u\n",
 914                         __func__, irq, gsi);
 915                 goto out;
 916         }
 917
 918         irq = xen_allocate_irq_gsi(gsi);
 919         if (irq < 0)
 920                 goto out;
 921
 922         irq_op.irq = irq;
 923         irq_op.vector = 0;
 924
 925         /* Only the privileged domain can do this. For non-priv, the pcifront
 926          * driver provides a PCI bus that does the call to do exactly
 927          * this in the priv domain. */
 928         if (xen_initial_domain() &&
 929             HYPERVISOR_physdev_op(PHYSDEVOP_alloc_irq_vector, &irq_op)) {
 930                 xen_free_irq(irq);
 931                 irq = -ENOSPC;
 932                 goto out;
 933         }
 934
 935         ret = xen_irq_info_pirq_setup(irq, 0, pirq, gsi, DOMID_SELF,
 936                                shareable ? PIRQ_SHAREABLE : 0);
 937         if (ret < 0) {
 938                 __unbind_from_irq(irq);
 939                 irq = ret;
 940                 goto out;
 941         }
 942
 943         pirq_query_unmask(irq);
 944         /* We try to use the handler with the appropriate semantic for the
 945          * type of interrupt: if the interrupt is an edge triggered
 946          * interrupt we use handle_edge_irq.
 947          *
 948          * On the other hand if the interrupt is level triggered we use
 949          * handle_fasteoi_irq like the native code does for this kind of
 950          * interrupts.
 951          *
 952          * Depending on the Xen version, pirq_needs_eoi might return true
 953          * not only for level triggered interrupts but for edge triggered
 954          * interrupts too. In any case Xen always honors the eoi mechanism,
 955          * not injecting any more pirqs of the same kind if the first one
 956          * hasn't received an eoi yet. Therefore using the fasteoi handler
 957          * is the right choice either way.
 958          */
 959         if (shareable)
 960                 irq_set_chip_and_handler_name(irq, &xen_pirq_chip,
 961                                 handle_fasteoi_irq, name);
 962         else
 963                 irq_set_chip_and_handler_name(irq, &xen_pirq_chip,
 964                                 handle_edge_irq, name);
 965
 966 out:
 967         mutex_unlock(&irq_mapping_update_lock);
 968
 969         return irq;
 970 }
 971
 972 #ifdef CONFIG_PCI_MSI
 973 int xen_allocate_pirq_msi(struct pci_dev *dev, struct msi_desc *msidesc)
 974 {
 975         int rc;
 976         struct physdev_get_free_pirq op_get_free_pirq;
 977
 978         op_get_free_pirq.type = MAP_PIRQ_TYPE_MSI;
 979         rc = HYPERVISOR_physdev_op(PHYSDEVOP_get_free_pirq, &op_get_free_pirq);
 980
 981         WARN_ONCE(rc == -ENOSYS,
 982                   "hypervisor does not support the PHYSDEVOP_get_free_pirq interface\n");
 983
 984         return rc ? -1 : op_get_free_pirq.pirq;
 985 }
 986
 987 int xen_bind_pirq_msi_to_irq(struct pci_dev *dev, struct msi_desc *msidesc,
 988                              int pirq, int nvec, const char *name, domid_t domid)
 989 {
 990         int i, irq, ret;
 991
 992         mutex_lock(&irq_mapping_update_lock);
 993
 994         irq = xen_allocate_irqs_dynamic(nvec);
 995         if (irq < 0)
 996                 goto out;
 997
 998         for (i = 0; i < nvec; i++) {
 999                 irq_set_chip_and_handler_name(irq + i, &xen_pirq_chip, handle_edge_irq, name);
1000
1001                 ret = xen_irq_info_pirq_setup(irq + i, 0, pirq + i, 0, domid,
1002                                               i == 0 ? 0 : PIRQ_MSI_GROUP);
1003                 if (ret < 0)
1004                         goto error_irq;
1005         }
1006
1007         ret = irq_set_msi_desc(irq, msidesc);
1008         if (ret < 0)
1009                 goto error_irq;
1010 out:
1011         mutex_unlock(&irq_mapping_update_lock);
1012         return irq;
1013 error_irq:
1014         while (nvec--)
1015                 __unbind_from_irq(irq + nvec);
1016         mutex_unlock(&irq_mapping_update_lock);
1017         return ret;
1018 }
1019 #endif
1020
1021 int xen_destroy_irq(int irq)
1022 {
1023         struct physdev_unmap_pirq unmap_irq;
1024         struct irq_info *info = info_for_irq(irq);
1025         int rc = -ENOENT;
1026
1027         mutex_lock(&irq_mapping_update_lock);
1028
1029         /*
1030          * If trying to remove a vector in a MSI group different
1031          * than the first one skip the PIRQ unmap unless this vector
1032          * is the first one in the group.
1033          */
1034         if (xen_initial_domain() && !(info->u.pirq.flags & PIRQ_MSI_GROUP)) {
1035                 unmap_irq.pirq = info->u.pirq.pirq;
1036                 unmap_irq.domid = info->u.pirq.domid;
1037                 rc = HYPERVISOR_physdev_op(PHYSDEVOP_unmap_pirq, &unmap_irq);
1038                 /* If another domain quits without making the pci_disable_msix
1039                  * call, the Xen hypervisor takes care of freeing the PIRQs
1040                  * (free_domain_pirqs).
1041                  */
1042                 if ((rc == -ESRCH && info->u.pirq.domid != DOMID_SELF))
1043                         pr_info("domain %d does not have %d anymore\n",
1044                                 info->u.pirq.domid, info->u.pirq.pirq);
1045                 else if (rc) {
1046                         pr_warn("unmap irq failed %d\n", rc);
1047                         goto out;
1048                 }
1049         }
1050
1051         xen_free_irq(irq);
1052
1053 out:
1054         mutex_unlock(&irq_mapping_update_lock);
1055         return rc;
1056 }
1057
1058 int xen_irq_from_pirq(unsigned pirq)
1059 {
1060         int irq;
1061
1062         struct irq_info *info;
1063
1064         mutex_lock(&irq_mapping_update_lock);
1065
1066         list_for_each_entry(info, &xen_irq_list_head, list) {
1067                 if (info->type != IRQT_PIRQ)
1068                         continue;
1069                 irq = info->irq;
1070                 if (info->u.pirq.pirq == pirq)
1071                         goto out;
1072         }
1073         irq = -1;
1074 out:
1075         mutex_unlock(&irq_mapping_update_lock);
1076
1077         return irq;
1078 }
1079
1080
1081 int xen_pirq_from_irq(unsigned irq)
1082 {
1083         return pirq_from_irq(irq);
1084 }
1085 EXPORT_SYMBOL_GPL(xen_pirq_from_irq);
1086
1087 static int bind_evtchn_to_irq_chip(evtchn_port_t evtchn, struct irq_chip *chip)
1088 {
1089         int irq;
1090         int ret;
1091
1092         if (evtchn >= xen_evtchn_max_channels())
1093                 return -ENOMEM;
1094
1095         mutex_lock(&irq_mapping_update_lock);
1096
1097         irq = get_evtchn_to_irq(evtchn);
1098
1099         if (irq == -1) {
1100                 irq = xen_allocate_irq_dynamic();
1101                 if (irq < 0)
1102                         goto out;
1103
1104                 irq_set_chip_and_handler_name(irq, chip,
1105                                               handle_edge_irq, "event");
1106
1107                 ret = xen_irq_info_evtchn_setup(irq, evtchn);
1108                 if (ret < 0) {
1109                         __unbind_from_irq(irq);
1110                         irq = ret;
1111                         goto out;
1112                 }
1113                 /* New interdomain events are bound to VCPU 0. */
1114                 bind_evtchn_to_cpu(evtchn, 0);
1115         } else {
1116                 struct irq_info *info = info_for_irq(irq);
1117                 WARN_ON(info == NULL || info->type != IRQT_EVTCHN);
1118         }
1119
1120 out:
1121         mutex_unlock(&irq_mapping_update_lock);
1122
1123         return irq;
1124 }
1125
1126 int bind_evtchn_to_irq(evtchn_port_t evtchn)
1127 {
1128         return bind_evtchn_to_irq_chip(evtchn, &xen_dynamic_chip);
1129 }
1130 EXPORT_SYMBOL_GPL(bind_evtchn_to_irq);
1131
1132 int bind_evtchn_to_irq_lateeoi(evtchn_port_t evtchn)
1133 {
1134         return bind_evtchn_to_irq_chip(evtchn, &xen_lateeoi_chip);
1135 }
1136 EXPORT_SYMBOL_GPL(bind_evtchn_to_irq_lateeoi);
1137
1138 static int bind_ipi_to_irq(unsigned int ipi, unsigned int cpu)
1139 {
1140         struct evtchn_bind_ipi bind_ipi;
1141         int evtchn, irq;
1142         int ret;
1143
1144         mutex_lock(&irq_mapping_update_lock);
1145
1146         irq = per_cpu(ipi_to_irq, cpu)[ipi];
1147
1148         if (irq == -1) {
1149                 irq = xen_allocate_irq_dynamic();
1150                 if (irq < 0)
1151                         goto out;
1152
1153                 irq_set_chip_and_handler_name(irq, &xen_percpu_chip,
1154                                               handle_percpu_irq, "ipi");
1155
1156                 bind_ipi.vcpu = xen_vcpu_nr(cpu);
1157                 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_ipi,
1158                                                 &bind_ipi) != 0)
1159                         BUG();
1160                 evtchn = bind_ipi.port;
1161
1162                 ret = xen_irq_info_ipi_setup(cpu, irq, evtchn, ipi);
1163                 if (ret < 0) {
1164                         __unbind_from_irq(irq);
1165                         irq = ret;
1166                         goto out;
1167                 }
1168                 bind_evtchn_to_cpu(evtchn, cpu);
1169         } else {
1170                 struct irq_info *info = info_for_irq(irq);
1171                 WARN_ON(info == NULL || info->type != IRQT_IPI);
1172         }
1173
1174  out:
1175         mutex_unlock(&irq_mapping_update_lock);
1176         return irq;
1177 }
1178
1179 static int bind_interdomain_evtchn_to_irq_chip(unsigned int remote_domain,
1180                                                evtchn_port_t remote_port,
1181                                                struct irq_chip *chip)
1182 {
1183         struct evtchn_bind_interdomain bind_interdomain;
1184         int err;
1185
1186         bind_interdomain.remote_dom  = remote_domain;
1187         bind_interdomain.remote_port = remote_port;
1188
1189         err = HYPERVISOR_event_channel_op(EVTCHNOP_bind_interdomain,
1190                                           &bind_interdomain);
1191
1192         return err ? : bind_evtchn_to_irq_chip(bind_interdomain.local_port,
1193                                                chip);
1194 }
1195
1196 int bind_interdomain_evtchn_to_irq(unsigned int remote_domain,
1197                                    evtchn_port_t remote_port)
1198 {
1199         return bind_interdomain_evtchn_to_irq_chip(remote_domain, remote_port,
1200                                                    &xen_dynamic_chip);
1201 }
1202 EXPORT_SYMBOL_GPL(bind_interdomain_evtchn_to_irq);
1203
1204 int bind_interdomain_evtchn_to_irq_lateeoi(unsigned int remote_domain,
1205                                            evtchn_port_t remote_port)
1206 {
1207         return bind_interdomain_evtchn_to_irq_chip(remote_domain, remote_port,
1208                                                    &xen_lateeoi_chip);
1209 }
1210 EXPORT_SYMBOL_GPL(bind_interdomain_evtchn_to_irq_lateeoi);
1211
1212 static int find_virq(unsigned int virq, unsigned int cpu)
1213 {
1214         struct evtchn_status status;
1215         int port, rc = -ENOENT;
1216
1217         memset(&status, 0, sizeof(status));
1218         for (port = 0; port < xen_evtchn_max_channels(); port++) {
1219                 status.dom = DOMID_SELF;
1220                 status.port = port;
1221                 rc = HYPERVISOR_event_channel_op(EVTCHNOP_status, &status);
1222                 if (rc < 0)
1223                         continue;
1224                 if (status.status != EVTCHNSTAT_virq)
1225                         continue;
1226                 if (status.u.virq == virq && status.vcpu == xen_vcpu_nr(cpu)) {
1227                         rc = port;
1228                         break;
1229                 }
1230         }
1231         return rc;
1232 }
1233
1234 /**
1235  * xen_evtchn_nr_channels - number of usable event channel ports
1236  *
1237  * This may be less than the maximum supported by the current
1238  * hypervisor ABI. Use xen_evtchn_max_channels() for the maximum
1239  * supported.
1240  */
1241 unsigned xen_evtchn_nr_channels(void)
1242 {
1243         return evtchn_ops->nr_channels();
1244 }
1245 EXPORT_SYMBOL_GPL(xen_evtchn_nr_channels);
1246
1247 int bind_virq_to_irq(unsigned int virq, unsigned int cpu, bool percpu)
1248 {
1249         struct evtchn_bind_virq bind_virq;
1250         int evtchn, irq, ret;
1251
1252         mutex_lock(&irq_mapping_update_lock);
1253
1254         irq = per_cpu(virq_to_irq, cpu)[virq];
1255
1256         if (irq == -1) {
1257                 irq = xen_allocate_irq_dynamic();
1258                 if (irq < 0)
1259                         goto out;
1260
1261                 if (percpu)
1262                         irq_set_chip_and_handler_name(irq, &xen_percpu_chip,
1263                                                       handle_percpu_irq, "virq");
1264                 else
1265                         irq_set_chip_and_handler_name(irq, &xen_dynamic_chip,
1266                                                       handle_edge_irq, "virq");
1267
1268                 bind_virq.virq = virq;
1269                 bind_virq.vcpu = xen_vcpu_nr(cpu);
1270                 ret = HYPERVISOR_event_channel_op(EVTCHNOP_bind_virq,
1271                                                 &bind_virq);
1272                 if (ret == 0)
1273                         evtchn = bind_virq.port;
1274                 else {
1275                         if (ret == -EEXIST)
1276                                 ret = find_virq(virq, cpu);
1277                         BUG_ON(ret < 0);
1278                         evtchn = ret;
1279                 }
1280
1281                 ret = xen_irq_info_virq_setup(cpu, irq, evtchn, virq);
1282                 if (ret < 0) {
1283                         __unbind_from_irq(irq);
1284                         irq = ret;
1285                         goto out;
1286                 }
1287
1288                 bind_evtchn_to_cpu(evtchn, cpu);
1289         } else {
1290                 struct irq_info *info = info_for_irq(irq);
1291                 WARN_ON(info == NULL || info->type != IRQT_VIRQ);
1292         }
1293
1294 out:
1295         mutex_unlock(&irq_mapping_update_lock);
1296
1297         return irq;
1298 }
1299
1300 static void unbind_from_irq(unsigned int irq)
1301 {
1302         mutex_lock(&irq_mapping_update_lock);
1303         __unbind_from_irq(irq);
1304         mutex_unlock(&irq_mapping_update_lock);
1305 }
1306
1307 static int bind_evtchn_to_irqhandler_chip(evtchn_port_t evtchn,
1308                                           irq_handler_t handler,
1309                                           unsigned long irqflags,
1310                                           const char *devname, void *dev_id,
1311                                           struct irq_chip *chip)
1312 {
1313         int irq, retval;
1314
1315         irq = bind_evtchn_to_irq_chip(evtchn, chip);
1316         if (irq < 0)
1317                 return irq;
1318         retval = request_irq(irq, handler, irqflags, devname, dev_id);
1319         if (retval != 0) {
1320                 unbind_from_irq(irq);
1321                 return retval;
1322         }
1323
1324         return irq;
1325 }
1326
1327 int bind_evtchn_to_irqhandler(evtchn_port_t evtchn,
1328                               irq_handler_t handler,
1329                               unsigned long irqflags,
1330                               const char *devname, void *dev_id)
1331 {
1332         return bind_evtchn_to_irqhandler_chip(evtchn, handler, irqflags,
1333                                               devname, dev_id,
1334                                               &xen_dynamic_chip);
1335 }
1336 EXPORT_SYMBOL_GPL(bind_evtchn_to_irqhandler);
1337
1338 int bind_evtchn_to_irqhandler_lateeoi(evtchn_port_t evtchn,
1339                                       irq_handler_t handler,
1340                                       unsigned long irqflags,
1341                                       const char *devname, void *dev_id)
1342 {
1343         return bind_evtchn_to_irqhandler_chip(evtchn, handler, irqflags,
1344                                               devname, dev_id,
1345                                               &xen_lateeoi_chip);
1346 }
1347 EXPORT_SYMBOL_GPL(bind_evtchn_to_irqhandler_lateeoi);
1348
1349 static int bind_interdomain_evtchn_to_irqhandler_chip(
1350                 unsigned int remote_domain, evtchn_port_t remote_port,
1351                 irq_handler_t handler, unsigned long irqflags,
1352                 const char *devname, void *dev_id, struct irq_chip *chip)
1353 {
1354         int irq, retval;
1355
1356         irq = bind_interdomain_evtchn_to_irq_chip(remote_domain, remote_port,
1357                                                   chip);
1358         if (irq < 0)
1359                 return irq;
1360
1361         retval = request_irq(irq, handler, irqflags, devname, dev_id);
1362         if (retval != 0) {
1363                 unbind_from_irq(irq);
1364                 return retval;
1365         }
1366
1367         return irq;
1368 }
1369
1370 int bind_interdomain_evtchn_to_irqhandler(unsigned int remote_domain,
1371                                           evtchn_port_t remote_port,
1372                                           irq_handler_t handler,
1373                                           unsigned long irqflags,
1374                                           const char *devname,
1375                                           void *dev_id)
1376 {
1377         return bind_interdomain_evtchn_to_irqhandler_chip(remote_domain,
1378                                 remote_port, handler, irqflags, devname,
1379                                 dev_id, &xen_dynamic_chip);
1380 }
1381 EXPORT_SYMBOL_GPL(bind_interdomain_evtchn_to_irqhandler);
1382
1383 int bind_interdomain_evtchn_to_irqhandler_lateeoi(unsigned int remote_domain,
1384                                                   evtchn_port_t remote_port,
1385                                                   irq_handler_t handler,
1386                                                   unsigned long irqflags,
1387                                                   const char *devname,
1388                                                   void *dev_id)
1389 {
1390         return bind_interdomain_evtchn_to_irqhandler_chip(remote_domain,
1391                                 remote_port, handler, irqflags, devname,
1392                                 dev_id, &xen_lateeoi_chip);
1393 }
1394 EXPORT_SYMBOL_GPL(bind_interdomain_evtchn_to_irqhandler_lateeoi);
1395
1396 int bind_virq_to_irqhandler(unsigned int virq, unsigned int cpu,
1397                             irq_handler_t handler,
1398                             unsigned long irqflags, const char *devname, void *dev_id)
1399 {
1400         int irq, retval;
1401
1402         irq = bind_virq_to_irq(virq, cpu, irqflags & IRQF_PERCPU);
1403         if (irq < 0)
1404                 return irq;
1405         retval = request_irq(irq, handler, irqflags, devname, dev_id);
1406         if (retval != 0) {
1407                 unbind_from_irq(irq);
1408                 return retval;
1409         }
1410
1411         return irq;
1412 }
1413 EXPORT_SYMBOL_GPL(bind_virq_to_irqhandler);
1414
1415 int bind_ipi_to_irqhandler(enum ipi_vector ipi,
1416                            unsigned int cpu,
1417                            irq_handler_t handler,
1418                            unsigned long irqflags,
1419                            const char *devname,
1420                            void *dev_id)
1421 {
1422         int irq, retval;
1423
1424         irq = bind_ipi_to_irq(ipi, cpu);
1425         if (irq < 0)
1426                 return irq;
1427
1428         irqflags |= IRQF_NO_SUSPEND | IRQF_FORCE_RESUME | IRQF_EARLY_RESUME;
1429         retval = request_irq(irq, handler, irqflags, devname, dev_id);
1430         if (retval != 0) {
1431                 unbind_from_irq(irq);
1432                 return retval;
1433         }
1434
1435         return irq;
1436 }
1437
1438 void unbind_from_irqhandler(unsigned int irq, void *dev_id)
1439 {
1440         struct irq_info *info = info_for_irq(irq);
1441
1442         if (WARN_ON(!info))
1443                 return;
1444         free_irq(irq, dev_id);
1445         unbind_from_irq(irq);
1446 }
1447 EXPORT_SYMBOL_GPL(unbind_from_irqhandler);
1448
1449 /**
1450  * xen_set_irq_priority() - set an event channel priority.
1451  * @irq:irq bound to an event channel.
1452  * @priority: priority between XEN_IRQ_PRIORITY_MAX and XEN_IRQ_PRIORITY_MIN.
1453  */
1454 int xen_set_irq_priority(unsigned irq, unsigned priority)
1455 {
1456         struct evtchn_set_priority set_priority;
1457
1458         set_priority.port = evtchn_from_irq(irq);
1459         set_priority.priority = priority;
1460
1461         return HYPERVISOR_event_channel_op(EVTCHNOP_set_priority,
1462                                            &set_priority);
1463 }
1464 EXPORT_SYMBOL_GPL(xen_set_irq_priority);
1465
1466 int evtchn_make_refcounted(unsigned int evtchn)
1467 {
1468         int irq = get_evtchn_to_irq(evtchn);
1469         struct irq_info *info;
1470
1471         if (irq == -1)
1472                 return -ENOENT;
1473
1474         info = info_for_irq(irq);
1475
1476         if (!info)
1477                 return -ENOENT;
1478
1479         WARN_ON(info->refcnt != -1);
1480
1481         info->refcnt = 1;
1482
1483         return 0;
1484 }
1485 EXPORT_SYMBOL_GPL(evtchn_make_refcounted);
1486
1487 int evtchn_get(unsigned int evtchn)
1488 {
1489         int irq;
1490         struct irq_info *info;
1491         int err = -ENOENT;
1492
1493         if (evtchn >= xen_evtchn_max_channels())
1494                 return -EINVAL;
1495
1496         mutex_lock(&irq_mapping_update_lock);
1497
1498         irq = get_evtchn_to_irq(evtchn);
1499         if (irq == -1)
1500                 goto done;
1501
1502         info = info_for_irq(irq);
1503
1504         if (!info)
1505                 goto done;
1506
1507         err = -EINVAL;
1508         if (info->refcnt <= 0 || info->refcnt == SHRT_MAX)
1509                 goto done;
1510
1511         info->refcnt++;
1512         err = 0;
1513  done:
1514         mutex_unlock(&irq_mapping_update_lock);
1515
1516         return err;
1517 }
1518 EXPORT_SYMBOL_GPL(evtchn_get);
1519
1520 void evtchn_put(unsigned int evtchn)
1521 {
1522         int irq = get_evtchn_to_irq(evtchn);
1523         if (WARN_ON(irq == -1))
1524                 return;
1525         unbind_from_irq(irq);
1526 }
1527 EXPORT_SYMBOL_GPL(evtchn_put);
1528
1529 void xen_send_IPI_one(unsigned int cpu, enum ipi_vector vector)
1530 {
1531         int irq;
1532
1533 #ifdef CONFIG_X86
1534         if (unlikely(vector == XEN_NMI_VECTOR)) {
1535                 int rc =  HYPERVISOR_vcpu_op(VCPUOP_send_nmi, xen_vcpu_nr(cpu),
1536                                              NULL);
1537                 if (rc < 0)
1538                         printk(KERN_WARNING "Sending nmi to CPU%d failed (rc:%d)\n", cpu, rc);
1539                 return;
1540         }
1541 #endif
1542         irq = per_cpu(ipi_to_irq, cpu)[vector];
1543         BUG_ON(irq < 0);
1544         notify_remote_via_irq(irq);
1545 }
1546
1547 struct evtchn_loop_ctrl {
1548         ktime_t timeout;
1549         unsigned count;
1550         bool defer_eoi;
1551 };
1552
1553 void handle_irq_for_port(evtchn_port_t port, struct evtchn_loop_ctrl *ctrl)
1554 {
1555         int irq;
1556         struct irq_info *info;
1557
1558         irq = get_evtchn_to_irq(port);
1559         if (irq == -1)
1560                 return;
1561
1562         /*
1563          * Check for timeout every 256 events.
1564          * We are setting the timeout value only after the first 256
1565          * events in order to not hurt the common case of few loop
1566          * iterations. The 256 is basically an arbitrary value.
1567          *
1568          * In case we are hitting the timeout we need to defer all further
1569          * EOIs in order to ensure to leave the event handling loop rather
1570          * sooner than later.
1571          */
1572         if (!ctrl->defer_eoi && !(++ctrl->count & 0xff)) {
1573                 ktime_t kt = ktime_get();
1574
1575                 if (!ctrl->timeout) {
1576                         kt = ktime_add_ms(kt,
1577                                           jiffies_to_msecs(event_loop_timeout));
1578                         ctrl->timeout = kt;
1579                 } else if (kt > ctrl->timeout) {
1580                         ctrl->defer_eoi = true;
1581                 }
1582         }
1583
1584         info = info_for_irq(irq);
1585         if (xchg_acquire(&info->is_active, 1))
1586                 return;
1587
1588         if (ctrl->defer_eoi) {
1589                 info->eoi_cpu = smp_processor_id();
1590                 info->irq_epoch = __this_cpu_read(irq_epoch);
1591                 info->eoi_time = get_jiffies_64() + event_eoi_delay;
1592         }
1593
1594         generic_handle_irq(irq);
1595 }
1596
1597 static DEFINE_PER_CPU(unsigned, xed_nesting_count);
1598
1599 static void __xen_evtchn_do_upcall(void)
1600 {
1601         struct vcpu_info *vcpu_info = __this_cpu_read(xen_vcpu);
1602         int cpu = get_cpu();
1603         unsigned count;
1604         struct evtchn_loop_ctrl ctrl = { 0 };
1605
1606         read_lock(&evtchn_rwlock);
1607
1608         do {
1609                 vcpu_info->evtchn_upcall_pending = 0;
1610
1611                 if (__this_cpu_inc_return(xed_nesting_count) - 1)
1612                         goto out;
1613
1614                 xen_evtchn_handle_events(cpu, &ctrl);
1615
1616                 BUG_ON(!irqs_disabled());
1617
1618                 count = __this_cpu_read(xed_nesting_count);
1619                 __this_cpu_write(xed_nesting_count, 0);
1620         } while (count != 1 || vcpu_info->evtchn_upcall_pending);
1621
1622 out:
1623         read_unlock(&evtchn_rwlock);
1624
1625         /*
1626          * Increment irq_epoch only now to defer EOIs only for
1627          * xen_irq_lateeoi() invocations occurring from inside the loop
1628          * above.
1629          */
1630         __this_cpu_inc(irq_epoch);
1631
1632         put_cpu();
1633 }
1634
1635 void xen_evtchn_do_upcall(struct pt_regs *regs)
1636 {
1637         struct pt_regs *old_regs = set_irq_regs(regs);
1638
1639         irq_enter();
1640 #ifdef CONFIG_X86
1641         inc_irq_stat(irq_hv_callback_count);
1642 #endif
1643
1644         __xen_evtchn_do_upcall();
1645
1646         irq_exit();
1647         set_irq_regs(old_regs);
1648 }
1649
1650 void xen_hvm_evtchn_do_upcall(void)
1651 {
1652         __xen_evtchn_do_upcall();
1653 }
1654 EXPORT_SYMBOL_GPL(xen_hvm_evtchn_do_upcall);
1655
1656 /* Rebind a new event channel to an existing irq. */
1657 void rebind_evtchn_irq(int evtchn, int irq)
1658 {
1659         struct irq_info *info = info_for_irq(irq);
1660
1661         if (WARN_ON(!info))
1662                 return;
1663
1664         /* Make sure the irq is masked, since the new event channel
1665            will also be masked. */
1666         disable_irq(irq);
1667
1668         mutex_lock(&irq_mapping_update_lock);
1669
1670         /* After resume the irq<->evtchn mappings are all cleared out */
1671         BUG_ON(get_evtchn_to_irq(evtchn) != -1);
1672         /* Expect irq to have been bound before,
1673            so there should be a proper type */
1674         BUG_ON(info->type == IRQT_UNBOUND);
1675
1676         (void)xen_irq_info_evtchn_setup(irq, evtchn);
1677
1678         mutex_unlock(&irq_mapping_update_lock);
1679
1680         bind_evtchn_to_cpu(evtchn, info->cpu);
1681         /* This will be deferred until interrupt is processed */
1682         irq_set_affinity(irq, cpumask_of(info->cpu));
1683
1684         /* Unmask the event channel. */
1685         enable_irq(irq);
1686 }
1687
1688 /* Rebind an evtchn so that it gets delivered to a specific cpu */
1689 static int xen_rebind_evtchn_to_cpu(struct irq_info *info, unsigned int tcpu)
1690 {
1691         struct evtchn_bind_vcpu bind_vcpu;
1692         evtchn_port_t evtchn = info ? info->evtchn : 0;
1693
1694         if (!VALID_EVTCHN(evtchn))
1695                 return -1;
1696
1697         if (!xen_support_evtchn_rebind())
1698                 return -1;
1699
1700         /* Send future instances of this interrupt to other vcpu. */
1701         bind_vcpu.port = evtchn;
1702         bind_vcpu.vcpu = xen_vcpu_nr(tcpu);
1703
1704         /*
1705          * Mask the event while changing the VCPU binding to prevent
1706          * it being delivered on an unexpected VCPU.
1707          */
1708         do_mask(info, EVT_MASK_REASON_TEMPORARY);
1709
1710         /*
1711          * If this fails, it usually just indicates that we're dealing with a
1712          * virq or IPI channel, which don't actually need to be rebound. Ignore
1713          * it, but don't do the xenlinux-level rebind in that case.
1714          */
1715         if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_vcpu, &bind_vcpu) >= 0)
1716                 bind_evtchn_to_cpu(evtchn, tcpu);
1717
1718         do_unmask(info, EVT_MASK_REASON_TEMPORARY);
1719
1720         return 0;
1721 }
1722
1723 static int set_affinity_irq(struct irq_data *data, const struct cpumask *dest,
1724                             bool force)
1725 {
1726         unsigned tcpu = cpumask_first_and(dest, cpu_online_mask);
1727         int ret = xen_rebind_evtchn_to_cpu(info_for_irq(data->irq), tcpu);
1728
1729         if (!ret)
1730                 irq_data_update_effective_affinity(data, cpumask_of(tcpu));
1731
1732         return ret;
1733 }
1734
1735 /* To be called with desc->lock held. */
1736 int xen_set_affinity_evtchn(struct irq_desc *desc, unsigned int tcpu)
1737 {
1738         struct irq_data *d = irq_desc_get_irq_data(desc);
1739
1740         return set_affinity_irq(d, cpumask_of(tcpu), false);
1741 }
1742 EXPORT_SYMBOL_GPL(xen_set_affinity_evtchn);
1743
1744 static void enable_dynirq(struct irq_data *data)
1745 {
1746         struct irq_info *info = info_for_irq(data->irq);
1747         evtchn_port_t evtchn = info ? info->evtchn : 0;
1748
1749         if (VALID_EVTCHN(evtchn))
1750                 do_unmask(info, EVT_MASK_REASON_EXPLICIT);
1751 }
1752
1753 static void disable_dynirq(struct irq_data *data)
1754 {
1755         struct irq_info *info = info_for_irq(data->irq);
1756         evtchn_port_t evtchn = info ? info->evtchn : 0;
1757
1758         if (VALID_EVTCHN(evtchn))
1759                 do_mask(info, EVT_MASK_REASON_EXPLICIT);
1760 }
1761
1762 static void ack_dynirq(struct irq_data *data)
1763 {
1764         struct irq_info *info = info_for_irq(data->irq);
1765         evtchn_port_t evtchn = info ? info->evtchn : 0;
1766
1767         if (!VALID_EVTCHN(evtchn))
1768                 return;
1769
1770         if (unlikely(irqd_is_setaffinity_pending(data)) &&
1771             likely(!irqd_irq_disabled(data))) {
1772                 do_mask(info, EVT_MASK_REASON_TEMPORARY);
1773
1774                 event_handler_exit(info);
1775
1776                 irq_move_masked_irq(data);
1777
1778                 do_unmask(info, EVT_MASK_REASON_TEMPORARY);
1779         } else
1780                 event_handler_exit(info);
1781 }
1782
1783 static void mask_ack_dynirq(struct irq_data *data)
1784 {
1785         disable_dynirq(data);
1786         ack_dynirq(data);
1787 }
1788
1789 static void lateeoi_ack_dynirq(struct irq_data *data)
1790 {
1791         struct irq_info *info = info_for_irq(data->irq);
1792         evtchn_port_t evtchn = info ? info->evtchn : 0;
1793
1794         if (!VALID_EVTCHN(evtchn))
1795                 return;
1796
1797         do_mask(info, EVT_MASK_REASON_EOI_PENDING);
1798
1799         if (unlikely(irqd_is_setaffinity_pending(data)) &&
1800             likely(!irqd_irq_disabled(data))) {
1801                 do_mask(info, EVT_MASK_REASON_TEMPORARY);
1802
1803                 clear_evtchn(evtchn);
1804
1805                 irq_move_masked_irq(data);
1806
1807                 do_unmask(info, EVT_MASK_REASON_TEMPORARY);
1808         } else
1809                 clear_evtchn(evtchn);
1810 }
1811
1812 static void lateeoi_mask_ack_dynirq(struct irq_data *data)
1813 {
1814         struct irq_info *info = info_for_irq(data->irq);
1815         evtchn_port_t evtchn = info ? info->evtchn : 0;
1816
1817         if (VALID_EVTCHN(evtchn)) {
1818                 do_mask(info, EVT_MASK_REASON_EXPLICIT);
1819                 ack_dynirq(data);
1820         }
1821 }
1822
1823 static int retrigger_dynirq(struct irq_data *data)
1824 {
1825         struct irq_info *info = info_for_irq(data->irq);
1826         evtchn_port_t evtchn = info ? info->evtchn : 0;
1827
1828         if (!VALID_EVTCHN(evtchn))
1829                 return 0;
1830
1831         do_mask(info, EVT_MASK_REASON_TEMPORARY);
1832         set_evtchn(evtchn);
1833         do_unmask(info, EVT_MASK_REASON_TEMPORARY);
1834
1835         return 1;
1836 }
1837
1838 static void restore_pirqs(void)
1839 {
1840         int pirq, rc, irq, gsi;
1841         struct physdev_map_pirq map_irq;
1842         struct irq_info *info;
1843
1844         list_for_each_entry(info, &xen_irq_list_head, list) {
1845                 if (info->type != IRQT_PIRQ)
1846                         continue;
1847
1848                 pirq = info->u.pirq.pirq;
1849                 gsi = info->u.pirq.gsi;
1850                 irq = info->irq;
1851
1852                 /* save/restore of PT devices doesn't work, so at this point the
1853                  * only devices present are GSI based emulated devices */
1854                 if (!gsi)
1855                         continue;
1856
1857                 map_irq.domid = DOMID_SELF;
1858                 map_irq.type = MAP_PIRQ_TYPE_GSI;
1859                 map_irq.index = gsi;
1860                 map_irq.pirq = pirq;
1861
1862                 rc = HYPERVISOR_physdev_op(PHYSDEVOP_map_pirq, &map_irq);
1863                 if (rc) {
1864                         pr_warn("xen map irq failed gsi=%d irq=%d pirq=%d rc=%d\n",
1865                                 gsi, irq, pirq, rc);
1866                         xen_free_irq(irq);
1867                         continue;
1868                 }
1869
1870                 printk(KERN_DEBUG "xen: --> irq=%d, pirq=%d\n", irq, map_irq.pirq);
1871
1872                 __startup_pirq(irq);
1873         }
1874 }
1875
1876 static void restore_cpu_virqs(unsigned int cpu)
1877 {
1878         struct evtchn_bind_virq bind_virq;
1879         int virq, irq, evtchn;
1880
1881         for (virq = 0; virq < NR_VIRQS; virq++) {
1882                 if ((irq = per_cpu(virq_to_irq, cpu)[virq]) == -1)
1883                         continue;
1884
1885                 BUG_ON(virq_from_irq(irq) != virq);
1886
1887                 /* Get a new binding from Xen. */
1888                 bind_virq.virq = virq;
1889                 bind_virq.vcpu = xen_vcpu_nr(cpu);
1890                 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_virq,
1891                                                 &bind_virq) != 0)
1892                         BUG();
1893                 evtchn = bind_virq.port;
1894
1895                 /* Record the new mapping. */
1896                 (void)xen_irq_info_virq_setup(cpu, irq, evtchn, virq);
1897                 bind_evtchn_to_cpu(evtchn, cpu);
1898         }
1899 }
1900
1901 static void restore_cpu_ipis(unsigned int cpu)
1902 {
1903         struct evtchn_bind_ipi bind_ipi;
1904         int ipi, irq, evtchn;
1905
1906         for (ipi = 0; ipi < XEN_NR_IPIS; ipi++) {
1907                 if ((irq = per_cpu(ipi_to_irq, cpu)[ipi]) == -1)
1908                         continue;
1909
1910                 BUG_ON(ipi_from_irq(irq) != ipi);
1911
1912                 /* Get a new binding from Xen. */
1913                 bind_ipi.vcpu = xen_vcpu_nr(cpu);
1914                 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_ipi,
1915                                                 &bind_ipi) != 0)
1916                         BUG();
1917                 evtchn = bind_ipi.port;
1918
1919                 /* Record the new mapping. */
1920                 (void)xen_irq_info_ipi_setup(cpu, irq, evtchn, ipi);
1921                 bind_evtchn_to_cpu(evtchn, cpu);
1922         }
1923 }
1924
1925 /* Clear an irq's pending state, in preparation for polling on it */
1926 void xen_clear_irq_pending(int irq)
1927 {
1928         struct irq_info *info = info_for_irq(irq);
1929         evtchn_port_t evtchn = info ? info->evtchn : 0;
1930
1931         if (VALID_EVTCHN(evtchn))
1932                 event_handler_exit(info);
1933 }
1934 EXPORT_SYMBOL(xen_clear_irq_pending);
1935 void xen_set_irq_pending(int irq)
1936 {
1937         int evtchn = evtchn_from_irq(irq);
1938
1939         if (VALID_EVTCHN(evtchn))
1940                 set_evtchn(evtchn);
1941 }
1942
1943 bool xen_test_irq_pending(int irq)
1944 {
1945         int evtchn = evtchn_from_irq(irq);
1946         bool ret = false;
1947
1948         if (VALID_EVTCHN(evtchn))
1949                 ret = test_evtchn(evtchn);
1950
1951         return ret;
1952 }
1953
1954 /* Poll waiting for an irq to become pending with timeout.  In the usual case,
1955  * the irq will be disabled so it won't deliver an interrupt. */
1956 void xen_poll_irq_timeout(int irq, u64 timeout)
1957 {
1958         evtchn_port_t evtchn = evtchn_from_irq(irq);
1959
1960         if (VALID_EVTCHN(evtchn)) {
1961                 struct sched_poll poll;
1962
1963                 poll.nr_ports = 1;
1964                 poll.timeout = timeout;
1965                 set_xen_guest_handle(poll.ports, &evtchn);
1966
1967                 if (HYPERVISOR_sched_op(SCHEDOP_poll, &poll) != 0)
1968                         BUG();
1969         }
1970 }
1971 EXPORT_SYMBOL(xen_poll_irq_timeout);
1972 /* Poll waiting for an irq to become pending.  In the usual case, the
1973  * irq will be disabled so it won't deliver an interrupt. */
1974 void xen_poll_irq(int irq)
1975 {
1976         xen_poll_irq_timeout(irq, 0 /* no timeout */);
1977 }
1978
1979 /* Check whether the IRQ line is shared with other guests. */
1980 int xen_test_irq_shared(int irq)
1981 {
1982         struct irq_info *info = info_for_irq(irq);
1983         struct physdev_irq_status_query irq_status;
1984
1985         if (WARN_ON(!info))
1986                 return -ENOENT;
1987
1988         irq_status.irq = info->u.pirq.pirq;
1989
1990         if (HYPERVISOR_physdev_op(PHYSDEVOP_irq_status_query, &irq_status))
1991                 return 0;
1992         return !(irq_status.flags & XENIRQSTAT_shared);
1993 }
1994 EXPORT_SYMBOL_GPL(xen_test_irq_shared);
1995
1996 void xen_irq_resume(void)
1997 {
1998         unsigned int cpu;
1999         struct irq_info *info;
2000
2001         /* New event-channel space is not 'live' yet. */
2002         xen_evtchn_resume();
2003
2004         /* No IRQ <-> event-channel mappings. */
2005         list_for_each_entry(info, &xen_irq_list_head, list)
2006                 info->evtchn = 0; /* zap event-channel binding */
2007
2008         clear_evtchn_to_irq_all();
2009
2010         for_each_possible_cpu(cpu) {
2011                 restore_cpu_virqs(cpu);
2012                 restore_cpu_ipis(cpu);
2013         }
2014
2015         restore_pirqs();
2016 }
2017
2018 static struct irq_chip xen_dynamic_chip __read_mostly = {
2019         .name                   = "xen-dyn",
2020
2021         .irq_disable            = disable_dynirq,
2022         .irq_mask               = disable_dynirq,
2023         .irq_unmask             = enable_dynirq,
2024
2025         .irq_ack                = ack_dynirq,
2026         .irq_mask_ack           = mask_ack_dynirq,
2027
2028         .irq_set_affinity       = set_affinity_irq,
2029         .irq_retrigger          = retrigger_dynirq,
2030 };
2031
2032 static struct irq_chip xen_lateeoi_chip __read_mostly = {
2033         /* The chip name needs to contain "xen-dyn" for irqbalance to work. */
2034         .name                   = "xen-dyn-lateeoi",
2035
2036         .irq_disable            = disable_dynirq,
2037         .irq_mask               = disable_dynirq,
2038         .irq_unmask             = enable_dynirq,
2039
2040         .irq_ack                = lateeoi_ack_dynirq,
2041         .irq_mask_ack           = lateeoi_mask_ack_dynirq,
2042
2043         .irq_set_affinity       = set_affinity_irq,
2044         .irq_retrigger          = retrigger_dynirq,
2045 };
2046
2047 static struct irq_chip xen_pirq_chip __read_mostly = {
2048         .name                   = "xen-pirq",
2049
2050         .irq_startup            = startup_pirq,
2051         .irq_shutdown           = shutdown_pirq,
2052         .irq_enable             = enable_pirq,
2053         .irq_disable            = disable_pirq,
2054
2055         .irq_mask               = disable_dynirq,
2056         .irq_unmask             = enable_dynirq,
2057
2058         .irq_ack                = eoi_pirq,
2059         .irq_eoi                = eoi_pirq,
2060         .irq_mask_ack           = mask_ack_pirq,
2061
2062         .irq_set_affinity       = set_affinity_irq,
2063
2064         .irq_retrigger          = retrigger_dynirq,
2065 };
2066
2067 static struct irq_chip xen_percpu_chip __read_mostly = {
2068         .name                   = "xen-percpu",
2069
2070         .irq_disable            = disable_dynirq,
2071         .irq_mask               = disable_dynirq,
2072         .irq_unmask             = enable_dynirq,
2073
2074         .irq_ack                = ack_dynirq,
2075 };
2076
2077 #ifdef CONFIG_XEN_PVHVM
2078 /* Vector callbacks are better than PCI interrupts to receive event
2079  * channel notifications because we can receive vector callbacks on any
2080  * vcpu and we don't need PCI support or APIC interactions. */
2081 void xen_callback_vector(void)
2082 {
2083         int rc;
2084         uint64_t callback_via;
2085
2086         if (xen_have_vector_callback) {
2087                 callback_via = HVM_CALLBACK_VECTOR(HYPERVISOR_CALLBACK_VECTOR);
2088                 rc = xen_set_callback_via(callback_via);
2089                 if (rc) {
2090                         pr_err("Request for Xen HVM callback vector failed\n");
2091                         xen_have_vector_callback = 0;
2092                         return;
2093                 }
2094                 pr_info_once("Xen HVM callback vector for event delivery is enabled\n");
2095                 alloc_intr_gate(HYPERVISOR_CALLBACK_VECTOR,
2096                                 xen_hvm_callback_vector);
2097         }
2098 }
2099 #else
2100 void xen_callback_vector(void) {}
2101 #endif
2102
2103 static bool fifo_events = true;
2104 module_param(fifo_events, bool, 0);
2105
2106 static int xen_evtchn_cpu_prepare(unsigned int cpu)
2107 {
2108         int ret = 0;
2109
2110         xen_cpu_init_eoi(cpu);
2111
2112         if (evtchn_ops->percpu_init)
2113                 ret = evtchn_ops->percpu_init(cpu);
2114
2115         return ret;
2116 }
2117
2118 static int xen_evtchn_cpu_dead(unsigned int cpu)
2119 {
2120         int ret = 0;
2121
2122         if (evtchn_ops->percpu_deinit)
2123                 ret = evtchn_ops->percpu_deinit(cpu);
2124
2125         return ret;
2126 }
2127
2128 void __init xen_init_IRQ(void)
2129 {
2130         int ret = -EINVAL;
2131         unsigned int evtchn;
2132
2133         if (fifo_events)
2134                 ret = xen_evtchn_fifo_init();
2135         if (ret < 0)
2136                 xen_evtchn_2l_init();
2137
2138         xen_cpu_init_eoi(smp_processor_id());
2139
2140         cpuhp_setup_state_nocalls(CPUHP_XEN_EVTCHN_PREPARE,
2141                                   "xen/evtchn:prepare",
2142                                   xen_evtchn_cpu_prepare, xen_evtchn_cpu_dead);
2143
2144         evtchn_to_irq = kcalloc(EVTCHN_ROW(xen_evtchn_max_channels()),
2145                                 sizeof(*evtchn_to_irq), GFP_KERNEL);
2146         BUG_ON(!evtchn_to_irq);
2147
2148         /* No event channels are 'live' right now. */
2149         for (evtchn = 0; evtchn < xen_evtchn_nr_channels(); evtchn++)
2150                 mask_evtchn(evtchn);
2151
2152         pirq_needs_eoi = pirq_needs_eoi_flag;
2153
2154 #ifdef CONFIG_X86
2155         if (xen_pv_domain()) {
2156                 irq_ctx_init(smp_processor_id());
2157                 if (xen_initial_domain())
2158                         pci_xen_initial_domain();
2159         }
2160         if (xen_feature(XENFEAT_hvm_callback_vector))
2161                 xen_callback_vector();
2162
2163         if (xen_hvm_domain()) {
2164                 native_init_IRQ();
2165                 /* pci_xen_hvm_init must be called after native_init_IRQ so that
2166                  * __acpi_register_gsi can point at the right function */
2167                 pci_xen_hvm_init();
2168         } else {
2169                 int rc;
2170                 struct physdev_pirq_eoi_gmfn eoi_gmfn;
2171
2172                 pirq_eoi_map = (void *)__get_free_page(GFP_KERNEL|__GFP_ZERO);
2173                 eoi_gmfn.gmfn = virt_to_gfn(pirq_eoi_map);
2174                 rc = HYPERVISOR_physdev_op(PHYSDEVOP_pirq_eoi_gmfn_v2, &eoi_gmfn);
2175                 if (rc != 0) {
2176                         free_page((unsigned long) pirq_eoi_map);
2177                         pirq_eoi_map = NULL;
2178                 } else
2179                         pirq_needs_eoi = pirq_check_eoi_map;
2180         }
2181 #endif
2182 }