arch/x86/kvm/ioapic.c

   1 /*
   2  *  Copyright (C) 2001  MandrakeSoft S.A.
   3  *  Copyright 2010 Red Hat, Inc. and/or its affiliates.
   4  *
   5  *    MandrakeSoft S.A.
   6  *    43, rue d'Aboukir
   7  *    75002 Paris - France
   8  *    http://www.linux-mandrake.com/
   9  *    http://www.mandrakesoft.com/
  10  *
  11  *  This library is free software; you can redistribute it and/or
  12  *  modify it under the terms of the GNU Lesser General Public
  13  *  License as published by the Free Software Foundation; either
  14  *  version 2 of the License, or (at your option) any later version.
  15  *
  16  *  This library is distributed in the hope that it will be useful,
  17  *  but WITHOUT ANY WARRANTY; without even the implied warranty of
  18  *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  19  *  Lesser General Public License for more details.
  20  *
  21  *  You should have received a copy of the GNU Lesser General Public
  22  *  License along with this library; if not, write to the Free Software
  23  *  Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307 USA
  24  *
  25  *  Yunhong Jiang <yunhong.jiang@intel.com>
  26  *  Yaozu (Eddie) Dong <eddie.dong@intel.com>
  27  *  Based on Xen 3.1 code.
  28  */
  29
  30 #include <linux/kvm_host.h>
  31 #include <linux/kvm.h>
  32 #include <linux/mm.h>
  33 #include <linux/highmem.h>
  34 #include <linux/smp.h>
  35 #include <linux/hrtimer.h>
  36 #include <linux/io.h>
  37 #include <linux/slab.h>
  38 #include <linux/export.h>
  39 #include <asm/processor.h>
  40 #include <asm/page.h>
  41 #include <asm/current.h>
  42 #include <trace/events/kvm.h>
  43
  44 #include "ioapic.h"
  45 #include "lapic.h"
  46 #include "irq.h"
  47
  48 #if 0
  49 #define ioapic_debug(fmt,arg...) printk(KERN_WARNING fmt,##arg)
  50 #else
  51 #define ioapic_debug(fmt, arg...)
  52 #endif
  53 static int ioapic_service(struct kvm_ioapic *vioapic, int irq,
  54                 bool line_status);
  55
  56 static unsigned long ioapic_read_indirect(struct kvm_ioapic *ioapic,
  57                                           unsigned long addr,
  58                                           unsigned long length)
  59 {
  60         unsigned long result = 0;
  61
  62         switch (ioapic->ioregsel) {
  63         case IOAPIC_REG_VERSION:
  64                 result = ((((IOAPIC_NUM_PINS - 1) & 0xff) << 16)
  65                           | (IOAPIC_VERSION_ID & 0xff));
  66                 break;
  67
  68         case IOAPIC_REG_APIC_ID:
  69         case IOAPIC_REG_ARB_ID:
  70                 result = ((ioapic->id & 0xf) << 24);
  71                 break;
  72
  73         default:
  74                 {
  75                         u32 redir_index = (ioapic->ioregsel - 0x10) >> 1;
  76                         u64 redir_content;
  77
  78                         if (redir_index < IOAPIC_NUM_PINS)
  79                                 redir_content =
  80                                         ioapic->redirtbl[redir_index].bits;
  81                         else
  82                                 redir_content = ~0ULL;
  83
  84                         result = (ioapic->ioregsel & 0x1) ?
  85                             (redir_content >> 32) & 0xffffffff :
  86                             redir_content & 0xffffffff;
  87                         break;
  88                 }
  89         }
  90
  91         return result;
  92 }
  93
  94 static void rtc_irq_eoi_tracking_reset(struct kvm_ioapic *ioapic)
  95 {
  96         ioapic->rtc_status.pending_eoi = 0;
  97         bitmap_zero(ioapic->rtc_status.dest_map.map, KVM_MAX_VCPU_ID);
  98 }
  99
 100 static void kvm_rtc_eoi_tracking_restore_all(struct kvm_ioapic *ioapic);
 101
 102 static void rtc_status_pending_eoi_check_valid(struct kvm_ioapic *ioapic)
 103 {
 104         if (WARN_ON(ioapic->rtc_status.pending_eoi < 0))
 105                 kvm_rtc_eoi_tracking_restore_all(ioapic);
 106 }
 107
 108 static void __rtc_irq_eoi_tracking_restore_one(struct kvm_vcpu *vcpu)
 109 {
 110         bool new_val, old_val;
 111         struct kvm_ioapic *ioapic = vcpu->kvm->arch.vioapic;
 112         struct dest_map *dest_map = &ioapic->rtc_status.dest_map;
 113         union kvm_ioapic_redirect_entry *e;
 114
 115         e = &ioapic->redirtbl[RTC_GSI];
 116         if (!kvm_apic_match_dest(vcpu, NULL, 0, e->fields.dest_id,
 117                                 e->fields.dest_mode))
 118                 return;
 119
 120         new_val = kvm_apic_pending_eoi(vcpu, e->fields.vector);
 121         old_val = test_bit(vcpu->vcpu_id, dest_map->map);
 122
 123         if (new_val == old_val)
 124                 return;
 125
 126         if (new_val) {
 127                 __set_bit(vcpu->vcpu_id, dest_map->map);
 128                 dest_map->vectors[vcpu->vcpu_id] = e->fields.vector;
 129                 ioapic->rtc_status.pending_eoi++;
 130         } else {
 131                 __clear_bit(vcpu->vcpu_id, dest_map->map);
 132                 ioapic->rtc_status.pending_eoi--;
 133                 rtc_status_pending_eoi_check_valid(ioapic);
 134         }
 135 }
 136
 137 void kvm_rtc_eoi_tracking_restore_one(struct kvm_vcpu *vcpu)
 138 {
 139         struct kvm_ioapic *ioapic = vcpu->kvm->arch.vioapic;
 140
 141         spin_lock(&ioapic->lock);
 142         __rtc_irq_eoi_tracking_restore_one(vcpu);
 143         spin_unlock(&ioapic->lock);
 144 }
 145
 146 static void kvm_rtc_eoi_tracking_restore_all(struct kvm_ioapic *ioapic)
 147 {
 148         struct kvm_vcpu *vcpu;
 149         int i;
 150
 151         if (RTC_GSI >= IOAPIC_NUM_PINS)
 152                 return;
 153
 154         rtc_irq_eoi_tracking_reset(ioapic);
 155         kvm_for_each_vcpu(i, vcpu, ioapic->kvm)
 156             __rtc_irq_eoi_tracking_restore_one(vcpu);
 157 }
 158
 159 static void rtc_irq_eoi(struct kvm_ioapic *ioapic, struct kvm_vcpu *vcpu)
 160 {
 161         if (test_and_clear_bit(vcpu->vcpu_id,
 162                                ioapic->rtc_status.dest_map.map)) {
 163                 --ioapic->rtc_status.pending_eoi;
 164                 rtc_status_pending_eoi_check_valid(ioapic);
 165         }
 166 }
 167
 168 static bool rtc_irq_check_coalesced(struct kvm_ioapic *ioapic)
 169 {
 170         if (ioapic->rtc_status.pending_eoi > 0)
 171                 return true; /* coalesced */
 172
 173         return false;
 174 }
 175
 176 static int ioapic_set_irq(struct kvm_ioapic *ioapic, unsigned int irq,
 177                 int irq_level, bool line_status)
 178 {
 179         union kvm_ioapic_redirect_entry entry;
 180         u32 mask = 1 << irq;
 181         u32 old_irr;
 182         int edge, ret;
 183
 184         entry = ioapic->redirtbl[irq];
 185         edge = (entry.fields.trig_mode == IOAPIC_EDGE_TRIG);
 186
 187         if (!irq_level) {
 188                 ioapic->irr &= ~mask;
 189                 ret = 1;
 190                 goto out;
 191         }
 192
 193         /*
 194          * Return 0 for coalesced interrupts; for edge-triggered interrupts,
 195          * this only happens if a previous edge has not been delivered due
 196          * do masking.  For level interrupts, the remote_irr field tells
 197          * us if the interrupt is waiting for an EOI.
 198          *
 199          * RTC is special: it is edge-triggered, but userspace likes to know
 200          * if it has been already ack-ed via EOI because coalesced RTC
 201          * interrupts lead to time drift in Windows guests.  So we track
 202          * EOI manually for the RTC interrupt.
 203          */
 204         if (irq == RTC_GSI && line_status &&
 205                 rtc_irq_check_coalesced(ioapic)) {
 206                 ret = 0;
 207                 goto out;
 208         }
 209
 210         old_irr = ioapic->irr;
 211         ioapic->irr |= mask;
 212         if (edge)
 213                 ioapic->irr_delivered &= ~mask;
 214         if ((edge && old_irr == ioapic->irr) ||
 215             (!edge && entry.fields.remote_irr)) {
 216                 ret = 0;
 217                 goto out;
 218         }
 219
 220         ret = ioapic_service(ioapic, irq, line_status);
 221
 222 out:
 223         trace_kvm_ioapic_set_irq(entry.bits, irq, ret == 0);
 224         return ret;
 225 }
 226
 227 static void kvm_ioapic_inject_all(struct kvm_ioapic *ioapic, unsigned long irr)
 228 {
 229         u32 idx;
 230
 231         rtc_irq_eoi_tracking_reset(ioapic);
 232         for_each_set_bit(idx, &irr, IOAPIC_NUM_PINS)
 233                 ioapic_set_irq(ioapic, idx, 1, true);
 234
 235         kvm_rtc_eoi_tracking_restore_all(ioapic);
 236 }
 237
 238
 239 void kvm_ioapic_scan_entry(struct kvm_vcpu *vcpu, ulong *ioapic_handled_vectors)
 240 {
 241         struct kvm_ioapic *ioapic = vcpu->kvm->arch.vioapic;
 242         struct dest_map *dest_map = &ioapic->rtc_status.dest_map;
 243         union kvm_ioapic_redirect_entry *e;
 244         int index;
 245
 246         spin_lock(&ioapic->lock);
 247
 248         /* Make sure we see any missing RTC EOI */
 249         if (test_bit(vcpu->vcpu_id, dest_map->map))
 250                 __set_bit(dest_map->vectors[vcpu->vcpu_id],
 251                           ioapic_handled_vectors);
 252
 253         for (index = 0; index < IOAPIC_NUM_PINS; index++) {
 254                 e = &ioapic->redirtbl[index];
 255                 if (e->fields.trig_mode == IOAPIC_LEVEL_TRIG ||
 256                     kvm_irq_has_notifier(ioapic->kvm, KVM_IRQCHIP_IOAPIC, index) ||
 257                     index == RTC_GSI) {
 258                         if (kvm_apic_match_dest(vcpu, NULL, 0,
 259                                      e->fields.dest_id, e->fields.dest_mode) ||
 260                             (e->fields.trig_mode == IOAPIC_EDGE_TRIG &&
 261                              kvm_apic_pending_eoi(vcpu, e->fields.vector)))
 262                                 __set_bit(e->fields.vector,
 263                                           ioapic_handled_vectors);
 264                 }
 265         }
 266         spin_unlock(&ioapic->lock);
 267 }
 268
 269 void kvm_vcpu_request_scan_ioapic(struct kvm *kvm)
 270 {
 271         struct kvm_ioapic *ioapic = kvm->arch.vioapic;
 272
 273         if (!ioapic)
 274                 return;
 275         kvm_make_scan_ioapic_request(kvm);
 276 }
 277
 278 static void ioapic_write_indirect(struct kvm_ioapic *ioapic, u32 val)
 279 {
 280         unsigned index;
 281         bool mask_before, mask_after;
 282         union kvm_ioapic_redirect_entry *e;
 283
 284         switch (ioapic->ioregsel) {
 285         case IOAPIC_REG_VERSION:
 286                 /* Writes are ignored. */
 287                 break;
 288
 289         case IOAPIC_REG_APIC_ID:
 290                 ioapic->id = (val >> 24) & 0xf;
 291                 break;
 292
 293         case IOAPIC_REG_ARB_ID:
 294                 break;
 295
 296         default:
 297                 index = (ioapic->ioregsel - 0x10) >> 1;
 298
 299                 ioapic_debug("change redir index %x val %x\n", index, val);
 300                 if (index >= IOAPIC_NUM_PINS)
 301                         return;
 302                 e = &ioapic->redirtbl[index];
 303                 mask_before = e->fields.mask;
 304                 if (ioapic->ioregsel & 1) {
 305                         e->bits &= 0xffffffff;
 306                         e->bits |= (u64) val << 32;
 307                 } else {
 308                         e->bits &= ~0xffffffffULL;
 309                         e->bits |= (u32) val;
 310                         e->fields.remote_irr = 0;
 311                 }
 312                 mask_after = e->fields.mask;
 313                 if (mask_before != mask_after)
 314                         kvm_fire_mask_notifiers(ioapic->kvm, KVM_IRQCHIP_IOAPIC, index, mask_after);
 315                 if (e->fields.trig_mode == IOAPIC_LEVEL_TRIG
 316                     && ioapic->irr & (1 << index))
 317                         ioapic_service(ioapic, index, false);
 318                 kvm_vcpu_request_scan_ioapic(ioapic->kvm);
 319                 break;
 320         }
 321 }
 322
 323 static int ioapic_service(struct kvm_ioapic *ioapic, int irq, bool line_status)
 324 {
 325         union kvm_ioapic_redirect_entry *entry = &ioapic->redirtbl[irq];
 326         struct kvm_lapic_irq irqe;
 327         int ret;
 328
 329         if (entry->fields.mask)
 330                 return -1;
 331
 332         ioapic_debug("dest=%x dest_mode=%x delivery_mode=%x "
 333                      "vector=%x trig_mode=%x\n",
 334                      entry->fields.dest_id, entry->fields.dest_mode,
 335                      entry->fields.delivery_mode, entry->fields.vector,
 336                      entry->fields.trig_mode);
 337
 338         irqe.dest_id = entry->fields.dest_id;
 339         irqe.vector = entry->fields.vector;
 340         irqe.dest_mode = entry->fields.dest_mode;
 341         irqe.trig_mode = entry->fields.trig_mode;
 342         irqe.delivery_mode = entry->fields.delivery_mode << 8;
 343         irqe.level = 1;
 344         irqe.shorthand = 0;
 345         irqe.msi_redir_hint = false;
 346
 347         if (irqe.trig_mode == IOAPIC_EDGE_TRIG)
 348                 ioapic->irr_delivered |= 1 << irq;
 349
 350         if (irq == RTC_GSI && line_status) {
 351                 /*
 352                  * pending_eoi cannot ever become negative (see
 353                  * rtc_status_pending_eoi_check_valid) and the caller
 354                  * ensures that it is only called if it is >= zero, namely
 355                  * if rtc_irq_check_coalesced returns false).
 356                  */
 357                 BUG_ON(ioapic->rtc_status.pending_eoi != 0);
 358                 ret = kvm_irq_delivery_to_apic(ioapic->kvm, NULL, &irqe,
 359                                                &ioapic->rtc_status.dest_map);
 360                 ioapic->rtc_status.pending_eoi = (ret < 0 ? 0 : ret);
 361         } else
 362                 ret = kvm_irq_delivery_to_apic(ioapic->kvm, NULL, &irqe, NULL);
 363
 364         if (ret && irqe.trig_mode == IOAPIC_LEVEL_TRIG)
 365                 entry->fields.remote_irr = 1;
 366
 367         return ret;
 368 }
 369
 370 int kvm_ioapic_set_irq(struct kvm_ioapic *ioapic, int irq, int irq_source_id,
 371                        int level, bool line_status)
 372 {
 373         int ret, irq_level;
 374
 375         BUG_ON(irq < 0 || irq >= IOAPIC_NUM_PINS);
 376
 377         spin_lock(&ioapic->lock);
 378         irq_level = __kvm_irq_line_state(&ioapic->irq_states[irq],
 379                                          irq_source_id, level);
 380         ret = ioapic_set_irq(ioapic, irq, irq_level, line_status);
 381
 382         spin_unlock(&ioapic->lock);
 383
 384         return ret;
 385 }
 386
 387 void kvm_ioapic_clear_all(struct kvm_ioapic *ioapic, int irq_source_id)
 388 {
 389         int i;
 390
 391         spin_lock(&ioapic->lock);
 392         for (i = 0; i < KVM_IOAPIC_NUM_PINS; i++)
 393                 __clear_bit(irq_source_id, &ioapic->irq_states[i]);
 394         spin_unlock(&ioapic->lock);
 395 }
 396
 397 static void kvm_ioapic_eoi_inject_work(struct work_struct *work)
 398 {
 399         int i;
 400         struct kvm_ioapic *ioapic = container_of(work, struct kvm_ioapic,
 401                                                  eoi_inject.work);
 402         spin_lock(&ioapic->lock);
 403         for (i = 0; i < IOAPIC_NUM_PINS; i++) {
 404                 union kvm_ioapic_redirect_entry *ent = &ioapic->redirtbl[i];
 405
 406                 if (ent->fields.trig_mode != IOAPIC_LEVEL_TRIG)
 407                         continue;
 408
 409                 if (ioapic->irr & (1 << i) && !ent->fields.remote_irr)
 410                         ioapic_service(ioapic, i, false);
 411         }
 412         spin_unlock(&ioapic->lock);
 413 }
 414
 415 #define IOAPIC_SUCCESSIVE_IRQ_MAX_COUNT 10000
 416
 417 static void __kvm_ioapic_update_eoi(struct kvm_vcpu *vcpu,
 418                         struct kvm_ioapic *ioapic, int vector, int trigger_mode)
 419 {
 420         struct dest_map *dest_map = &ioapic->rtc_status.dest_map;
 421         struct kvm_lapic *apic = vcpu->arch.apic;
 422         int i;
 423
 424         /* RTC special handling */
 425         if (test_bit(vcpu->vcpu_id, dest_map->map) &&
 426             vector == dest_map->vectors[vcpu->vcpu_id])
 427                 rtc_irq_eoi(ioapic, vcpu);
 428
 429         for (i = 0; i < IOAPIC_NUM_PINS; i++) {
 430                 union kvm_ioapic_redirect_entry *ent = &ioapic->redirtbl[i];
 431
 432                 if (ent->fields.vector != vector)
 433                         continue;
 434
 435                 /*
 436                  * We are dropping lock while calling ack notifiers because ack
 437                  * notifier callbacks for assigned devices call into IOAPIC
 438                  * recursively. Since remote_irr is cleared only after call
 439                  * to notifiers if the same vector will be delivered while lock
 440                  * is dropped it will be put into irr and will be delivered
 441                  * after ack notifier returns.
 442                  */
 443                 spin_unlock(&ioapic->lock);
 444                 kvm_notify_acked_irq(ioapic->kvm, KVM_IRQCHIP_IOAPIC, i);
 445                 spin_lock(&ioapic->lock);
 446
 447                 if (trigger_mode != IOAPIC_LEVEL_TRIG ||
 448                     kvm_lapic_get_reg(apic, APIC_SPIV) & APIC_SPIV_DIRECTED_EOI)
 449                         continue;
 450
 451                 ASSERT(ent->fields.trig_mode == IOAPIC_LEVEL_TRIG);
 452                 ent->fields.remote_irr = 0;
 453                 if (!ent->fields.mask && (ioapic->irr & (1 << i))) {
 454                         ++ioapic->irq_eoi[i];
 455                         if (ioapic->irq_eoi[i] == IOAPIC_SUCCESSIVE_IRQ_MAX_COUNT) {
 456                                 /*
 457                                  * Real hardware does not deliver the interrupt
 458                                  * immediately during eoi broadcast, and this
 459                                  * lets a buggy guest make slow progress
 460                                  * even if it does not correctly handle a
 461                                  * level-triggered interrupt.  Emulate this
 462                                  * behavior if we detect an interrupt storm.
 463                                  */
 464                                 schedule_delayed_work(&ioapic->eoi_inject, HZ / 100);
 465                                 ioapic->irq_eoi[i] = 0;
 466                                 trace_kvm_ioapic_delayed_eoi_inj(ent->bits);
 467                         } else {
 468                                 ioapic_service(ioapic, i, false);
 469                         }
 470                 } else {
 471                         ioapic->irq_eoi[i] = 0;
 472                 }
 473         }
 474 }
 475
 476 void kvm_ioapic_update_eoi(struct kvm_vcpu *vcpu, int vector, int trigger_mode)
 477 {
 478         struct kvm_ioapic *ioapic = vcpu->kvm->arch.vioapic;
 479
 480         spin_lock(&ioapic->lock);
 481         __kvm_ioapic_update_eoi(vcpu, ioapic, vector, trigger_mode);
 482         spin_unlock(&ioapic->lock);
 483 }
 484
 485 static inline struct kvm_ioapic *to_ioapic(struct kvm_io_device *dev)
 486 {
 487         return container_of(dev, struct kvm_ioapic, dev);
 488 }
 489
 490 static inline int ioapic_in_range(struct kvm_ioapic *ioapic, gpa_t addr)
 491 {
 492         return ((addr >= ioapic->base_address &&
 493                  (addr < ioapic->base_address + IOAPIC_MEM_LENGTH)));
 494 }
 495
 496 static int ioapic_mmio_read(struct kvm_vcpu *vcpu, struct kvm_io_device *this,
 497                                 gpa_t addr, int len, void *val)
 498 {
 499         struct kvm_ioapic *ioapic = to_ioapic(this);
 500         u32 result;
 501         if (!ioapic_in_range(ioapic, addr))
 502                 return -EOPNOTSUPP;
 503
 504         ioapic_debug("addr %lx\n", (unsigned long)addr);
 505         ASSERT(!(addr & 0xf));  /* check alignment */
 506
 507         addr &= 0xff;
 508         spin_lock(&ioapic->lock);
 509         switch (addr) {
 510         case IOAPIC_REG_SELECT:
 511                 result = ioapic->ioregsel;
 512                 break;
 513
 514         case IOAPIC_REG_WINDOW:
 515                 result = ioapic_read_indirect(ioapic, addr, len);
 516                 break;
 517
 518         default:
 519                 result = 0;
 520                 break;
 521         }
 522         spin_unlock(&ioapic->lock);
 523
 524         switch (len) {
 525         case 8:
 526                 *(u64 *) val = result;
 527                 break;
 528         case 1:
 529         case 2:
 530         case 4:
 531                 memcpy(val, (char *)&result, len);
 532                 break;
 533         default:
 534                 printk(KERN_WARNING "ioapic: wrong length %d\n", len);
 535         }
 536         return 0;
 537 }
 538
 539 static int ioapic_mmio_write(struct kvm_vcpu *vcpu, struct kvm_io_device *this,
 540                                  gpa_t addr, int len, const void *val)
 541 {
 542         struct kvm_ioapic *ioapic = to_ioapic(this);
 543         u32 data;
 544         if (!ioapic_in_range(ioapic, addr))
 545                 return -EOPNOTSUPP;
 546
 547         ioapic_debug("ioapic_mmio_write addr=%p len=%d val=%p\n",
 548                      (void*)addr, len, val);
 549         ASSERT(!(addr & 0xf));  /* check alignment */
 550
 551         switch (len) {
 552         case 8:
 553         case 4:
 554                 data = *(u32 *) val;
 555                 break;
 556         case 2:
 557                 data = *(u16 *) val;
 558                 break;
 559         case 1:
 560                 data = *(u8  *) val;
 561                 break;
 562         default:
 563                 printk(KERN_WARNING "ioapic: Unsupported size %d\n", len);
 564                 return 0;
 565         }
 566
 567         addr &= 0xff;
 568         spin_lock(&ioapic->lock);
 569         switch (addr) {
 570         case IOAPIC_REG_SELECT:
 571                 ioapic->ioregsel = data & 0xFF; /* 8-bit register */
 572                 break;
 573
 574         case IOAPIC_REG_WINDOW:
 575                 ioapic_write_indirect(ioapic, data);
 576                 break;
 577
 578         default:
 579                 break;
 580         }
 581         spin_unlock(&ioapic->lock);
 582         return 0;
 583 }
 584
 585 static void kvm_ioapic_reset(struct kvm_ioapic *ioapic)
 586 {
 587         int i;
 588
 589         cancel_delayed_work_sync(&ioapic->eoi_inject);
 590         for (i = 0; i < IOAPIC_NUM_PINS; i++)
 591                 ioapic->redirtbl[i].fields.mask = 1;
 592         ioapic->base_address = IOAPIC_DEFAULT_BASE_ADDRESS;
 593         ioapic->ioregsel = 0;
 594         ioapic->irr = 0;
 595         ioapic->irr_delivered = 0;
 596         ioapic->id = 0;
 597         memset(ioapic->irq_eoi, 0x00, sizeof(ioapic->irq_eoi));
 598         rtc_irq_eoi_tracking_reset(ioapic);
 599 }
 600
 601 static const struct kvm_io_device_ops ioapic_mmio_ops = {
 602         .read     = ioapic_mmio_read,
 603         .write    = ioapic_mmio_write,
 604 };
 605
 606 int kvm_ioapic_init(struct kvm *kvm)
 607 {
 608         struct kvm_ioapic *ioapic;
 609         int ret;
 610
 611         ioapic = kzalloc(sizeof(struct kvm_ioapic), GFP_KERNEL);
 612         if (!ioapic)
 613                 return -ENOMEM;
 614         spin_lock_init(&ioapic->lock);
 615         INIT_DELAYED_WORK(&ioapic->eoi_inject, kvm_ioapic_eoi_inject_work);
 616         kvm->arch.vioapic = ioapic;
 617         kvm_ioapic_reset(ioapic);
 618         kvm_iodevice_init(&ioapic->dev, &ioapic_mmio_ops);
 619         ioapic->kvm = kvm;
 620         mutex_lock(&kvm->slots_lock);
 621         ret = kvm_io_bus_register_dev(kvm, KVM_MMIO_BUS, ioapic->base_address,
 622                                       IOAPIC_MEM_LENGTH, &ioapic->dev);
 623         mutex_unlock(&kvm->slots_lock);
 624         if (ret < 0) {
 625                 kvm->arch.vioapic = NULL;
 626                 kfree(ioapic);
 627                 return ret;
 628         }
 629
 630         kvm_vcpu_request_scan_ioapic(kvm);
 631         return ret;
 632 }
 633
 634 void kvm_ioapic_destroy(struct kvm *kvm)
 635 {
 636         struct kvm_ioapic *ioapic = kvm->arch.vioapic;
 637
 638         if (!ioapic)
 639                 return;
 640
 641         cancel_delayed_work_sync(&ioapic->eoi_inject);
 642         kvm_io_bus_unregister_dev(kvm, KVM_MMIO_BUS, &ioapic->dev);
 643         kvm->arch.vioapic = NULL;
 644         kfree(ioapic);
 645 }
 646
 647 int kvm_get_ioapic(struct kvm *kvm, struct kvm_ioapic_state *state)
 648 {
 649         struct kvm_ioapic *ioapic = ioapic_irqchip(kvm);
 650         if (!ioapic)
 651                 return -EINVAL;
 652
 653         spin_lock(&ioapic->lock);
 654         memcpy(state, ioapic, sizeof(struct kvm_ioapic_state));
 655         state->irr &= ~ioapic->irr_delivered;
 656         spin_unlock(&ioapic->lock);
 657         return 0;
 658 }
 659
 660 int kvm_set_ioapic(struct kvm *kvm, struct kvm_ioapic_state *state)
 661 {
 662         struct kvm_ioapic *ioapic = ioapic_irqchip(kvm);
 663         if (!ioapic)
 664                 return -EINVAL;
 665
 666         spin_lock(&ioapic->lock);
 667         memcpy(ioapic, state, sizeof(struct kvm_ioapic_state));
 668         ioapic->irr = 0;
 669         ioapic->irr_delivered = 0;
 670         kvm_vcpu_request_scan_ioapic(kvm);
 671         kvm_ioapic_inject_all(ioapic, state->irr);
 672         spin_unlock(&ioapic->lock);
 673         return 0;
 674 }