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 <linux/nospec.h>
  40 #include <asm/processor.h>
  41 #include <asm/page.h>
  42 #include <asm/current.h>
  43 #include <trace/events/kvm.h>
  44
  45 #include "ioapic.h"
  46 #include "lapic.h"
  47 #include "irq.h"
  48
  49 #if 0
  50 #define ioapic_debug(fmt,arg...) printk(KERN_WARNING fmt,##arg)
  51 #else
  52 #define ioapic_debug(fmt, arg...)
  53 #endif
  54 static int ioapic_service(struct kvm_ioapic *vioapic, int irq,
  55                 bool line_status);
  56
  57 static unsigned long ioapic_read_indirect(struct kvm_ioapic *ioapic,
  58                                           unsigned long addr,
  59                                           unsigned long length)
  60 {
  61         unsigned long result = 0;
  62
  63         switch (ioapic->ioregsel) {
  64         case IOAPIC_REG_VERSION:
  65                 result = ((((IOAPIC_NUM_PINS - 1) & 0xff) << 16)
  66                           | (IOAPIC_VERSION_ID & 0xff));
  67                 break;
  68
  69         case IOAPIC_REG_APIC_ID:
  70         case IOAPIC_REG_ARB_ID:
  71                 result = ((ioapic->id & 0xf) << 24);
  72                 break;
  73
  74         default:
  75                 {
  76                         u32 redir_index = (ioapic->ioregsel - 0x10) >> 1;
  77                         u64 redir_content = ~0ULL;
  78
  79                         if (redir_index < IOAPIC_NUM_PINS) {
  80                                 u32 index = array_index_nospec(
  81                                         redir_index, IOAPIC_NUM_PINS);
  82
  83                                 redir_content = ioapic->redirtbl[index].bits;
  84                         }
  85
  86                         result = (ioapic->ioregsel & 0x1) ?
  87                             (redir_content >> 32) & 0xffffffff :
  88                             redir_content & 0xffffffff;
  89                         break;
  90                 }
  91         }
  92
  93         return result;
  94 }
  95
  96 static void rtc_irq_eoi_tracking_reset(struct kvm_ioapic *ioapic)
  97 {
  98         ioapic->rtc_status.pending_eoi = 0;
  99         bitmap_zero(ioapic->rtc_status.dest_map.map, KVM_MAX_VCPU_ID);
 100 }
 101
 102 static void kvm_rtc_eoi_tracking_restore_all(struct kvm_ioapic *ioapic);
 103
 104 static void rtc_status_pending_eoi_check_valid(struct kvm_ioapic *ioapic)
 105 {
 106         if (WARN_ON(ioapic->rtc_status.pending_eoi < 0))
 107                 kvm_rtc_eoi_tracking_restore_all(ioapic);
 108 }
 109
 110 static void __rtc_irq_eoi_tracking_restore_one(struct kvm_vcpu *vcpu)
 111 {
 112         bool new_val, old_val;
 113         struct kvm_ioapic *ioapic = vcpu->kvm->arch.vioapic;
 114         struct dest_map *dest_map = &ioapic->rtc_status.dest_map;
 115         union kvm_ioapic_redirect_entry *e;
 116
 117         e = &ioapic->redirtbl[RTC_GSI];
 118         if (!kvm_apic_match_dest(vcpu, NULL, 0, e->fields.dest_id,
 119                                 e->fields.dest_mode))
 120                 return;
 121
 122         new_val = kvm_apic_pending_eoi(vcpu, e->fields.vector);
 123         old_val = test_bit(vcpu->vcpu_id, dest_map->map);
 124
 125         if (new_val == old_val)
 126                 return;
 127
 128         if (new_val) {
 129                 __set_bit(vcpu->vcpu_id, dest_map->map);
 130                 dest_map->vectors[vcpu->vcpu_id] = e->fields.vector;
 131                 ioapic->rtc_status.pending_eoi++;
 132         } else {
 133                 __clear_bit(vcpu->vcpu_id, dest_map->map);
 134                 ioapic->rtc_status.pending_eoi--;
 135                 rtc_status_pending_eoi_check_valid(ioapic);
 136         }
 137 }
 138
 139 void kvm_rtc_eoi_tracking_restore_one(struct kvm_vcpu *vcpu)
 140 {
 141         struct kvm_ioapic *ioapic = vcpu->kvm->arch.vioapic;
 142
 143         spin_lock(&ioapic->lock);
 144         __rtc_irq_eoi_tracking_restore_one(vcpu);
 145         spin_unlock(&ioapic->lock);
 146 }
 147
 148 static void kvm_rtc_eoi_tracking_restore_all(struct kvm_ioapic *ioapic)
 149 {
 150         struct kvm_vcpu *vcpu;
 151         int i;
 152
 153         if (RTC_GSI >= IOAPIC_NUM_PINS)
 154                 return;
 155
 156         rtc_irq_eoi_tracking_reset(ioapic);
 157         kvm_for_each_vcpu(i, vcpu, ioapic->kvm)
 158             __rtc_irq_eoi_tracking_restore_one(vcpu);
 159 }
 160
 161 static void rtc_irq_eoi(struct kvm_ioapic *ioapic, struct kvm_vcpu *vcpu)
 162 {
 163         if (test_and_clear_bit(vcpu->vcpu_id,
 164                                ioapic->rtc_status.dest_map.map)) {
 165                 --ioapic->rtc_status.pending_eoi;
 166                 rtc_status_pending_eoi_check_valid(ioapic);
 167         }
 168 }
 169
 170 static bool rtc_irq_check_coalesced(struct kvm_ioapic *ioapic)
 171 {
 172         if (ioapic->rtc_status.pending_eoi > 0)
 173                 return true; /* coalesced */
 174
 175         return false;
 176 }
 177
 178 static int ioapic_set_irq(struct kvm_ioapic *ioapic, unsigned int irq,
 179                 int irq_level, bool line_status)
 180 {
 181         union kvm_ioapic_redirect_entry entry;
 182         u32 mask = 1 << irq;
 183         u32 old_irr;
 184         int edge, ret;
 185
 186         entry = ioapic->redirtbl[irq];
 187         edge = (entry.fields.trig_mode == IOAPIC_EDGE_TRIG);
 188
 189         if (!irq_level) {
 190                 ioapic->irr &= ~mask;
 191                 ret = 1;
 192                 goto out;
 193         }
 194
 195         /*
 196          * Return 0 for coalesced interrupts; for edge-triggered interrupts,
 197          * this only happens if a previous edge has not been delivered due
 198          * do masking.  For level interrupts, the remote_irr field tells
 199          * us if the interrupt is waiting for an EOI.
 200          *
 201          * RTC is special: it is edge-triggered, but userspace likes to know
 202          * if it has been already ack-ed via EOI because coalesced RTC
 203          * interrupts lead to time drift in Windows guests.  So we track
 204          * EOI manually for the RTC interrupt.
 205          */
 206         if (irq == RTC_GSI && line_status &&
 207                 rtc_irq_check_coalesced(ioapic)) {
 208                 ret = 0;
 209                 goto out;
 210         }
 211
 212         old_irr = ioapic->irr;
 213         ioapic->irr |= mask;
 214         if (edge)
 215                 ioapic->irr_delivered &= ~mask;
 216         if ((edge && old_irr == ioapic->irr) ||
 217             (!edge && entry.fields.remote_irr)) {
 218                 ret = 0;
 219                 goto out;
 220         }
 221
 222         ret = ioapic_service(ioapic, irq, line_status);
 223
 224 out:
 225         trace_kvm_ioapic_set_irq(entry.bits, irq, ret == 0);
 226         return ret;
 227 }
 228
 229 static void kvm_ioapic_inject_all(struct kvm_ioapic *ioapic, unsigned long irr)
 230 {
 231         u32 idx;
 232
 233         rtc_irq_eoi_tracking_reset(ioapic);
 234         for_each_set_bit(idx, &irr, IOAPIC_NUM_PINS)
 235                 ioapic_set_irq(ioapic, idx, 1, true);
 236
 237         kvm_rtc_eoi_tracking_restore_all(ioapic);
 238 }
 239
 240
 241 void kvm_ioapic_scan_entry(struct kvm_vcpu *vcpu, ulong *ioapic_handled_vectors)
 242 {
 243         struct kvm_ioapic *ioapic = vcpu->kvm->arch.vioapic;
 244         struct dest_map *dest_map = &ioapic->rtc_status.dest_map;
 245         union kvm_ioapic_redirect_entry *e;
 246         int index;
 247
 248         spin_lock(&ioapic->lock);
 249
 250         /* Make sure we see any missing RTC EOI */
 251         if (test_bit(vcpu->vcpu_id, dest_map->map))
 252                 __set_bit(dest_map->vectors[vcpu->vcpu_id],
 253                           ioapic_handled_vectors);
 254
 255         for (index = 0; index < IOAPIC_NUM_PINS; index++) {
 256                 e = &ioapic->redirtbl[index];
 257                 if (e->fields.trig_mode == IOAPIC_LEVEL_TRIG ||
 258                     kvm_irq_has_notifier(ioapic->kvm, KVM_IRQCHIP_IOAPIC, index) ||
 259                     index == RTC_GSI) {
 260                         if (kvm_apic_match_dest(vcpu, NULL, 0,
 261                                      e->fields.dest_id, e->fields.dest_mode) ||
 262                             kvm_apic_pending_eoi(vcpu, e->fields.vector))
 263                                 __set_bit(e->fields.vector,
 264                                           ioapic_handled_vectors);
 265                 }
 266         }
 267         spin_unlock(&ioapic->lock);
 268 }
 269
 270 void kvm_arch_post_irq_ack_notifier_list_update(struct kvm *kvm)
 271 {
 272         if (!ioapic_in_kernel(kvm))
 273                 return;
 274         kvm_make_scan_ioapic_request(kvm);
 275 }
 276
 277 static void ioapic_write_indirect(struct kvm_ioapic *ioapic, u32 val)
 278 {
 279         unsigned index;
 280         bool mask_before, mask_after;
 281         int old_remote_irr, old_delivery_status;
 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                 index = array_index_nospec(index, IOAPIC_NUM_PINS);
 303                 e = &ioapic->redirtbl[index];
 304                 mask_before = e->fields.mask;
 305                 /* Preserve read-only fields */
 306                 old_remote_irr = e->fields.remote_irr;
 307                 old_delivery_status = e->fields.delivery_status;
 308                 if (ioapic->ioregsel & 1) {
 309                         e->bits &= 0xffffffff;
 310                         e->bits |= (u64) val << 32;
 311                 } else {
 312                         e->bits &= ~0xffffffffULL;
 313                         e->bits |= (u32) val;
 314                 }
 315                 e->fields.remote_irr = old_remote_irr;
 316                 e->fields.delivery_status = old_delivery_status;
 317
 318                 /*
 319                  * Some OSes (Linux, Xen) assume that Remote IRR bit will
 320                  * be cleared by IOAPIC hardware when the entry is configured
 321                  * as edge-triggered. This behavior is used to simulate an
 322                  * explicit EOI on IOAPICs that don't have the EOI register.
 323                  */
 324                 if (e->fields.trig_mode == IOAPIC_EDGE_TRIG)
 325                         e->fields.remote_irr = 0;
 326
 327                 mask_after = e->fields.mask;
 328                 if (mask_before != mask_after)
 329                         kvm_fire_mask_notifiers(ioapic->kvm, KVM_IRQCHIP_IOAPIC, index, mask_after);
 330                 if (e->fields.trig_mode == IOAPIC_LEVEL_TRIG
 331                     && ioapic->irr & (1 << index))
 332                         ioapic_service(ioapic, index, false);
 333                 kvm_make_scan_ioapic_request(ioapic->kvm);
 334                 break;
 335         }
 336 }
 337
 338 static int ioapic_service(struct kvm_ioapic *ioapic, int irq, bool line_status)
 339 {
 340         union kvm_ioapic_redirect_entry *entry = &ioapic->redirtbl[irq];
 341         struct kvm_lapic_irq irqe;
 342         int ret;
 343
 344         if (entry->fields.mask)
 345                 return -1;
 346
 347         ioapic_debug("dest=%x dest_mode=%x delivery_mode=%x "
 348                      "vector=%x trig_mode=%x\n",
 349                      entry->fields.dest_id, entry->fields.dest_mode,
 350                      entry->fields.delivery_mode, entry->fields.vector,
 351                      entry->fields.trig_mode);
 352
 353         irqe.dest_id = entry->fields.dest_id;
 354         irqe.vector = entry->fields.vector;
 355         irqe.dest_mode = entry->fields.dest_mode;
 356         irqe.trig_mode = entry->fields.trig_mode;
 357         irqe.delivery_mode = entry->fields.delivery_mode << 8;
 358         irqe.level = 1;
 359         irqe.shorthand = 0;
 360         irqe.msi_redir_hint = false;
 361
 362         if (irqe.trig_mode == IOAPIC_EDGE_TRIG)
 363                 ioapic->irr_delivered |= 1 << irq;
 364
 365         if (irq == RTC_GSI && line_status) {
 366                 /*
 367                  * pending_eoi cannot ever become negative (see
 368                  * rtc_status_pending_eoi_check_valid) and the caller
 369                  * ensures that it is only called if it is >= zero, namely
 370                  * if rtc_irq_check_coalesced returns false).
 371                  */
 372                 BUG_ON(ioapic->rtc_status.pending_eoi != 0);
 373                 ret = kvm_irq_delivery_to_apic(ioapic->kvm, NULL, &irqe,
 374                                                &ioapic->rtc_status.dest_map);
 375                 ioapic->rtc_status.pending_eoi = (ret < 0 ? 0 : ret);
 376         } else
 377                 ret = kvm_irq_delivery_to_apic(ioapic->kvm, NULL, &irqe, NULL);
 378
 379         if (ret && irqe.trig_mode == IOAPIC_LEVEL_TRIG)
 380                 entry->fields.remote_irr = 1;
 381
 382         return ret;
 383 }
 384
 385 int kvm_ioapic_set_irq(struct kvm_ioapic *ioapic, int irq, int irq_source_id,
 386                        int level, bool line_status)
 387 {
 388         int ret, irq_level;
 389
 390         BUG_ON(irq < 0 || irq >= IOAPIC_NUM_PINS);
 391
 392         spin_lock(&ioapic->lock);
 393         irq_level = __kvm_irq_line_state(&ioapic->irq_states[irq],
 394                                          irq_source_id, level);
 395         ret = ioapic_set_irq(ioapic, irq, irq_level, line_status);
 396
 397         spin_unlock(&ioapic->lock);
 398
 399         return ret;
 400 }
 401
 402 void kvm_ioapic_clear_all(struct kvm_ioapic *ioapic, int irq_source_id)
 403 {
 404         int i;
 405
 406         spin_lock(&ioapic->lock);
 407         for (i = 0; i < KVM_IOAPIC_NUM_PINS; i++)
 408                 __clear_bit(irq_source_id, &ioapic->irq_states[i]);
 409         spin_unlock(&ioapic->lock);
 410 }
 411
 412 static void kvm_ioapic_eoi_inject_work(struct work_struct *work)
 413 {
 414         int i;
 415         struct kvm_ioapic *ioapic = container_of(work, struct kvm_ioapic,
 416                                                  eoi_inject.work);
 417         spin_lock(&ioapic->lock);
 418         for (i = 0; i < IOAPIC_NUM_PINS; i++) {
 419                 union kvm_ioapic_redirect_entry *ent = &ioapic->redirtbl[i];
 420
 421                 if (ent->fields.trig_mode != IOAPIC_LEVEL_TRIG)
 422                         continue;
 423
 424                 if (ioapic->irr & (1 << i) && !ent->fields.remote_irr)
 425                         ioapic_service(ioapic, i, false);
 426         }
 427         spin_unlock(&ioapic->lock);
 428 }
 429
 430 #define IOAPIC_SUCCESSIVE_IRQ_MAX_COUNT 10000
 431
 432 static void __kvm_ioapic_update_eoi(struct kvm_vcpu *vcpu,
 433                         struct kvm_ioapic *ioapic, int vector, int trigger_mode)
 434 {
 435         struct dest_map *dest_map = &ioapic->rtc_status.dest_map;
 436         struct kvm_lapic *apic = vcpu->arch.apic;
 437         int i;
 438
 439         /* RTC special handling */
 440         if (test_bit(vcpu->vcpu_id, dest_map->map) &&
 441             vector == dest_map->vectors[vcpu->vcpu_id])
 442                 rtc_irq_eoi(ioapic, vcpu);
 443
 444         for (i = 0; i < IOAPIC_NUM_PINS; i++) {
 445                 union kvm_ioapic_redirect_entry *ent = &ioapic->redirtbl[i];
 446
 447                 if (ent->fields.vector != vector)
 448                         continue;
 449
 450                 /*
 451                  * We are dropping lock while calling ack notifiers because ack
 452                  * notifier callbacks for assigned devices call into IOAPIC
 453                  * recursively. Since remote_irr is cleared only after call
 454                  * to notifiers if the same vector will be delivered while lock
 455                  * is dropped it will be put into irr and will be delivered
 456                  * after ack notifier returns.
 457                  */
 458                 spin_unlock(&ioapic->lock);
 459                 kvm_notify_acked_irq(ioapic->kvm, KVM_IRQCHIP_IOAPIC, i);
 460                 spin_lock(&ioapic->lock);
 461
 462                 if (trigger_mode != IOAPIC_LEVEL_TRIG ||
 463                     kvm_lapic_get_reg(apic, APIC_SPIV) & APIC_SPIV_DIRECTED_EOI)
 464                         continue;
 465
 466                 ASSERT(ent->fields.trig_mode == IOAPIC_LEVEL_TRIG);
 467                 ent->fields.remote_irr = 0;
 468                 if (!ent->fields.mask && (ioapic->irr & (1 << i))) {
 469                         ++ioapic->irq_eoi[i];
 470                         if (ioapic->irq_eoi[i] == IOAPIC_SUCCESSIVE_IRQ_MAX_COUNT) {
 471                                 /*
 472                                  * Real hardware does not deliver the interrupt
 473                                  * immediately during eoi broadcast, and this
 474                                  * lets a buggy guest make slow progress
 475                                  * even if it does not correctly handle a
 476                                  * level-triggered interrupt.  Emulate this
 477                                  * behavior if we detect an interrupt storm.
 478                                  */
 479                                 schedule_delayed_work(&ioapic->eoi_inject, HZ / 100);
 480                                 ioapic->irq_eoi[i] = 0;
 481                                 trace_kvm_ioapic_delayed_eoi_inj(ent->bits);
 482                         } else {
 483                                 ioapic_service(ioapic, i, false);
 484                         }
 485                 } else {
 486                         ioapic->irq_eoi[i] = 0;
 487                 }
 488         }
 489 }
 490
 491 void kvm_ioapic_update_eoi(struct kvm_vcpu *vcpu, int vector, int trigger_mode)
 492 {
 493         struct kvm_ioapic *ioapic = vcpu->kvm->arch.vioapic;
 494
 495         spin_lock(&ioapic->lock);
 496         __kvm_ioapic_update_eoi(vcpu, ioapic, vector, trigger_mode);
 497         spin_unlock(&ioapic->lock);
 498 }
 499
 500 static inline struct kvm_ioapic *to_ioapic(struct kvm_io_device *dev)
 501 {
 502         return container_of(dev, struct kvm_ioapic, dev);
 503 }
 504
 505 static inline int ioapic_in_range(struct kvm_ioapic *ioapic, gpa_t addr)
 506 {
 507         return ((addr >= ioapic->base_address &&
 508                  (addr < ioapic->base_address + IOAPIC_MEM_LENGTH)));
 509 }
 510
 511 static int ioapic_mmio_read(struct kvm_vcpu *vcpu, struct kvm_io_device *this,
 512                                 gpa_t addr, int len, void *val)
 513 {
 514         struct kvm_ioapic *ioapic = to_ioapic(this);
 515         u32 result;
 516         if (!ioapic_in_range(ioapic, addr))
 517                 return -EOPNOTSUPP;
 518
 519         ioapic_debug("addr %lx\n", (unsigned long)addr);
 520         ASSERT(!(addr & 0xf));  /* check alignment */
 521
 522         addr &= 0xff;
 523         spin_lock(&ioapic->lock);
 524         switch (addr) {
 525         case IOAPIC_REG_SELECT:
 526                 result = ioapic->ioregsel;
 527                 break;
 528
 529         case IOAPIC_REG_WINDOW:
 530                 result = ioapic_read_indirect(ioapic, addr, len);
 531                 break;
 532
 533         default:
 534                 result = 0;
 535                 break;
 536         }
 537         spin_unlock(&ioapic->lock);
 538
 539         switch (len) {
 540         case 8:
 541                 *(u64 *) val = result;
 542                 break;
 543         case 1:
 544         case 2:
 545         case 4:
 546                 memcpy(val, (char *)&result, len);
 547                 break;
 548         default:
 549                 printk(KERN_WARNING "ioapic: wrong length %d\n", len);
 550         }
 551         return 0;
 552 }
 553
 554 static int ioapic_mmio_write(struct kvm_vcpu *vcpu, struct kvm_io_device *this,
 555                                  gpa_t addr, int len, const void *val)
 556 {
 557         struct kvm_ioapic *ioapic = to_ioapic(this);
 558         u32 data;
 559         if (!ioapic_in_range(ioapic, addr))
 560                 return -EOPNOTSUPP;
 561
 562         ioapic_debug("ioapic_mmio_write addr=%p len=%d val=%p\n",
 563                      (void*)addr, len, val);
 564         ASSERT(!(addr & 0xf));  /* check alignment */
 565
 566         switch (len) {
 567         case 8:
 568         case 4:
 569                 data = *(u32 *) val;
 570                 break;
 571         case 2:
 572                 data = *(u16 *) val;
 573                 break;
 574         case 1:
 575                 data = *(u8  *) val;
 576                 break;
 577         default:
 578                 printk(KERN_WARNING "ioapic: Unsupported size %d\n", len);
 579                 return 0;
 580         }
 581
 582         addr &= 0xff;
 583         spin_lock(&ioapic->lock);
 584         switch (addr) {
 585         case IOAPIC_REG_SELECT:
 586                 ioapic->ioregsel = data & 0xFF; /* 8-bit register */
 587                 break;
 588
 589         case IOAPIC_REG_WINDOW:
 590                 ioapic_write_indirect(ioapic, data);
 591                 break;
 592
 593         default:
 594                 break;
 595         }
 596         spin_unlock(&ioapic->lock);
 597         return 0;
 598 }
 599
 600 static void kvm_ioapic_reset(struct kvm_ioapic *ioapic)
 601 {
 602         int i;
 603
 604         cancel_delayed_work_sync(&ioapic->eoi_inject);
 605         for (i = 0; i < IOAPIC_NUM_PINS; i++)
 606                 ioapic->redirtbl[i].fields.mask = 1;
 607         ioapic->base_address = IOAPIC_DEFAULT_BASE_ADDRESS;
 608         ioapic->ioregsel = 0;
 609         ioapic->irr = 0;
 610         ioapic->irr_delivered = 0;
 611         ioapic->id = 0;
 612         memset(ioapic->irq_eoi, 0x00, sizeof(ioapic->irq_eoi));
 613         rtc_irq_eoi_tracking_reset(ioapic);
 614 }
 615
 616 static const struct kvm_io_device_ops ioapic_mmio_ops = {
 617         .read     = ioapic_mmio_read,
 618         .write    = ioapic_mmio_write,
 619 };
 620
 621 int kvm_ioapic_init(struct kvm *kvm)
 622 {
 623         struct kvm_ioapic *ioapic;
 624         int ret;
 625
 626         ioapic = kzalloc(sizeof(struct kvm_ioapic), GFP_KERNEL);
 627         if (!ioapic)
 628                 return -ENOMEM;
 629         spin_lock_init(&ioapic->lock);
 630         INIT_DELAYED_WORK(&ioapic->eoi_inject, kvm_ioapic_eoi_inject_work);
 631         kvm->arch.vioapic = ioapic;
 632         kvm_ioapic_reset(ioapic);
 633         kvm_iodevice_init(&ioapic->dev, &ioapic_mmio_ops);
 634         ioapic->kvm = kvm;
 635         mutex_lock(&kvm->slots_lock);
 636         ret = kvm_io_bus_register_dev(kvm, KVM_MMIO_BUS, ioapic->base_address,
 637                                       IOAPIC_MEM_LENGTH, &ioapic->dev);
 638         mutex_unlock(&kvm->slots_lock);
 639         if (ret < 0) {
 640                 kvm->arch.vioapic = NULL;
 641                 kfree(ioapic);
 642         }
 643
 644         return ret;
 645 }
 646
 647 void kvm_ioapic_destroy(struct kvm *kvm)
 648 {
 649         struct kvm_ioapic *ioapic = kvm->arch.vioapic;
 650
 651         if (!ioapic)
 652                 return;
 653
 654         cancel_delayed_work_sync(&ioapic->eoi_inject);
 655         mutex_lock(&kvm->slots_lock);
 656         kvm_io_bus_unregister_dev(kvm, KVM_MMIO_BUS, &ioapic->dev);
 657         mutex_unlock(&kvm->slots_lock);
 658         kvm->arch.vioapic = NULL;
 659         kfree(ioapic);
 660 }
 661
 662 void kvm_get_ioapic(struct kvm *kvm, struct kvm_ioapic_state *state)
 663 {
 664         struct kvm_ioapic *ioapic = kvm->arch.vioapic;
 665
 666         spin_lock(&ioapic->lock);
 667         memcpy(state, ioapic, sizeof(struct kvm_ioapic_state));
 668         state->irr &= ~ioapic->irr_delivered;
 669         spin_unlock(&ioapic->lock);
 670 }
 671
 672 void kvm_set_ioapic(struct kvm *kvm, struct kvm_ioapic_state *state)
 673 {
 674         struct kvm_ioapic *ioapic = kvm->arch.vioapic;
 675
 676         spin_lock(&ioapic->lock);
 677         memcpy(ioapic, state, sizeof(struct kvm_ioapic_state));
 678         ioapic->irr = 0;
 679         ioapic->irr_delivered = 0;
 680         kvm_make_scan_ioapic_request(kvm);
 681         kvm_ioapic_inject_all(ioapic, state->irr);
 682         spin_unlock(&ioapic->lock);
 683 }