err.no Git - linux-2.6/blob - drivers/kvm/x86.c

   1 /*
   2  * Kernel-based Virtual Machine driver for Linux
   3  *
   4  * derived from drivers/kvm/kvm_main.c
   5  *
   6  * Copyright (C) 2006 Qumranet, Inc.
   7  *
   8  * Authors:
   9  *   Avi Kivity   <avi@qumranet.com>
  10  *   Yaniv Kamay  <yaniv@qumranet.com>
  11  *
  12  * This work is licensed under the terms of the GNU GPL, version 2.  See
  13  * the COPYING file in the top-level directory.
  14  *
  15  */
  16
  17 #include "kvm.h"
  18 #include "x86.h"
  19 #include "x86_emulate.h"
  20 #include "segment_descriptor.h"
  21 #include "irq.h"
  22
  23 #include <linux/kvm.h>
  24 #include <linux/fs.h>
  25 #include <linux/vmalloc.h>
  26 #include <linux/module.h>
  27
  28 #include <asm/uaccess.h>
  29 #include <asm/msr.h>
  30
  31 #define MAX_IO_MSRS 256
  32 #define CR0_RESERVED_BITS                                               \
  33         (~(unsigned long)(X86_CR0_PE | X86_CR0_MP | X86_CR0_EM | X86_CR0_TS \
  34                           | X86_CR0_ET | X86_CR0_NE | X86_CR0_WP | X86_CR0_AM \
  35                           | X86_CR0_NW | X86_CR0_CD | X86_CR0_PG))
  36 #define CR4_RESERVED_BITS                                               \
  37         (~(unsigned long)(X86_CR4_VME | X86_CR4_PVI | X86_CR4_TSD | X86_CR4_DE\
  38                           | X86_CR4_PSE | X86_CR4_PAE | X86_CR4_MCE     \
  39                           | X86_CR4_PGE | X86_CR4_PCE | X86_CR4_OSFXSR  \
  40                           | X86_CR4_OSXMMEXCPT | X86_CR4_VMXE))
  41
  42 #define CR8_RESERVED_BITS (~(unsigned long)X86_CR8_TPR)
  43 #define EFER_RESERVED_BITS 0xfffffffffffff2fe
  44
  45 #define STAT_OFFSET(x) offsetof(struct kvm_vcpu, stat.x)
  46
  47 struct kvm_x86_ops *kvm_x86_ops;
  48
  49 struct kvm_stats_debugfs_item debugfs_entries[] = {
  50         { "pf_fixed", STAT_OFFSET(pf_fixed) },
  51         { "pf_guest", STAT_OFFSET(pf_guest) },
  52         { "tlb_flush", STAT_OFFSET(tlb_flush) },
  53         { "invlpg", STAT_OFFSET(invlpg) },
  54         { "exits", STAT_OFFSET(exits) },
  55         { "io_exits", STAT_OFFSET(io_exits) },
  56         { "mmio_exits", STAT_OFFSET(mmio_exits) },
  57         { "signal_exits", STAT_OFFSET(signal_exits) },
  58         { "irq_window", STAT_OFFSET(irq_window_exits) },
  59         { "halt_exits", STAT_OFFSET(halt_exits) },
  60         { "halt_wakeup", STAT_OFFSET(halt_wakeup) },
  61         { "request_irq", STAT_OFFSET(request_irq_exits) },
  62         { "irq_exits", STAT_OFFSET(irq_exits) },
  63         { "light_exits", STAT_OFFSET(light_exits) },
  64         { "efer_reload", STAT_OFFSET(efer_reload) },
  65         { NULL }
  66 };
  67
  68
  69 unsigned long segment_base(u16 selector)
  70 {
  71         struct descriptor_table gdt;
  72         struct segment_descriptor *d;
  73         unsigned long table_base;
  74         unsigned long v;
  75
  76         if (selector == 0)
  77                 return 0;
  78
  79         asm("sgdt %0" : "=m"(gdt));
  80         table_base = gdt.base;
  81
  82         if (selector & 4) {           /* from ldt */
  83                 u16 ldt_selector;
  84
  85                 asm("sldt %0" : "=g"(ldt_selector));
  86                 table_base = segment_base(ldt_selector);
  87         }
  88         d = (struct segment_descriptor *)(table_base + (selector & ~7));
  89         v = d->base_low | ((unsigned long)d->base_mid << 16) |
  90                 ((unsigned long)d->base_high << 24);
  91 #ifdef CONFIG_X86_64
  92         if (d->system == 0 && (d->type == 2 || d->type == 9 || d->type == 11))
  93                 v |= ((unsigned long) \
  94                       ((struct segment_descriptor_64 *)d)->base_higher) << 32;
  95 #endif
  96         return v;
  97 }
  98 EXPORT_SYMBOL_GPL(segment_base);
  99
 100 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
 101 {
 102         if (irqchip_in_kernel(vcpu->kvm))
 103                 return vcpu->apic_base;
 104         else
 105                 return vcpu->apic_base;
 106 }
 107 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
 108
 109 void kvm_set_apic_base(struct kvm_vcpu *vcpu, u64 data)
 110 {
 111         /* TODO: reserve bits check */
 112         if (irqchip_in_kernel(vcpu->kvm))
 113                 kvm_lapic_set_base(vcpu, data);
 114         else
 115                 vcpu->apic_base = data;
 116 }
 117 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
 118
 119 static void inject_gp(struct kvm_vcpu *vcpu)
 120 {
 121         kvm_x86_ops->inject_gp(vcpu, 0);
 122 }
 123
 124 /*
 125  * Load the pae pdptrs.  Return true is they are all valid.
 126  */
 127 int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
 128 {
 129         gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
 130         unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
 131         int i;
 132         int ret;
 133         u64 pdpte[ARRAY_SIZE(vcpu->pdptrs)];
 134
 135         mutex_lock(&vcpu->kvm->lock);
 136         ret = kvm_read_guest_page(vcpu->kvm, pdpt_gfn, pdpte,
 137                                   offset * sizeof(u64), sizeof(pdpte));
 138         if (ret < 0) {
 139                 ret = 0;
 140                 goto out;
 141         }
 142         for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
 143                 if ((pdpte[i] & 1) && (pdpte[i] & 0xfffffff0000001e6ull)) {
 144                         ret = 0;
 145                         goto out;
 146                 }
 147         }
 148         ret = 1;
 149
 150         memcpy(vcpu->pdptrs, pdpte, sizeof(vcpu->pdptrs));
 151 out:
 152         mutex_unlock(&vcpu->kvm->lock);
 153
 154         return ret;
 155 }
 156
 157 void set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
 158 {
 159         if (cr0 & CR0_RESERVED_BITS) {
 160                 printk(KERN_DEBUG "set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
 161                        cr0, vcpu->cr0);
 162                 inject_gp(vcpu);
 163                 return;
 164         }
 165
 166         if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD)) {
 167                 printk(KERN_DEBUG "set_cr0: #GP, CD == 0 && NW == 1\n");
 168                 inject_gp(vcpu);
 169                 return;
 170         }
 171
 172         if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE)) {
 173                 printk(KERN_DEBUG "set_cr0: #GP, set PG flag "
 174                        "and a clear PE flag\n");
 175                 inject_gp(vcpu);
 176                 return;
 177         }
 178
 179         if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
 180 #ifdef CONFIG_X86_64
 181                 if ((vcpu->shadow_efer & EFER_LME)) {
 182                         int cs_db, cs_l;
 183
 184                         if (!is_pae(vcpu)) {
 185                                 printk(KERN_DEBUG "set_cr0: #GP, start paging "
 186                                        "in long mode while PAE is disabled\n");
 187                                 inject_gp(vcpu);
 188                                 return;
 189                         }
 190                         kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
 191                         if (cs_l) {
 192                                 printk(KERN_DEBUG "set_cr0: #GP, start paging "
 193                                        "in long mode while CS.L == 1\n");
 194                                 inject_gp(vcpu);
 195                                 return;
 196
 197                         }
 198                 } else
 199 #endif
 200                 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->cr3)) {
 201                         printk(KERN_DEBUG "set_cr0: #GP, pdptrs "
 202                                "reserved bits\n");
 203                         inject_gp(vcpu);
 204                         return;
 205                 }
 206
 207         }
 208
 209         kvm_x86_ops->set_cr0(vcpu, cr0);
 210         vcpu->cr0 = cr0;
 211
 212         mutex_lock(&vcpu->kvm->lock);
 213         kvm_mmu_reset_context(vcpu);
 214         mutex_unlock(&vcpu->kvm->lock);
 215         return;
 216 }
 217 EXPORT_SYMBOL_GPL(set_cr0);
 218
 219 void lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
 220 {
 221         set_cr0(vcpu, (vcpu->cr0 & ~0x0ful) | (msw & 0x0f));
 222 }
 223 EXPORT_SYMBOL_GPL(lmsw);
 224
 225 void set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
 226 {
 227         if (cr4 & CR4_RESERVED_BITS) {
 228                 printk(KERN_DEBUG "set_cr4: #GP, reserved bits\n");
 229                 inject_gp(vcpu);
 230                 return;
 231         }
 232
 233         if (is_long_mode(vcpu)) {
 234                 if (!(cr4 & X86_CR4_PAE)) {
 235                         printk(KERN_DEBUG "set_cr4: #GP, clearing PAE while "
 236                                "in long mode\n");
 237                         inject_gp(vcpu);
 238                         return;
 239                 }
 240         } else if (is_paging(vcpu) && !is_pae(vcpu) && (cr4 & X86_CR4_PAE)
 241                    && !load_pdptrs(vcpu, vcpu->cr3)) {
 242                 printk(KERN_DEBUG "set_cr4: #GP, pdptrs reserved bits\n");
 243                 inject_gp(vcpu);
 244                 return;
 245         }
 246
 247         if (cr4 & X86_CR4_VMXE) {
 248                 printk(KERN_DEBUG "set_cr4: #GP, setting VMXE\n");
 249                 inject_gp(vcpu);
 250                 return;
 251         }
 252         kvm_x86_ops->set_cr4(vcpu, cr4);
 253         vcpu->cr4 = cr4;
 254         mutex_lock(&vcpu->kvm->lock);
 255         kvm_mmu_reset_context(vcpu);
 256         mutex_unlock(&vcpu->kvm->lock);
 257 }
 258 EXPORT_SYMBOL_GPL(set_cr4);
 259
 260 void set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
 261 {
 262         if (is_long_mode(vcpu)) {
 263                 if (cr3 & CR3_L_MODE_RESERVED_BITS) {
 264                         printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
 265                         inject_gp(vcpu);
 266                         return;
 267                 }
 268         } else {
 269                 if (is_pae(vcpu)) {
 270                         if (cr3 & CR3_PAE_RESERVED_BITS) {
 271                                 printk(KERN_DEBUG
 272                                        "set_cr3: #GP, reserved bits\n");
 273                                 inject_gp(vcpu);
 274                                 return;
 275                         }
 276                         if (is_paging(vcpu) && !load_pdptrs(vcpu, cr3)) {
 277                                 printk(KERN_DEBUG "set_cr3: #GP, pdptrs "
 278                                        "reserved bits\n");
 279                                 inject_gp(vcpu);
 280                                 return;
 281                         }
 282                 }
 283                 /*
 284                  * We don't check reserved bits in nonpae mode, because
 285                  * this isn't enforced, and VMware depends on this.
 286                  */
 287         }
 288
 289         mutex_lock(&vcpu->kvm->lock);
 290         /*
 291          * Does the new cr3 value map to physical memory? (Note, we
 292          * catch an invalid cr3 even in real-mode, because it would
 293          * cause trouble later on when we turn on paging anyway.)
 294          *
 295          * A real CPU would silently accept an invalid cr3 and would
 296          * attempt to use it - with largely undefined (and often hard
 297          * to debug) behavior on the guest side.
 298          */
 299         if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
 300                 inject_gp(vcpu);
 301         else {
 302                 vcpu->cr3 = cr3;
 303                 vcpu->mmu.new_cr3(vcpu);
 304         }
 305         mutex_unlock(&vcpu->kvm->lock);
 306 }
 307 EXPORT_SYMBOL_GPL(set_cr3);
 308
 309 void set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
 310 {
 311         if (cr8 & CR8_RESERVED_BITS) {
 312                 printk(KERN_DEBUG "set_cr8: #GP, reserved bits 0x%lx\n", cr8);
 313                 inject_gp(vcpu);
 314                 return;
 315         }
 316         if (irqchip_in_kernel(vcpu->kvm))
 317                 kvm_lapic_set_tpr(vcpu, cr8);
 318         else
 319                 vcpu->cr8 = cr8;
 320 }
 321 EXPORT_SYMBOL_GPL(set_cr8);
 322
 323 unsigned long get_cr8(struct kvm_vcpu *vcpu)
 324 {
 325         if (irqchip_in_kernel(vcpu->kvm))
 326                 return kvm_lapic_get_cr8(vcpu);
 327         else
 328                 return vcpu->cr8;
 329 }
 330 EXPORT_SYMBOL_GPL(get_cr8);
 331
 332 /*
 333  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
 334  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
 335  *
 336  * This list is modified at module load time to reflect the
 337  * capabilities of the host cpu.
 338  */
 339 static u32 msrs_to_save[] = {
 340         MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
 341         MSR_K6_STAR,
 342 #ifdef CONFIG_X86_64
 343         MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
 344 #endif
 345         MSR_IA32_TIME_STAMP_COUNTER,
 346 };
 347
 348 static unsigned num_msrs_to_save;
 349
 350 static u32 emulated_msrs[] = {
 351         MSR_IA32_MISC_ENABLE,
 352 };
 353
 354 #ifdef CONFIG_X86_64
 355
 356 static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
 357 {
 358         if (efer & EFER_RESERVED_BITS) {
 359                 printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
 360                        efer);
 361                 inject_gp(vcpu);
 362                 return;
 363         }
 364
 365         if (is_paging(vcpu)
 366             && (vcpu->shadow_efer & EFER_LME) != (efer & EFER_LME)) {
 367                 printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
 368                 inject_gp(vcpu);
 369                 return;
 370         }
 371
 372         kvm_x86_ops->set_efer(vcpu, efer);
 373
 374         efer &= ~EFER_LMA;
 375         efer |= vcpu->shadow_efer & EFER_LMA;
 376
 377         vcpu->shadow_efer = efer;
 378 }
 379
 380 #endif
 381
 382 /*
 383  * Writes msr value into into the appropriate "register".
 384  * Returns 0 on success, non-0 otherwise.
 385  * Assumes vcpu_load() was already called.
 386  */
 387 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
 388 {
 389         return kvm_x86_ops->set_msr(vcpu, msr_index, data);
 390 }
 391
 392 /*
 393  * Adapt set_msr() to msr_io()'s calling convention
 394  */
 395 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
 396 {
 397         return kvm_set_msr(vcpu, index, *data);
 398 }
 399
 400
 401 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
 402 {
 403         switch (msr) {
 404 #ifdef CONFIG_X86_64
 405         case MSR_EFER:
 406                 set_efer(vcpu, data);
 407                 break;
 408 #endif
 409         case MSR_IA32_MC0_STATUS:
 410                 pr_unimpl(vcpu, "%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n",
 411                        __FUNCTION__, data);
 412                 break;
 413         case MSR_IA32_MCG_STATUS:
 414                 pr_unimpl(vcpu, "%s: MSR_IA32_MCG_STATUS 0x%llx, nop\n",
 415                         __FUNCTION__, data);
 416                 break;
 417         case MSR_IA32_UCODE_REV:
 418         case MSR_IA32_UCODE_WRITE:
 419         case 0x200 ... 0x2ff: /* MTRRs */
 420                 break;
 421         case MSR_IA32_APICBASE:
 422                 kvm_set_apic_base(vcpu, data);
 423                 break;
 424         case MSR_IA32_MISC_ENABLE:
 425                 vcpu->ia32_misc_enable_msr = data;
 426                 break;
 427         default:
 428                 pr_unimpl(vcpu, "unhandled wrmsr: 0x%x\n", msr);
 429                 return 1;
 430         }
 431         return 0;
 432 }
 433 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
 434
 435
 436 /*
 437  * Reads an msr value (of 'msr_index') into 'pdata'.
 438  * Returns 0 on success, non-0 otherwise.
 439  * Assumes vcpu_load() was already called.
 440  */
 441 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
 442 {
 443         return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
 444 }
 445
 446 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
 447 {
 448         u64 data;
 449
 450         switch (msr) {
 451         case 0xc0010010: /* SYSCFG */
 452         case 0xc0010015: /* HWCR */
 453         case MSR_IA32_PLATFORM_ID:
 454         case MSR_IA32_P5_MC_ADDR:
 455         case MSR_IA32_P5_MC_TYPE:
 456         case MSR_IA32_MC0_CTL:
 457         case MSR_IA32_MCG_STATUS:
 458         case MSR_IA32_MCG_CAP:
 459         case MSR_IA32_MC0_MISC:
 460         case MSR_IA32_MC0_MISC+4:
 461         case MSR_IA32_MC0_MISC+8:
 462         case MSR_IA32_MC0_MISC+12:
 463         case MSR_IA32_MC0_MISC+16:
 464         case MSR_IA32_UCODE_REV:
 465         case MSR_IA32_PERF_STATUS:
 466         case MSR_IA32_EBL_CR_POWERON:
 467                 /* MTRR registers */
 468         case 0xfe:
 469         case 0x200 ... 0x2ff:
 470                 data = 0;
 471                 break;
 472         case 0xcd: /* fsb frequency */
 473                 data = 3;
 474                 break;
 475         case MSR_IA32_APICBASE:
 476                 data = kvm_get_apic_base(vcpu);
 477                 break;
 478         case MSR_IA32_MISC_ENABLE:
 479                 data = vcpu->ia32_misc_enable_msr;
 480                 break;
 481 #ifdef CONFIG_X86_64
 482         case MSR_EFER:
 483                 data = vcpu->shadow_efer;
 484                 break;
 485 #endif
 486         default:
 487                 pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
 488                 return 1;
 489         }
 490         *pdata = data;
 491         return 0;
 492 }
 493 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
 494
 495 /*
 496  * Read or write a bunch of msrs. All parameters are kernel addresses.
 497  *
 498  * @return number of msrs set successfully.
 499  */
 500 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
 501                     struct kvm_msr_entry *entries,
 502                     int (*do_msr)(struct kvm_vcpu *vcpu,
 503                                   unsigned index, u64 *data))
 504 {
 505         int i;
 506
 507         vcpu_load(vcpu);
 508
 509         for (i = 0; i < msrs->nmsrs; ++i)
 510                 if (do_msr(vcpu, entries[i].index, &entries[i].data))
 511                         break;
 512
 513         vcpu_put(vcpu);
 514
 515         return i;
 516 }
 517
 518 /*
 519  * Read or write a bunch of msrs. Parameters are user addresses.
 520  *
 521  * @return number of msrs set successfully.
 522  */
 523 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
 524                   int (*do_msr)(struct kvm_vcpu *vcpu,
 525                                 unsigned index, u64 *data),
 526                   int writeback)
 527 {
 528         struct kvm_msrs msrs;
 529         struct kvm_msr_entry *entries;
 530         int r, n;
 531         unsigned size;
 532
 533         r = -EFAULT;
 534         if (copy_from_user(&msrs, user_msrs, sizeof msrs))
 535                 goto out;
 536
 537         r = -E2BIG;
 538         if (msrs.nmsrs >= MAX_IO_MSRS)
 539                 goto out;
 540
 541         r = -ENOMEM;
 542         size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
 543         entries = vmalloc(size);
 544         if (!entries)
 545                 goto out;
 546
 547         r = -EFAULT;
 548         if (copy_from_user(entries, user_msrs->entries, size))
 549                 goto out_free;
 550
 551         r = n = __msr_io(vcpu, &msrs, entries, do_msr);
 552         if (r < 0)
 553                 goto out_free;
 554
 555         r = -EFAULT;
 556         if (writeback && copy_to_user(user_msrs->entries, entries, size))
 557                 goto out_free;
 558
 559         r = n;
 560
 561 out_free:
 562         vfree(entries);
 563 out:
 564         return r;
 565 }
 566
 567 /*
 568  * Make sure that a cpu that is being hot-unplugged does not have any vcpus
 569  * cached on it.
 570  */
 571 void decache_vcpus_on_cpu(int cpu)
 572 {
 573         struct kvm *vm;
 574         struct kvm_vcpu *vcpu;
 575         int i;
 576
 577         spin_lock(&kvm_lock);
 578         list_for_each_entry(vm, &vm_list, vm_list)
 579                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 580                         vcpu = vm->vcpus[i];
 581                         if (!vcpu)
 582                                 continue;
 583                         /*
 584                          * If the vcpu is locked, then it is running on some
 585                          * other cpu and therefore it is not cached on the
 586                          * cpu in question.
 587                          *
 588                          * If it's not locked, check the last cpu it executed
 589                          * on.
 590                          */
 591                         if (mutex_trylock(&vcpu->mutex)) {
 592                                 if (vcpu->cpu == cpu) {
 593                                         kvm_x86_ops->vcpu_decache(vcpu);
 594                                         vcpu->cpu = -1;
 595                                 }
 596                                 mutex_unlock(&vcpu->mutex);
 597                         }
 598                 }
 599         spin_unlock(&kvm_lock);
 600 }
 601
 602 long kvm_arch_dev_ioctl(struct file *filp,
 603                         unsigned int ioctl, unsigned long arg)
 604 {
 605         void __user *argp = (void __user *)arg;
 606         long r;
 607
 608         switch (ioctl) {
 609         case KVM_GET_MSR_INDEX_LIST: {
 610                 struct kvm_msr_list __user *user_msr_list = argp;
 611                 struct kvm_msr_list msr_list;
 612                 unsigned n;
 613
 614                 r = -EFAULT;
 615                 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
 616                         goto out;
 617                 n = msr_list.nmsrs;
 618                 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
 619                 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
 620                         goto out;
 621                 r = -E2BIG;
 622                 if (n < num_msrs_to_save)
 623                         goto out;
 624                 r = -EFAULT;
 625                 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
 626                                  num_msrs_to_save * sizeof(u32)))
 627                         goto out;
 628                 if (copy_to_user(user_msr_list->indices
 629                                  + num_msrs_to_save * sizeof(u32),
 630                                  &emulated_msrs,
 631                                  ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
 632                         goto out;
 633                 r = 0;
 634                 break;
 635         }
 636         default:
 637                 r = -EINVAL;
 638         }
 639 out:
 640         return r;
 641 }
 642
 643 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
 644 {
 645         kvm_x86_ops->vcpu_load(vcpu, cpu);
 646 }
 647
 648 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
 649 {
 650         kvm_x86_ops->vcpu_put(vcpu);
 651 }
 652
 653 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
 654 {
 655         u64 efer;
 656         int i;
 657         struct kvm_cpuid_entry *e, *entry;
 658
 659         rdmsrl(MSR_EFER, efer);
 660         entry = NULL;
 661         for (i = 0; i < vcpu->cpuid_nent; ++i) {
 662                 e = &vcpu->cpuid_entries[i];
 663                 if (e->function == 0x80000001) {
 664                         entry = e;
 665                         break;
 666                 }
 667         }
 668         if (entry && (entry->edx & (1 << 20)) && !(efer & EFER_NX)) {
 669                 entry->edx &= ~(1 << 20);
 670                 printk(KERN_INFO "kvm: guest NX capability removed\n");
 671         }
 672 }
 673
 674 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
 675                                     struct kvm_cpuid *cpuid,
 676                                     struct kvm_cpuid_entry __user *entries)
 677 {
 678         int r;
 679
 680         r = -E2BIG;
 681         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
 682                 goto out;
 683         r = -EFAULT;
 684         if (copy_from_user(&vcpu->cpuid_entries, entries,
 685                            cpuid->nent * sizeof(struct kvm_cpuid_entry)))
 686                 goto out;
 687         vcpu->cpuid_nent = cpuid->nent;
 688         cpuid_fix_nx_cap(vcpu);
 689         return 0;
 690
 691 out:
 692         return r;
 693 }
 694
 695 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
 696                                     struct kvm_lapic_state *s)
 697 {
 698         vcpu_load(vcpu);
 699         memcpy(s->regs, vcpu->apic->regs, sizeof *s);
 700         vcpu_put(vcpu);
 701
 702         return 0;
 703 }
 704
 705 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
 706                                     struct kvm_lapic_state *s)
 707 {
 708         vcpu_load(vcpu);
 709         memcpy(vcpu->apic->regs, s->regs, sizeof *s);
 710         kvm_apic_post_state_restore(vcpu);
 711         vcpu_put(vcpu);
 712
 713         return 0;
 714 }
 715
 716 long kvm_arch_vcpu_ioctl(struct file *filp,
 717                          unsigned int ioctl, unsigned long arg)
 718 {
 719         struct kvm_vcpu *vcpu = filp->private_data;
 720         void __user *argp = (void __user *)arg;
 721         int r;
 722
 723         switch (ioctl) {
 724         case KVM_GET_LAPIC: {
 725                 struct kvm_lapic_state lapic;
 726
 727                 memset(&lapic, 0, sizeof lapic);
 728                 r = kvm_vcpu_ioctl_get_lapic(vcpu, &lapic);
 729                 if (r)
 730                         goto out;
 731                 r = -EFAULT;
 732                 if (copy_to_user(argp, &lapic, sizeof lapic))
 733                         goto out;
 734                 r = 0;
 735                 break;
 736         }
 737         case KVM_SET_LAPIC: {
 738                 struct kvm_lapic_state lapic;
 739
 740                 r = -EFAULT;
 741                 if (copy_from_user(&lapic, argp, sizeof lapic))
 742                         goto out;
 743                 r = kvm_vcpu_ioctl_set_lapic(vcpu, &lapic);;
 744                 if (r)
 745                         goto out;
 746                 r = 0;
 747                 break;
 748         }
 749         case KVM_SET_CPUID: {
 750                 struct kvm_cpuid __user *cpuid_arg = argp;
 751                 struct kvm_cpuid cpuid;
 752
 753                 r = -EFAULT;
 754                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
 755                         goto out;
 756                 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
 757                 if (r)
 758                         goto out;
 759                 break;
 760         }
 761         case KVM_GET_MSRS:
 762                 r = msr_io(vcpu, argp, kvm_get_msr, 1);
 763                 break;
 764         case KVM_SET_MSRS:
 765                 r = msr_io(vcpu, argp, do_set_msr, 0);
 766                 break;
 767         default:
 768                 r = -EINVAL;
 769         }
 770 out:
 771         return r;
 772 }
 773
 774 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
 775 {
 776         int ret;
 777
 778         if (addr > (unsigned int)(-3 * PAGE_SIZE))
 779                 return -1;
 780         ret = kvm_x86_ops->set_tss_addr(kvm, addr);
 781         return ret;
 782 }
 783
 784 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
 785                                           u32 kvm_nr_mmu_pages)
 786 {
 787         if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
 788                 return -EINVAL;
 789
 790         mutex_lock(&kvm->lock);
 791
 792         kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
 793         kvm->n_requested_mmu_pages = kvm_nr_mmu_pages;
 794
 795         mutex_unlock(&kvm->lock);
 796         return 0;
 797 }
 798
 799 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
 800 {
 801         return kvm->n_alloc_mmu_pages;
 802 }
 803
 804 /*
 805  * Set a new alias region.  Aliases map a portion of physical memory into
 806  * another portion.  This is useful for memory windows, for example the PC
 807  * VGA region.
 808  */
 809 static int kvm_vm_ioctl_set_memory_alias(struct kvm *kvm,
 810                                          struct kvm_memory_alias *alias)
 811 {
 812         int r, n;
 813         struct kvm_mem_alias *p;
 814
 815         r = -EINVAL;
 816         /* General sanity checks */
 817         if (alias->memory_size & (PAGE_SIZE - 1))
 818                 goto out;
 819         if (alias->guest_phys_addr & (PAGE_SIZE - 1))
 820                 goto out;
 821         if (alias->slot >= KVM_ALIAS_SLOTS)
 822                 goto out;
 823         if (alias->guest_phys_addr + alias->memory_size
 824             < alias->guest_phys_addr)
 825                 goto out;
 826         if (alias->target_phys_addr + alias->memory_size
 827             < alias->target_phys_addr)
 828                 goto out;
 829
 830         mutex_lock(&kvm->lock);
 831
 832         p = &kvm->aliases[alias->slot];
 833         p->base_gfn = alias->guest_phys_addr >> PAGE_SHIFT;
 834         p->npages = alias->memory_size >> PAGE_SHIFT;
 835         p->target_gfn = alias->target_phys_addr >> PAGE_SHIFT;
 836
 837         for (n = KVM_ALIAS_SLOTS; n > 0; --n)
 838                 if (kvm->aliases[n - 1].npages)
 839                         break;
 840         kvm->naliases = n;
 841
 842         kvm_mmu_zap_all(kvm);
 843
 844         mutex_unlock(&kvm->lock);
 845
 846         return 0;
 847
 848 out:
 849         return r;
 850 }
 851
 852 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
 853 {
 854         int r;
 855
 856         r = 0;
 857         switch (chip->chip_id) {
 858         case KVM_IRQCHIP_PIC_MASTER:
 859                 memcpy(&chip->chip.pic,
 860                         &pic_irqchip(kvm)->pics[0],
 861                         sizeof(struct kvm_pic_state));
 862                 break;
 863         case KVM_IRQCHIP_PIC_SLAVE:
 864                 memcpy(&chip->chip.pic,
 865                         &pic_irqchip(kvm)->pics[1],
 866                         sizeof(struct kvm_pic_state));
 867                 break;
 868         case KVM_IRQCHIP_IOAPIC:
 869                 memcpy(&chip->chip.ioapic,
 870                         ioapic_irqchip(kvm),
 871                         sizeof(struct kvm_ioapic_state));
 872                 break;
 873         default:
 874                 r = -EINVAL;
 875                 break;
 876         }
 877         return r;
 878 }
 879
 880 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
 881 {
 882         int r;
 883
 884         r = 0;
 885         switch (chip->chip_id) {
 886         case KVM_IRQCHIP_PIC_MASTER:
 887                 memcpy(&pic_irqchip(kvm)->pics[0],
 888                         &chip->chip.pic,
 889                         sizeof(struct kvm_pic_state));
 890                 break;
 891         case KVM_IRQCHIP_PIC_SLAVE:
 892                 memcpy(&pic_irqchip(kvm)->pics[1],
 893                         &chip->chip.pic,
 894                         sizeof(struct kvm_pic_state));
 895                 break;
 896         case KVM_IRQCHIP_IOAPIC:
 897                 memcpy(ioapic_irqchip(kvm),
 898                         &chip->chip.ioapic,
 899                         sizeof(struct kvm_ioapic_state));
 900                 break;
 901         default:
 902                 r = -EINVAL;
 903                 break;
 904         }
 905         kvm_pic_update_irq(pic_irqchip(kvm));
 906         return r;
 907 }
 908
 909 long kvm_arch_vm_ioctl(struct file *filp,
 910                        unsigned int ioctl, unsigned long arg)
 911 {
 912         struct kvm *kvm = filp->private_data;
 913         void __user *argp = (void __user *)arg;
 914         int r = -EINVAL;
 915
 916         switch (ioctl) {
 917         case KVM_SET_TSS_ADDR:
 918                 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
 919                 if (r < 0)
 920                         goto out;
 921                 break;
 922         case KVM_SET_MEMORY_REGION: {
 923                 struct kvm_memory_region kvm_mem;
 924                 struct kvm_userspace_memory_region kvm_userspace_mem;
 925
 926                 r = -EFAULT;
 927                 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
 928                         goto out;
 929                 kvm_userspace_mem.slot = kvm_mem.slot;
 930                 kvm_userspace_mem.flags = kvm_mem.flags;
 931                 kvm_userspace_mem.guest_phys_addr = kvm_mem.guest_phys_addr;
 932                 kvm_userspace_mem.memory_size = kvm_mem.memory_size;
 933                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 0);
 934                 if (r)
 935                         goto out;
 936                 break;
 937         }
 938         case KVM_SET_NR_MMU_PAGES:
 939                 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
 940                 if (r)
 941                         goto out;
 942                 break;
 943         case KVM_GET_NR_MMU_PAGES:
 944                 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
 945                 break;
 946         case KVM_SET_MEMORY_ALIAS: {
 947                 struct kvm_memory_alias alias;
 948
 949                 r = -EFAULT;
 950                 if (copy_from_user(&alias, argp, sizeof alias))
 951                         goto out;
 952                 r = kvm_vm_ioctl_set_memory_alias(kvm, &alias);
 953                 if (r)
 954                         goto out;
 955                 break;
 956         }
 957         case KVM_CREATE_IRQCHIP:
 958                 r = -ENOMEM;
 959                 kvm->vpic = kvm_create_pic(kvm);
 960                 if (kvm->vpic) {
 961                         r = kvm_ioapic_init(kvm);
 962                         if (r) {
 963                                 kfree(kvm->vpic);
 964                                 kvm->vpic = NULL;
 965                                 goto out;
 966                         }
 967                 } else
 968                         goto out;
 969                 break;
 970         case KVM_IRQ_LINE: {
 971                 struct kvm_irq_level irq_event;
 972
 973                 r = -EFAULT;
 974                 if (copy_from_user(&irq_event, argp, sizeof irq_event))
 975                         goto out;
 976                 if (irqchip_in_kernel(kvm)) {
 977                         mutex_lock(&kvm->lock);
 978                         if (irq_event.irq < 16)
 979                                 kvm_pic_set_irq(pic_irqchip(kvm),
 980                                         irq_event.irq,
 981                                         irq_event.level);
 982                         kvm_ioapic_set_irq(kvm->vioapic,
 983                                         irq_event.irq,
 984                                         irq_event.level);
 985                         mutex_unlock(&kvm->lock);
 986                         r = 0;
 987                 }
 988                 break;
 989         }
 990         case KVM_GET_IRQCHIP: {
 991                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
 992                 struct kvm_irqchip chip;
 993
 994                 r = -EFAULT;
 995                 if (copy_from_user(&chip, argp, sizeof chip))
 996                         goto out;
 997                 r = -ENXIO;
 998                 if (!irqchip_in_kernel(kvm))
 999                         goto out;
1000                 r = kvm_vm_ioctl_get_irqchip(kvm, &chip);
1001                 if (r)
1002                         goto out;
1003                 r = -EFAULT;
1004                 if (copy_to_user(argp, &chip, sizeof chip))
1005                         goto out;
1006                 r = 0;
1007                 break;
1008         }
1009         case KVM_SET_IRQCHIP: {
1010                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
1011                 struct kvm_irqchip chip;
1012
1013                 r = -EFAULT;
1014                 if (copy_from_user(&chip, argp, sizeof chip))
1015                         goto out;
1016                 r = -ENXIO;
1017                 if (!irqchip_in_kernel(kvm))
1018                         goto out;
1019                 r = kvm_vm_ioctl_set_irqchip(kvm, &chip);
1020                 if (r)
1021                         goto out;
1022                 r = 0;
1023                 break;
1024         }
1025         default:
1026                 ;
1027         }
1028 out:
1029         return r;
1030 }
1031
1032 static __init void kvm_init_msr_list(void)
1033 {
1034         u32 dummy[2];
1035         unsigned i, j;
1036
1037         for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
1038                 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
1039                         continue;
1040                 if (j < i)
1041                         msrs_to_save[j] = msrs_to_save[i];
1042                 j++;
1043         }
1044         num_msrs_to_save = j;
1045 }
1046
1047 /*
1048  * Only apic need an MMIO device hook, so shortcut now..
1049  */
1050 static struct kvm_io_device *vcpu_find_pervcpu_dev(struct kvm_vcpu *vcpu,
1051                                                 gpa_t addr)
1052 {
1053         struct kvm_io_device *dev;
1054
1055         if (vcpu->apic) {
1056                 dev = &vcpu->apic->dev;
1057                 if (dev->in_range(dev, addr))
1058                         return dev;
1059         }
1060         return NULL;
1061 }
1062
1063
1064 static struct kvm_io_device *vcpu_find_mmio_dev(struct kvm_vcpu *vcpu,
1065                                                 gpa_t addr)
1066 {
1067         struct kvm_io_device *dev;
1068
1069         dev = vcpu_find_pervcpu_dev(vcpu, addr);
1070         if (dev == NULL)
1071                 dev = kvm_io_bus_find_dev(&vcpu->kvm->mmio_bus, addr);
1072         return dev;
1073 }
1074
1075 int emulator_read_std(unsigned long addr,
1076                              void *val,
1077                              unsigned int bytes,
1078                              struct kvm_vcpu *vcpu)
1079 {
1080         void *data = val;
1081
1082         while (bytes) {
1083                 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1084                 unsigned offset = addr & (PAGE_SIZE-1);
1085                 unsigned tocopy = min(bytes, (unsigned)PAGE_SIZE - offset);
1086                 int ret;
1087
1088                 if (gpa == UNMAPPED_GVA)
1089                         return X86EMUL_PROPAGATE_FAULT;
1090                 ret = kvm_read_guest(vcpu->kvm, gpa, data, tocopy);
1091                 if (ret < 0)
1092                         return X86EMUL_UNHANDLEABLE;
1093
1094                 bytes -= tocopy;
1095                 data += tocopy;
1096                 addr += tocopy;
1097         }
1098
1099         return X86EMUL_CONTINUE;
1100 }
1101 EXPORT_SYMBOL_GPL(emulator_read_std);
1102
1103 static int emulator_write_std(unsigned long addr,
1104                               const void *val,
1105                               unsigned int bytes,
1106                               struct kvm_vcpu *vcpu)
1107 {
1108         pr_unimpl(vcpu, "emulator_write_std: addr %lx n %d\n", addr, bytes);
1109         return X86EMUL_UNHANDLEABLE;
1110 }
1111
1112 static int emulator_read_emulated(unsigned long addr,
1113                                   void *val,
1114                                   unsigned int bytes,
1115                                   struct kvm_vcpu *vcpu)
1116 {
1117         struct kvm_io_device *mmio_dev;
1118         gpa_t                 gpa;
1119
1120         if (vcpu->mmio_read_completed) {
1121                 memcpy(val, vcpu->mmio_data, bytes);
1122                 vcpu->mmio_read_completed = 0;
1123                 return X86EMUL_CONTINUE;
1124         }
1125
1126         gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1127
1128         /* For APIC access vmexit */
1129         if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
1130                 goto mmio;
1131
1132         if (emulator_read_std(addr, val, bytes, vcpu)
1133                         == X86EMUL_CONTINUE)
1134                 return X86EMUL_CONTINUE;
1135         if (gpa == UNMAPPED_GVA)
1136                 return X86EMUL_PROPAGATE_FAULT;
1137
1138 mmio:
1139         /*
1140          * Is this MMIO handled locally?
1141          */
1142         mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1143         if (mmio_dev) {
1144                 kvm_iodevice_read(mmio_dev, gpa, bytes, val);
1145                 return X86EMUL_CONTINUE;
1146         }
1147
1148         vcpu->mmio_needed = 1;
1149         vcpu->mmio_phys_addr = gpa;
1150         vcpu->mmio_size = bytes;
1151         vcpu->mmio_is_write = 0;
1152
1153         return X86EMUL_UNHANDLEABLE;
1154 }
1155
1156 static int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
1157                                const void *val, int bytes)
1158 {
1159         int ret;
1160
1161         ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
1162         if (ret < 0)
1163                 return 0;
1164         kvm_mmu_pte_write(vcpu, gpa, val, bytes);
1165         return 1;
1166 }
1167
1168 static int emulator_write_emulated_onepage(unsigned long addr,
1169                                            const void *val,
1170                                            unsigned int bytes,
1171                                            struct kvm_vcpu *vcpu)
1172 {
1173         struct kvm_io_device *mmio_dev;
1174         gpa_t                 gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1175
1176         if (gpa == UNMAPPED_GVA) {
1177                 kvm_x86_ops->inject_page_fault(vcpu, addr, 2);
1178                 return X86EMUL_PROPAGATE_FAULT;
1179         }
1180
1181         /* For APIC access vmexit */
1182         if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
1183                 goto mmio;
1184
1185         if (emulator_write_phys(vcpu, gpa, val, bytes))
1186                 return X86EMUL_CONTINUE;
1187
1188 mmio:
1189         /*
1190          * Is this MMIO handled locally?
1191          */
1192         mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1193         if (mmio_dev) {
1194                 kvm_iodevice_write(mmio_dev, gpa, bytes, val);
1195                 return X86EMUL_CONTINUE;
1196         }
1197
1198         vcpu->mmio_needed = 1;
1199         vcpu->mmio_phys_addr = gpa;
1200         vcpu->mmio_size = bytes;
1201         vcpu->mmio_is_write = 1;
1202         memcpy(vcpu->mmio_data, val, bytes);
1203
1204         return X86EMUL_CONTINUE;
1205 }
1206
1207 int emulator_write_emulated(unsigned long addr,
1208                                    const void *val,
1209                                    unsigned int bytes,
1210                                    struct kvm_vcpu *vcpu)
1211 {
1212         /* Crossing a page boundary? */
1213         if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
1214                 int rc, now;
1215
1216                 now = -addr & ~PAGE_MASK;
1217                 rc = emulator_write_emulated_onepage(addr, val, now, vcpu);
1218                 if (rc != X86EMUL_CONTINUE)
1219                         return rc;
1220                 addr += now;
1221                 val += now;
1222                 bytes -= now;
1223         }
1224         return emulator_write_emulated_onepage(addr, val, bytes, vcpu);
1225 }
1226 EXPORT_SYMBOL_GPL(emulator_write_emulated);
1227
1228 static int emulator_cmpxchg_emulated(unsigned long addr,
1229                                      const void *old,
1230                                      const void *new,
1231                                      unsigned int bytes,
1232                                      struct kvm_vcpu *vcpu)
1233 {
1234         static int reported;
1235
1236         if (!reported) {
1237                 reported = 1;
1238                 printk(KERN_WARNING "kvm: emulating exchange as write\n");
1239         }
1240         return emulator_write_emulated(addr, new, bytes, vcpu);
1241 }
1242
1243 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
1244 {
1245         return kvm_x86_ops->get_segment_base(vcpu, seg);
1246 }
1247
1248 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
1249 {
1250         return X86EMUL_CONTINUE;
1251 }
1252
1253 int emulate_clts(struct kvm_vcpu *vcpu)
1254 {
1255         kvm_x86_ops->set_cr0(vcpu, vcpu->cr0 & ~X86_CR0_TS);
1256         return X86EMUL_CONTINUE;
1257 }
1258
1259 int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long *dest)
1260 {
1261         struct kvm_vcpu *vcpu = ctxt->vcpu;
1262
1263         switch (dr) {
1264         case 0 ... 3:
1265                 *dest = kvm_x86_ops->get_dr(vcpu, dr);
1266                 return X86EMUL_CONTINUE;
1267         default:
1268                 pr_unimpl(vcpu, "%s: unexpected dr %u\n", __FUNCTION__, dr);
1269                 return X86EMUL_UNHANDLEABLE;
1270         }
1271 }
1272
1273 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
1274 {
1275         unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
1276         int exception;
1277
1278         kvm_x86_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
1279         if (exception) {
1280                 /* FIXME: better handling */
1281                 return X86EMUL_UNHANDLEABLE;
1282         }
1283         return X86EMUL_CONTINUE;
1284 }
1285
1286 void kvm_report_emulation_failure(struct kvm_vcpu *vcpu, const char *context)
1287 {
1288         static int reported;
1289         u8 opcodes[4];
1290         unsigned long rip = vcpu->rip;
1291         unsigned long rip_linear;
1292
1293         rip_linear = rip + get_segment_base(vcpu, VCPU_SREG_CS);
1294
1295         if (reported)
1296                 return;
1297
1298         emulator_read_std(rip_linear, (void *)opcodes, 4, vcpu);
1299
1300         printk(KERN_ERR "emulation failed (%s) rip %lx %02x %02x %02x %02x\n",
1301                context, rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
1302         reported = 1;
1303 }
1304 EXPORT_SYMBOL_GPL(kvm_report_emulation_failure);
1305
1306 struct x86_emulate_ops emulate_ops = {
1307         .read_std            = emulator_read_std,
1308         .write_std           = emulator_write_std,
1309         .read_emulated       = emulator_read_emulated,
1310         .write_emulated      = emulator_write_emulated,
1311         .cmpxchg_emulated    = emulator_cmpxchg_emulated,
1312 };
1313
1314 int emulate_instruction(struct kvm_vcpu *vcpu,
1315                         struct kvm_run *run,
1316                         unsigned long cr2,
1317                         u16 error_code,
1318                         int no_decode)
1319 {
1320         int r;
1321
1322         vcpu->mmio_fault_cr2 = cr2;
1323         kvm_x86_ops->cache_regs(vcpu);
1324
1325         vcpu->mmio_is_write = 0;
1326         vcpu->pio.string = 0;
1327
1328         if (!no_decode) {
1329                 int cs_db, cs_l;
1330                 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
1331
1332                 vcpu->emulate_ctxt.vcpu = vcpu;
1333                 vcpu->emulate_ctxt.eflags = kvm_x86_ops->get_rflags(vcpu);
1334                 vcpu->emulate_ctxt.cr2 = cr2;
1335                 vcpu->emulate_ctxt.mode =
1336                         (vcpu->emulate_ctxt.eflags & X86_EFLAGS_VM)
1337                         ? X86EMUL_MODE_REAL : cs_l
1338                         ? X86EMUL_MODE_PROT64 : cs_db
1339                         ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
1340
1341                 if (vcpu->emulate_ctxt.mode == X86EMUL_MODE_PROT64) {
1342                         vcpu->emulate_ctxt.cs_base = 0;
1343                         vcpu->emulate_ctxt.ds_base = 0;
1344                         vcpu->emulate_ctxt.es_base = 0;
1345                         vcpu->emulate_ctxt.ss_base = 0;
1346                 } else {
1347                         vcpu->emulate_ctxt.cs_base =
1348                                         get_segment_base(vcpu, VCPU_SREG_CS);
1349                         vcpu->emulate_ctxt.ds_base =
1350                                         get_segment_base(vcpu, VCPU_SREG_DS);
1351                         vcpu->emulate_ctxt.es_base =
1352                                         get_segment_base(vcpu, VCPU_SREG_ES);
1353                         vcpu->emulate_ctxt.ss_base =
1354                                         get_segment_base(vcpu, VCPU_SREG_SS);
1355                 }
1356
1357                 vcpu->emulate_ctxt.gs_base =
1358                                         get_segment_base(vcpu, VCPU_SREG_GS);
1359                 vcpu->emulate_ctxt.fs_base =
1360                                         get_segment_base(vcpu, VCPU_SREG_FS);
1361
1362                 r = x86_decode_insn(&vcpu->emulate_ctxt, &emulate_ops);
1363                 if (r)  {
1364                         if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
1365                                 return EMULATE_DONE;
1366                         return EMULATE_FAIL;
1367                 }
1368         }
1369
1370         r = x86_emulate_insn(&vcpu->emulate_ctxt, &emulate_ops);
1371
1372         if (vcpu->pio.string)
1373                 return EMULATE_DO_MMIO;
1374
1375         if ((r || vcpu->mmio_is_write) && run) {
1376                 run->exit_reason = KVM_EXIT_MMIO;
1377                 run->mmio.phys_addr = vcpu->mmio_phys_addr;
1378                 memcpy(run->mmio.data, vcpu->mmio_data, 8);
1379                 run->mmio.len = vcpu->mmio_size;
1380                 run->mmio.is_write = vcpu->mmio_is_write;
1381         }
1382
1383         if (r) {
1384                 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
1385                         return EMULATE_DONE;
1386                 if (!vcpu->mmio_needed) {
1387                         kvm_report_emulation_failure(vcpu, "mmio");
1388                         return EMULATE_FAIL;
1389                 }
1390                 return EMULATE_DO_MMIO;
1391         }
1392
1393         kvm_x86_ops->decache_regs(vcpu);
1394         kvm_x86_ops->set_rflags(vcpu, vcpu->emulate_ctxt.eflags);
1395
1396         if (vcpu->mmio_is_write) {
1397                 vcpu->mmio_needed = 0;
1398                 return EMULATE_DO_MMIO;
1399         }
1400
1401         return EMULATE_DONE;
1402 }
1403 EXPORT_SYMBOL_GPL(emulate_instruction);
1404
1405 static void free_pio_guest_pages(struct kvm_vcpu *vcpu)
1406 {
1407         int i;
1408
1409         for (i = 0; i < ARRAY_SIZE(vcpu->pio.guest_pages); ++i)
1410                 if (vcpu->pio.guest_pages[i]) {
1411                         kvm_release_page(vcpu->pio.guest_pages[i]);
1412                         vcpu->pio.guest_pages[i] = NULL;
1413                 }
1414 }
1415
1416 static int pio_copy_data(struct kvm_vcpu *vcpu)
1417 {
1418         void *p = vcpu->pio_data;
1419         void *q;
1420         unsigned bytes;
1421         int nr_pages = vcpu->pio.guest_pages[1] ? 2 : 1;
1422
1423         q = vmap(vcpu->pio.guest_pages, nr_pages, VM_READ|VM_WRITE,
1424                  PAGE_KERNEL);
1425         if (!q) {
1426                 free_pio_guest_pages(vcpu);
1427                 return -ENOMEM;
1428         }
1429         q += vcpu->pio.guest_page_offset;
1430         bytes = vcpu->pio.size * vcpu->pio.cur_count;
1431         if (vcpu->pio.in)
1432                 memcpy(q, p, bytes);
1433         else
1434                 memcpy(p, q, bytes);
1435         q -= vcpu->pio.guest_page_offset;
1436         vunmap(q);
1437         free_pio_guest_pages(vcpu);
1438         return 0;
1439 }
1440
1441 int complete_pio(struct kvm_vcpu *vcpu)
1442 {
1443         struct kvm_pio_request *io = &vcpu->pio;
1444         long delta;
1445         int r;
1446
1447         kvm_x86_ops->cache_regs(vcpu);
1448
1449         if (!io->string) {
1450                 if (io->in)
1451                         memcpy(&vcpu->regs[VCPU_REGS_RAX], vcpu->pio_data,
1452                                io->size);
1453         } else {
1454                 if (io->in) {
1455                         r = pio_copy_data(vcpu);
1456                         if (r) {
1457                                 kvm_x86_ops->cache_regs(vcpu);
1458                                 return r;
1459                         }
1460                 }
1461
1462                 delta = 1;
1463                 if (io->rep) {
1464                         delta *= io->cur_count;
1465                         /*
1466                          * The size of the register should really depend on
1467                          * current address size.
1468                          */
1469                         vcpu->regs[VCPU_REGS_RCX] -= delta;
1470                 }
1471                 if (io->down)
1472                         delta = -delta;
1473                 delta *= io->size;
1474                 if (io->in)
1475                         vcpu->regs[VCPU_REGS_RDI] += delta;
1476                 else
1477                         vcpu->regs[VCPU_REGS_RSI] += delta;
1478         }
1479
1480         kvm_x86_ops->decache_regs(vcpu);
1481
1482         io->count -= io->cur_count;
1483         io->cur_count = 0;
1484
1485         return 0;
1486 }
1487
1488 static void kernel_pio(struct kvm_io_device *pio_dev,
1489                        struct kvm_vcpu *vcpu,
1490                        void *pd)
1491 {
1492         /* TODO: String I/O for in kernel device */
1493
1494         mutex_lock(&vcpu->kvm->lock);
1495         if (vcpu->pio.in)
1496                 kvm_iodevice_read(pio_dev, vcpu->pio.port,
1497                                   vcpu->pio.size,
1498                                   pd);
1499         else
1500                 kvm_iodevice_write(pio_dev, vcpu->pio.port,
1501                                    vcpu->pio.size,
1502                                    pd);
1503         mutex_unlock(&vcpu->kvm->lock);
1504 }
1505
1506 static void pio_string_write(struct kvm_io_device *pio_dev,
1507                              struct kvm_vcpu *vcpu)
1508 {
1509         struct kvm_pio_request *io = &vcpu->pio;
1510         void *pd = vcpu->pio_data;
1511         int i;
1512
1513         mutex_lock(&vcpu->kvm->lock);
1514         for (i = 0; i < io->cur_count; i++) {
1515                 kvm_iodevice_write(pio_dev, io->port,
1516                                    io->size,
1517                                    pd);
1518                 pd += io->size;
1519         }
1520         mutex_unlock(&vcpu->kvm->lock);
1521 }
1522
1523 static struct kvm_io_device *vcpu_find_pio_dev(struct kvm_vcpu *vcpu,
1524                                                gpa_t addr)
1525 {
1526         return kvm_io_bus_find_dev(&vcpu->kvm->pio_bus, addr);
1527 }
1528
1529 int kvm_emulate_pio(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
1530                   int size, unsigned port)
1531 {
1532         struct kvm_io_device *pio_dev;
1533
1534         vcpu->run->exit_reason = KVM_EXIT_IO;
1535         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
1536         vcpu->run->io.size = vcpu->pio.size = size;
1537         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
1538         vcpu->run->io.count = vcpu->pio.count = vcpu->pio.cur_count = 1;
1539         vcpu->run->io.port = vcpu->pio.port = port;
1540         vcpu->pio.in = in;
1541         vcpu->pio.string = 0;
1542         vcpu->pio.down = 0;
1543         vcpu->pio.guest_page_offset = 0;
1544         vcpu->pio.rep = 0;
1545
1546         kvm_x86_ops->cache_regs(vcpu);
1547         memcpy(vcpu->pio_data, &vcpu->regs[VCPU_REGS_RAX], 4);
1548         kvm_x86_ops->decache_regs(vcpu);
1549
1550         kvm_x86_ops->skip_emulated_instruction(vcpu);
1551
1552         pio_dev = vcpu_find_pio_dev(vcpu, port);
1553         if (pio_dev) {
1554                 kernel_pio(pio_dev, vcpu, vcpu->pio_data);
1555                 complete_pio(vcpu);
1556                 return 1;
1557         }
1558         return 0;
1559 }
1560 EXPORT_SYMBOL_GPL(kvm_emulate_pio);
1561
1562 int kvm_emulate_pio_string(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
1563                   int size, unsigned long count, int down,
1564                   gva_t address, int rep, unsigned port)
1565 {
1566         unsigned now, in_page;
1567         int i, ret = 0;
1568         int nr_pages = 1;
1569         struct page *page;
1570         struct kvm_io_device *pio_dev;
1571
1572         vcpu->run->exit_reason = KVM_EXIT_IO;
1573         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
1574         vcpu->run->io.size = vcpu->pio.size = size;
1575         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
1576         vcpu->run->io.count = vcpu->pio.count = vcpu->pio.cur_count = count;
1577         vcpu->run->io.port = vcpu->pio.port = port;
1578         vcpu->pio.in = in;
1579         vcpu->pio.string = 1;
1580         vcpu->pio.down = down;
1581         vcpu->pio.guest_page_offset = offset_in_page(address);
1582         vcpu->pio.rep = rep;
1583
1584         if (!count) {
1585                 kvm_x86_ops->skip_emulated_instruction(vcpu);
1586                 return 1;
1587         }
1588
1589         if (!down)
1590                 in_page = PAGE_SIZE - offset_in_page(address);
1591         else
1592                 in_page = offset_in_page(address) + size;
1593         now = min(count, (unsigned long)in_page / size);
1594         if (!now) {
1595                 /*
1596                  * String I/O straddles page boundary.  Pin two guest pages
1597                  * so that we satisfy atomicity constraints.  Do just one
1598                  * transaction to avoid complexity.
1599                  */
1600                 nr_pages = 2;
1601                 now = 1;
1602         }
1603         if (down) {
1604                 /*
1605                  * String I/O in reverse.  Yuck.  Kill the guest, fix later.
1606                  */
1607                 pr_unimpl(vcpu, "guest string pio down\n");
1608                 inject_gp(vcpu);
1609                 return 1;
1610         }
1611         vcpu->run->io.count = now;
1612         vcpu->pio.cur_count = now;
1613
1614         if (vcpu->pio.cur_count == vcpu->pio.count)
1615                 kvm_x86_ops->skip_emulated_instruction(vcpu);
1616
1617         for (i = 0; i < nr_pages; ++i) {
1618                 mutex_lock(&vcpu->kvm->lock);
1619                 page = gva_to_page(vcpu, address + i * PAGE_SIZE);
1620                 vcpu->pio.guest_pages[i] = page;
1621                 mutex_unlock(&vcpu->kvm->lock);
1622                 if (!page) {
1623                         inject_gp(vcpu);
1624                         free_pio_guest_pages(vcpu);
1625                         return 1;
1626                 }
1627         }
1628
1629         pio_dev = vcpu_find_pio_dev(vcpu, port);
1630         if (!vcpu->pio.in) {
1631                 /* string PIO write */
1632                 ret = pio_copy_data(vcpu);
1633                 if (ret >= 0 && pio_dev) {
1634                         pio_string_write(pio_dev, vcpu);
1635                         complete_pio(vcpu);
1636                         if (vcpu->pio.count == 0)
1637                                 ret = 1;
1638                 }
1639         } else if (pio_dev)
1640                 pr_unimpl(vcpu, "no string pio read support yet, "
1641                        "port %x size %d count %ld\n",
1642                         port, size, count);
1643
1644         return ret;
1645 }
1646 EXPORT_SYMBOL_GPL(kvm_emulate_pio_string);
1647
1648 int kvm_arch_init(void *opaque)
1649 {
1650         struct kvm_x86_ops *ops = (struct kvm_x86_ops *)opaque;
1651
1652         kvm_init_msr_list();
1653
1654         if (kvm_x86_ops) {
1655                 printk(KERN_ERR "kvm: already loaded the other module\n");
1656                 return -EEXIST;
1657         }
1658
1659         if (!ops->cpu_has_kvm_support()) {
1660                 printk(KERN_ERR "kvm: no hardware support\n");
1661                 return -EOPNOTSUPP;
1662         }
1663         if (ops->disabled_by_bios()) {
1664                 printk(KERN_ERR "kvm: disabled by bios\n");
1665                 return -EOPNOTSUPP;
1666         }
1667
1668         kvm_x86_ops = ops;
1669
1670         return 0;
1671 }
1672
1673 void kvm_arch_exit(void)
1674 {
1675         kvm_x86_ops = NULL;
1676  }
1677
1678 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
1679 {
1680         ++vcpu->stat.halt_exits;
1681         if (irqchip_in_kernel(vcpu->kvm)) {
1682                 vcpu->mp_state = VCPU_MP_STATE_HALTED;
1683                 kvm_vcpu_block(vcpu);
1684                 if (vcpu->mp_state != VCPU_MP_STATE_RUNNABLE)
1685                         return -EINTR;
1686                 return 1;
1687         } else {
1688                 vcpu->run->exit_reason = KVM_EXIT_HLT;
1689                 return 0;
1690         }
1691 }
1692 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
1693
1694 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
1695 {
1696         unsigned long nr, a0, a1, a2, a3, ret;
1697
1698         kvm_x86_ops->cache_regs(vcpu);
1699
1700         nr = vcpu->regs[VCPU_REGS_RAX];
1701         a0 = vcpu->regs[VCPU_REGS_RBX];
1702         a1 = vcpu->regs[VCPU_REGS_RCX];
1703         a2 = vcpu->regs[VCPU_REGS_RDX];
1704         a3 = vcpu->regs[VCPU_REGS_RSI];
1705
1706         if (!is_long_mode(vcpu)) {
1707                 nr &= 0xFFFFFFFF;
1708                 a0 &= 0xFFFFFFFF;
1709                 a1 &= 0xFFFFFFFF;
1710                 a2 &= 0xFFFFFFFF;
1711                 a3 &= 0xFFFFFFFF;
1712         }
1713
1714         switch (nr) {
1715         default:
1716                 ret = -KVM_ENOSYS;
1717                 break;
1718         }
1719         vcpu->regs[VCPU_REGS_RAX] = ret;
1720         kvm_x86_ops->decache_regs(vcpu);
1721         return 0;
1722 }
1723 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
1724
1725 int kvm_fix_hypercall(struct kvm_vcpu *vcpu)
1726 {
1727         char instruction[3];
1728         int ret = 0;
1729
1730         mutex_lock(&vcpu->kvm->lock);
1731
1732         /*
1733          * Blow out the MMU to ensure that no other VCPU has an active mapping
1734          * to ensure that the updated hypercall appears atomically across all
1735          * VCPUs.
1736          */
1737         kvm_mmu_zap_all(vcpu->kvm);
1738
1739         kvm_x86_ops->cache_regs(vcpu);
1740         kvm_x86_ops->patch_hypercall(vcpu, instruction);
1741         if (emulator_write_emulated(vcpu->rip, instruction, 3, vcpu)
1742             != X86EMUL_CONTINUE)
1743                 ret = -EFAULT;
1744
1745         mutex_unlock(&vcpu->kvm->lock);
1746
1747         return ret;
1748 }
1749
1750 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
1751 {
1752         return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
1753 }
1754
1755 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1756 {
1757         struct descriptor_table dt = { limit, base };
1758
1759         kvm_x86_ops->set_gdt(vcpu, &dt);
1760 }
1761
1762 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1763 {
1764         struct descriptor_table dt = { limit, base };
1765
1766         kvm_x86_ops->set_idt(vcpu, &dt);
1767 }
1768
1769 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
1770                    unsigned long *rflags)
1771 {
1772         lmsw(vcpu, msw);
1773         *rflags = kvm_x86_ops->get_rflags(vcpu);
1774 }
1775
1776 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
1777 {
1778         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
1779         switch (cr) {
1780         case 0:
1781                 return vcpu->cr0;
1782         case 2:
1783                 return vcpu->cr2;
1784         case 3:
1785                 return vcpu->cr3;
1786         case 4:
1787                 return vcpu->cr4;
1788         default:
1789                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1790                 return 0;
1791         }
1792 }
1793
1794 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
1795                      unsigned long *rflags)
1796 {
1797         switch (cr) {
1798         case 0:
1799                 set_cr0(vcpu, mk_cr_64(vcpu->cr0, val));
1800                 *rflags = kvm_x86_ops->get_rflags(vcpu);
1801                 break;
1802         case 2:
1803                 vcpu->cr2 = val;
1804                 break;
1805         case 3:
1806                 set_cr3(vcpu, val);
1807                 break;
1808         case 4:
1809                 set_cr4(vcpu, mk_cr_64(vcpu->cr4, val));
1810                 break;
1811         default:
1812                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1813         }
1814 }
1815
1816 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
1817 {
1818         int i;
1819         u32 function;
1820         struct kvm_cpuid_entry *e, *best;
1821
1822         kvm_x86_ops->cache_regs(vcpu);
1823         function = vcpu->regs[VCPU_REGS_RAX];
1824         vcpu->regs[VCPU_REGS_RAX] = 0;
1825         vcpu->regs[VCPU_REGS_RBX] = 0;
1826         vcpu->regs[VCPU_REGS_RCX] = 0;
1827         vcpu->regs[VCPU_REGS_RDX] = 0;
1828         best = NULL;
1829         for (i = 0; i < vcpu->cpuid_nent; ++i) {
1830                 e = &vcpu->cpuid_entries[i];
1831                 if (e->function == function) {
1832                         best = e;
1833                         break;
1834                 }
1835                 /*
1836                  * Both basic or both extended?
1837                  */
1838                 if (((e->function ^ function) & 0x80000000) == 0)
1839                         if (!best || e->function > best->function)
1840                                 best = e;
1841         }
1842         if (best) {
1843                 vcpu->regs[VCPU_REGS_RAX] = best->eax;
1844                 vcpu->regs[VCPU_REGS_RBX] = best->ebx;
1845                 vcpu->regs[VCPU_REGS_RCX] = best->ecx;
1846                 vcpu->regs[VCPU_REGS_RDX] = best->edx;
1847         }
1848         kvm_x86_ops->decache_regs(vcpu);
1849         kvm_x86_ops->skip_emulated_instruction(vcpu);
1850 }
1851 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
1852
1853 /*
1854  * Check if userspace requested an interrupt window, and that the
1855  * interrupt window is open.
1856  *
1857  * No need to exit to userspace if we already have an interrupt queued.
1858  */
1859 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu,
1860                                           struct kvm_run *kvm_run)
1861 {
1862         return (!vcpu->irq_summary &&
1863                 kvm_run->request_interrupt_window &&
1864                 vcpu->interrupt_window_open &&
1865                 (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF));
1866 }
1867
1868 static void post_kvm_run_save(struct kvm_vcpu *vcpu,
1869                               struct kvm_run *kvm_run)
1870 {
1871         kvm_run->if_flag = (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
1872         kvm_run->cr8 = get_cr8(vcpu);
1873         kvm_run->apic_base = kvm_get_apic_base(vcpu);
1874         if (irqchip_in_kernel(vcpu->kvm))
1875                 kvm_run->ready_for_interrupt_injection = 1;
1876         else
1877                 kvm_run->ready_for_interrupt_injection =
1878                                         (vcpu->interrupt_window_open &&
1879                                          vcpu->irq_summary == 0);
1880 }
1881
1882 static int __vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1883 {
1884         int r;
1885
1886         if (unlikely(vcpu->mp_state == VCPU_MP_STATE_SIPI_RECEIVED)) {
1887                 pr_debug("vcpu %d received sipi with vector # %x\n",
1888                        vcpu->vcpu_id, vcpu->sipi_vector);
1889                 kvm_lapic_reset(vcpu);
1890                 r = kvm_x86_ops->vcpu_reset(vcpu);
1891                 if (r)
1892                         return r;
1893                 vcpu->mp_state = VCPU_MP_STATE_RUNNABLE;
1894         }
1895
1896 preempted:
1897         if (vcpu->guest_debug.enabled)
1898                 kvm_x86_ops->guest_debug_pre(vcpu);
1899
1900 again:
1901         r = kvm_mmu_reload(vcpu);
1902         if (unlikely(r))
1903                 goto out;
1904
1905         kvm_inject_pending_timer_irqs(vcpu);
1906
1907         preempt_disable();
1908
1909         kvm_x86_ops->prepare_guest_switch(vcpu);
1910         kvm_load_guest_fpu(vcpu);
1911
1912         local_irq_disable();
1913
1914         if (signal_pending(current)) {
1915                 local_irq_enable();
1916                 preempt_enable();
1917                 r = -EINTR;
1918                 kvm_run->exit_reason = KVM_EXIT_INTR;
1919                 ++vcpu->stat.signal_exits;
1920                 goto out;
1921         }
1922
1923         if (irqchip_in_kernel(vcpu->kvm))
1924                 kvm_x86_ops->inject_pending_irq(vcpu);
1925         else if (!vcpu->mmio_read_completed)
1926                 kvm_x86_ops->inject_pending_vectors(vcpu, kvm_run);
1927
1928         vcpu->guest_mode = 1;
1929         kvm_guest_enter();
1930
1931         if (vcpu->requests)
1932                 if (test_and_clear_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
1933                         kvm_x86_ops->tlb_flush(vcpu);
1934
1935         kvm_x86_ops->run(vcpu, kvm_run);
1936
1937         vcpu->guest_mode = 0;
1938         local_irq_enable();
1939
1940         ++vcpu->stat.exits;
1941
1942         /*
1943          * We must have an instruction between local_irq_enable() and
1944          * kvm_guest_exit(), so the timer interrupt isn't delayed by
1945          * the interrupt shadow.  The stat.exits increment will do nicely.
1946          * But we need to prevent reordering, hence this barrier():
1947          */
1948         barrier();
1949
1950         kvm_guest_exit();
1951
1952         preempt_enable();
1953
1954         /*
1955          * Profile KVM exit RIPs:
1956          */
1957         if (unlikely(prof_on == KVM_PROFILING)) {
1958                 kvm_x86_ops->cache_regs(vcpu);
1959                 profile_hit(KVM_PROFILING, (void *)vcpu->rip);
1960         }
1961
1962         r = kvm_x86_ops->handle_exit(kvm_run, vcpu);
1963
1964         if (r > 0) {
1965                 if (dm_request_for_irq_injection(vcpu, kvm_run)) {
1966                         r = -EINTR;
1967                         kvm_run->exit_reason = KVM_EXIT_INTR;
1968                         ++vcpu->stat.request_irq_exits;
1969                         goto out;
1970                 }
1971                 if (!need_resched()) {
1972                         ++vcpu->stat.light_exits;
1973                         goto again;
1974                 }
1975         }
1976
1977 out:
1978         if (r > 0) {
1979                 kvm_resched(vcpu);
1980                 goto preempted;
1981         }
1982
1983         post_kvm_run_save(vcpu, kvm_run);
1984
1985         return r;
1986 }
1987
1988 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1989 {
1990         int r;
1991         sigset_t sigsaved;
1992
1993         vcpu_load(vcpu);
1994
1995         if (unlikely(vcpu->mp_state == VCPU_MP_STATE_UNINITIALIZED)) {
1996                 kvm_vcpu_block(vcpu);
1997                 vcpu_put(vcpu);
1998                 return -EAGAIN;
1999         }
2000
2001         if (vcpu->sigset_active)
2002                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
2003
2004         /* re-sync apic's tpr */
2005         if (!irqchip_in_kernel(vcpu->kvm))
2006                 set_cr8(vcpu, kvm_run->cr8);
2007
2008         if (vcpu->pio.cur_count) {
2009                 r = complete_pio(vcpu);
2010                 if (r)
2011                         goto out;
2012         }
2013 #if CONFIG_HAS_IOMEM
2014         if (vcpu->mmio_needed) {
2015                 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
2016                 vcpu->mmio_read_completed = 1;
2017                 vcpu->mmio_needed = 0;
2018                 r = emulate_instruction(vcpu, kvm_run,
2019                                         vcpu->mmio_fault_cr2, 0, 1);
2020                 if (r == EMULATE_DO_MMIO) {
2021                         /*
2022                          * Read-modify-write.  Back to userspace.
2023                          */
2024                         r = 0;
2025                         goto out;
2026                 }
2027         }
2028 #endif
2029         if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL) {
2030                 kvm_x86_ops->cache_regs(vcpu);
2031                 vcpu->regs[VCPU_REGS_RAX] = kvm_run->hypercall.ret;
2032                 kvm_x86_ops->decache_regs(vcpu);
2033         }
2034
2035         r = __vcpu_run(vcpu, kvm_run);
2036
2037 out:
2038         if (vcpu->sigset_active)
2039                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
2040
2041         vcpu_put(vcpu);
2042         return r;
2043 }
2044
2045 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
2046 {
2047         vcpu_load(vcpu);
2048
2049         kvm_x86_ops->cache_regs(vcpu);
2050
2051         regs->rax = vcpu->regs[VCPU_REGS_RAX];
2052         regs->rbx = vcpu->regs[VCPU_REGS_RBX];
2053         regs->rcx = vcpu->regs[VCPU_REGS_RCX];
2054         regs->rdx = vcpu->regs[VCPU_REGS_RDX];
2055         regs->rsi = vcpu->regs[VCPU_REGS_RSI];
2056         regs->rdi = vcpu->regs[VCPU_REGS_RDI];
2057         regs->rsp = vcpu->regs[VCPU_REGS_RSP];
2058         regs->rbp = vcpu->regs[VCPU_REGS_RBP];
2059 #ifdef CONFIG_X86_64
2060         regs->r8 = vcpu->regs[VCPU_REGS_R8];
2061         regs->r9 = vcpu->regs[VCPU_REGS_R9];
2062         regs->r10 = vcpu->regs[VCPU_REGS_R10];
2063         regs->r11 = vcpu->regs[VCPU_REGS_R11];
2064         regs->r12 = vcpu->regs[VCPU_REGS_R12];
2065         regs->r13 = vcpu->regs[VCPU_REGS_R13];
2066         regs->r14 = vcpu->regs[VCPU_REGS_R14];
2067         regs->r15 = vcpu->regs[VCPU_REGS_R15];
2068 #endif
2069
2070         regs->rip = vcpu->rip;
2071         regs->rflags = kvm_x86_ops->get_rflags(vcpu);
2072
2073         /*
2074          * Don't leak debug flags in case they were set for guest debugging
2075          */
2076         if (vcpu->guest_debug.enabled && vcpu->guest_debug.singlestep)
2077                 regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
2078
2079         vcpu_put(vcpu);
2080
2081         return 0;
2082 }
2083
2084 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
2085 {
2086         vcpu_load(vcpu);
2087
2088         vcpu->regs[VCPU_REGS_RAX] = regs->rax;
2089         vcpu->regs[VCPU_REGS_RBX] = regs->rbx;
2090         vcpu->regs[VCPU_REGS_RCX] = regs->rcx;
2091         vcpu->regs[VCPU_REGS_RDX] = regs->rdx;
2092         vcpu->regs[VCPU_REGS_RSI] = regs->rsi;
2093         vcpu->regs[VCPU_REGS_RDI] = regs->rdi;
2094         vcpu->regs[VCPU_REGS_RSP] = regs->rsp;
2095         vcpu->regs[VCPU_REGS_RBP] = regs->rbp;
2096 #ifdef CONFIG_X86_64
2097         vcpu->regs[VCPU_REGS_R8] = regs->r8;
2098         vcpu->regs[VCPU_REGS_R9] = regs->r9;
2099         vcpu->regs[VCPU_REGS_R10] = regs->r10;
2100         vcpu->regs[VCPU_REGS_R11] = regs->r11;
2101         vcpu->regs[VCPU_REGS_R12] = regs->r12;
2102         vcpu->regs[VCPU_REGS_R13] = regs->r13;
2103         vcpu->regs[VCPU_REGS_R14] = regs->r14;
2104         vcpu->regs[VCPU_REGS_R15] = regs->r15;
2105 #endif
2106
2107         vcpu->rip = regs->rip;
2108         kvm_x86_ops->set_rflags(vcpu, regs->rflags);
2109
2110         kvm_x86_ops->decache_regs(vcpu);
2111
2112         vcpu_put(vcpu);
2113
2114         return 0;
2115 }
2116
2117 static void get_segment(struct kvm_vcpu *vcpu,
2118                         struct kvm_segment *var, int seg)
2119 {
2120         return kvm_x86_ops->get_segment(vcpu, var, seg);
2121 }
2122
2123 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
2124 {
2125         struct kvm_segment cs;
2126
2127         get_segment(vcpu, &cs, VCPU_SREG_CS);
2128         *db = cs.db;
2129         *l = cs.l;
2130 }
2131 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
2132
2133 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
2134                                   struct kvm_sregs *sregs)
2135 {
2136         struct descriptor_table dt;
2137         int pending_vec;
2138
2139         vcpu_load(vcpu);
2140
2141         get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
2142         get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
2143         get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
2144         get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
2145         get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
2146         get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
2147
2148         get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
2149         get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2150
2151         kvm_x86_ops->get_idt(vcpu, &dt);
2152         sregs->idt.limit = dt.limit;
2153         sregs->idt.base = dt.base;
2154         kvm_x86_ops->get_gdt(vcpu, &dt);
2155         sregs->gdt.limit = dt.limit;
2156         sregs->gdt.base = dt.base;
2157
2158         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
2159         sregs->cr0 = vcpu->cr0;
2160         sregs->cr2 = vcpu->cr2;
2161         sregs->cr3 = vcpu->cr3;
2162         sregs->cr4 = vcpu->cr4;
2163         sregs->cr8 = get_cr8(vcpu);
2164         sregs->efer = vcpu->shadow_efer;
2165         sregs->apic_base = kvm_get_apic_base(vcpu);
2166
2167         if (irqchip_in_kernel(vcpu->kvm)) {
2168                 memset(sregs->interrupt_bitmap, 0,
2169                        sizeof sregs->interrupt_bitmap);
2170                 pending_vec = kvm_x86_ops->get_irq(vcpu);
2171                 if (pending_vec >= 0)
2172                         set_bit(pending_vec,
2173                                 (unsigned long *)sregs->interrupt_bitmap);
2174         } else
2175                 memcpy(sregs->interrupt_bitmap, vcpu->irq_pending,
2176                        sizeof sregs->interrupt_bitmap);
2177
2178         vcpu_put(vcpu);
2179
2180         return 0;
2181 }
2182
2183 static void set_segment(struct kvm_vcpu *vcpu,
2184                         struct kvm_segment *var, int seg)
2185 {
2186         return kvm_x86_ops->set_segment(vcpu, var, seg);
2187 }
2188
2189 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
2190                                   struct kvm_sregs *sregs)
2191 {
2192         int mmu_reset_needed = 0;
2193         int i, pending_vec, max_bits;
2194         struct descriptor_table dt;
2195
2196         vcpu_load(vcpu);
2197
2198         dt.limit = sregs->idt.limit;
2199         dt.base = sregs->idt.base;
2200         kvm_x86_ops->set_idt(vcpu, &dt);
2201         dt.limit = sregs->gdt.limit;
2202         dt.base = sregs->gdt.base;
2203         kvm_x86_ops->set_gdt(vcpu, &dt);
2204
2205         vcpu->cr2 = sregs->cr2;
2206         mmu_reset_needed |= vcpu->cr3 != sregs->cr3;
2207         vcpu->cr3 = sregs->cr3;
2208
2209         set_cr8(vcpu, sregs->cr8);
2210
2211         mmu_reset_needed |= vcpu->shadow_efer != sregs->efer;
2212 #ifdef CONFIG_X86_64
2213         kvm_x86_ops->set_efer(vcpu, sregs->efer);
2214 #endif
2215         kvm_set_apic_base(vcpu, sregs->apic_base);
2216
2217         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
2218
2219         mmu_reset_needed |= vcpu->cr0 != sregs->cr0;
2220         vcpu->cr0 = sregs->cr0;
2221         kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
2222
2223         mmu_reset_needed |= vcpu->cr4 != sregs->cr4;
2224         kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
2225         if (!is_long_mode(vcpu) && is_pae(vcpu))
2226                 load_pdptrs(vcpu, vcpu->cr3);
2227
2228         if (mmu_reset_needed)
2229                 kvm_mmu_reset_context(vcpu);
2230
2231         if (!irqchip_in_kernel(vcpu->kvm)) {
2232                 memcpy(vcpu->irq_pending, sregs->interrupt_bitmap,
2233                        sizeof vcpu->irq_pending);
2234                 vcpu->irq_summary = 0;
2235                 for (i = 0; i < ARRAY_SIZE(vcpu->irq_pending); ++i)
2236                         if (vcpu->irq_pending[i])
2237                                 __set_bit(i, &vcpu->irq_summary);
2238         } else {
2239                 max_bits = (sizeof sregs->interrupt_bitmap) << 3;
2240                 pending_vec = find_first_bit(
2241                         (const unsigned long *)sregs->interrupt_bitmap,
2242                         max_bits);
2243                 /* Only pending external irq is handled here */
2244                 if (pending_vec < max_bits) {
2245                         kvm_x86_ops->set_irq(vcpu, pending_vec);
2246                         pr_debug("Set back pending irq %d\n",
2247                                  pending_vec);
2248                 }
2249         }
2250
2251         set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
2252         set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
2253         set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
2254         set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
2255         set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
2256         set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
2257
2258         set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
2259         set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2260
2261         vcpu_put(vcpu);
2262
2263         return 0;
2264 }
2265
2266 int kvm_arch_vcpu_ioctl_debug_guest(struct kvm_vcpu *vcpu,
2267                                     struct kvm_debug_guest *dbg)
2268 {
2269         int r;
2270
2271         vcpu_load(vcpu);
2272
2273         r = kvm_x86_ops->set_guest_debug(vcpu, dbg);
2274
2275         vcpu_put(vcpu);
2276
2277         return r;
2278 }
2279
2280 /*
2281  * fxsave fpu state.  Taken from x86_64/processor.h.  To be killed when
2282  * we have asm/x86/processor.h
2283  */
2284 struct fxsave {
2285         u16     cwd;
2286         u16     swd;
2287         u16     twd;
2288         u16     fop;
2289         u64     rip;
2290         u64     rdp;
2291         u32     mxcsr;
2292         u32     mxcsr_mask;
2293         u32     st_space[32];   /* 8*16 bytes for each FP-reg = 128 bytes */
2294 #ifdef CONFIG_X86_64
2295         u32     xmm_space[64];  /* 16*16 bytes for each XMM-reg = 256 bytes */
2296 #else
2297         u32     xmm_space[32];  /* 8*16 bytes for each XMM-reg = 128 bytes */
2298 #endif
2299 };
2300
2301 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2302 {
2303         struct fxsave *fxsave = (struct fxsave *)&vcpu->guest_fx_image;
2304
2305         vcpu_load(vcpu);
2306
2307         memcpy(fpu->fpr, fxsave->st_space, 128);
2308         fpu->fcw = fxsave->cwd;
2309         fpu->fsw = fxsave->swd;
2310         fpu->ftwx = fxsave->twd;
2311         fpu->last_opcode = fxsave->fop;
2312         fpu->last_ip = fxsave->rip;
2313         fpu->last_dp = fxsave->rdp;
2314         memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
2315
2316         vcpu_put(vcpu);
2317
2318         return 0;
2319 }
2320
2321 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2322 {
2323         struct fxsave *fxsave = (struct fxsave *)&vcpu->guest_fx_image;
2324
2325         vcpu_load(vcpu);
2326
2327         memcpy(fxsave->st_space, fpu->fpr, 128);
2328         fxsave->cwd = fpu->fcw;
2329         fxsave->swd = fpu->fsw;
2330         fxsave->twd = fpu->ftwx;
2331         fxsave->fop = fpu->last_opcode;
2332         fxsave->rip = fpu->last_ip;
2333         fxsave->rdp = fpu->last_dp;
2334         memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
2335
2336         vcpu_put(vcpu);
2337
2338         return 0;
2339 }
2340
2341 void fx_init(struct kvm_vcpu *vcpu)
2342 {
2343         unsigned after_mxcsr_mask;
2344
2345         /* Initialize guest FPU by resetting ours and saving into guest's */
2346         preempt_disable();
2347         fx_save(&vcpu->host_fx_image);
2348         fpu_init();
2349         fx_save(&vcpu->guest_fx_image);
2350         fx_restore(&vcpu->host_fx_image);
2351         preempt_enable();
2352
2353         vcpu->cr0 |= X86_CR0_ET;
2354         after_mxcsr_mask = offsetof(struct i387_fxsave_struct, st_space);
2355         vcpu->guest_fx_image.mxcsr = 0x1f80;
2356         memset((void *)&vcpu->guest_fx_image + after_mxcsr_mask,
2357                0, sizeof(struct i387_fxsave_struct) - after_mxcsr_mask);
2358 }
2359 EXPORT_SYMBOL_GPL(fx_init);
2360
2361 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
2362 {
2363         if (!vcpu->fpu_active || vcpu->guest_fpu_loaded)
2364                 return;
2365
2366         vcpu->guest_fpu_loaded = 1;
2367         fx_save(&vcpu->host_fx_image);
2368         fx_restore(&vcpu->guest_fx_image);
2369 }
2370 EXPORT_SYMBOL_GPL(kvm_load_guest_fpu);
2371
2372 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
2373 {
2374         if (!vcpu->guest_fpu_loaded)
2375                 return;
2376
2377         vcpu->guest_fpu_loaded = 0;
2378         fx_save(&vcpu->guest_fx_image);
2379         fx_restore(&vcpu->host_fx_image);
2380 }
2381 EXPORT_SYMBOL_GPL(kvm_put_guest_fpu);
2382
2383 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
2384 {
2385         kvm_x86_ops->vcpu_free(vcpu);
2386 }
2387
2388 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
2389                                                 unsigned int id)
2390 {
2391         int r;
2392         struct kvm_vcpu *vcpu = kvm_x86_ops->vcpu_create(kvm, id);
2393
2394         if (IS_ERR(vcpu)) {
2395                 r = -ENOMEM;
2396                 goto fail;
2397         }
2398
2399         /* We do fxsave: this must be aligned. */
2400         BUG_ON((unsigned long)&vcpu->host_fx_image & 0xF);
2401
2402         vcpu_load(vcpu);
2403         r = kvm_arch_vcpu_reset(vcpu);
2404         if (r == 0)
2405                 r = kvm_mmu_setup(vcpu);
2406         vcpu_put(vcpu);
2407         if (r < 0)
2408                 goto free_vcpu;
2409
2410         return vcpu;
2411 free_vcpu:
2412         kvm_x86_ops->vcpu_free(vcpu);
2413 fail:
2414         return ERR_PTR(r);
2415 }
2416
2417 void kvm_arch_vcpu_destory(struct kvm_vcpu *vcpu)
2418 {
2419         vcpu_load(vcpu);
2420         kvm_mmu_unload(vcpu);
2421         vcpu_put(vcpu);
2422
2423         kvm_x86_ops->vcpu_free(vcpu);
2424 }
2425
2426 int kvm_arch_vcpu_reset(struct kvm_vcpu *vcpu)
2427 {
2428         return kvm_x86_ops->vcpu_reset(vcpu);
2429 }
2430
2431 void kvm_arch_hardware_enable(void *garbage)
2432 {
2433         kvm_x86_ops->hardware_enable(garbage);
2434 }
2435
2436 void kvm_arch_hardware_disable(void *garbage)
2437 {
2438         kvm_x86_ops->hardware_disable(garbage);
2439 }
2440
2441 int kvm_arch_hardware_setup(void)
2442 {
2443         return kvm_x86_ops->hardware_setup();
2444 }
2445
2446 void kvm_arch_hardware_unsetup(void)
2447 {
2448         kvm_x86_ops->hardware_unsetup();
2449 }
2450
2451 void kvm_arch_check_processor_compat(void *rtn)
2452 {
2453         kvm_x86_ops->check_processor_compatibility(rtn);
2454 }
2455
2456 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
2457 {
2458         struct page *page;
2459         struct kvm *kvm;
2460         int r;
2461
2462         BUG_ON(vcpu->kvm == NULL);
2463         kvm = vcpu->kvm;
2464
2465         vcpu->mmu.root_hpa = INVALID_PAGE;
2466         if (!irqchip_in_kernel(kvm) || vcpu->vcpu_id == 0)
2467                 vcpu->mp_state = VCPU_MP_STATE_RUNNABLE;
2468         else
2469                 vcpu->mp_state = VCPU_MP_STATE_UNINITIALIZED;
2470
2471         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2472         if (!page) {
2473                 r = -ENOMEM;
2474                 goto fail;
2475         }
2476         vcpu->pio_data = page_address(page);
2477
2478         r = kvm_mmu_create(vcpu);
2479         if (r < 0)
2480                 goto fail_free_pio_data;
2481
2482         if (irqchip_in_kernel(kvm)) {
2483                 r = kvm_create_lapic(vcpu);
2484                 if (r < 0)
2485                         goto fail_mmu_destroy;
2486         }
2487
2488         return 0;
2489
2490 fail_mmu_destroy:
2491         kvm_mmu_destroy(vcpu);
2492 fail_free_pio_data:
2493         free_page((unsigned long)vcpu->pio_data);
2494 fail:
2495         return r;
2496 }
2497
2498 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
2499 {
2500         kvm_free_lapic(vcpu);
2501         kvm_mmu_destroy(vcpu);
2502         free_page((unsigned long)vcpu->pio_data);
2503 }