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 VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
  46 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
  47
  48 struct kvm_x86_ops *kvm_x86_ops;
  49
  50 struct kvm_stats_debugfs_item debugfs_entries[] = {
  51         { "pf_fixed", VCPU_STAT(pf_fixed) },
  52         { "pf_guest", VCPU_STAT(pf_guest) },
  53         { "tlb_flush", VCPU_STAT(tlb_flush) },
  54         { "invlpg", VCPU_STAT(invlpg) },
  55         { "exits", VCPU_STAT(exits) },
  56         { "io_exits", VCPU_STAT(io_exits) },
  57         { "mmio_exits", VCPU_STAT(mmio_exits) },
  58         { "signal_exits", VCPU_STAT(signal_exits) },
  59         { "irq_window", VCPU_STAT(irq_window_exits) },
  60         { "halt_exits", VCPU_STAT(halt_exits) },
  61         { "halt_wakeup", VCPU_STAT(halt_wakeup) },
  62         { "request_irq", VCPU_STAT(request_irq_exits) },
  63         { "irq_exits", VCPU_STAT(irq_exits) },
  64         { "host_state_reload", VCPU_STAT(host_state_reload) },
  65         { "efer_reload", VCPU_STAT(efer_reload) },
  66         { "fpu_reload", VCPU_STAT(fpu_reload) },
  67         { "insn_emulation", VCPU_STAT(insn_emulation) },
  68         { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
  69         { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
  70         { "mmu_pte_write", VM_STAT(mmu_pte_write) },
  71         { "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
  72         { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
  73         { "mmu_flooded", VM_STAT(mmu_flooded) },
  74         { "mmu_recycled", VM_STAT(mmu_recycled) },
  75         { NULL }
  76 };
  77
  78
  79 unsigned long segment_base(u16 selector)
  80 {
  81         struct descriptor_table gdt;
  82         struct segment_descriptor *d;
  83         unsigned long table_base;
  84         unsigned long v;
  85
  86         if (selector == 0)
  87                 return 0;
  88
  89         asm("sgdt %0" : "=m"(gdt));
  90         table_base = gdt.base;
  91
  92         if (selector & 4) {           /* from ldt */
  93                 u16 ldt_selector;
  94
  95                 asm("sldt %0" : "=g"(ldt_selector));
  96                 table_base = segment_base(ldt_selector);
  97         }
  98         d = (struct segment_descriptor *)(table_base + (selector & ~7));
  99         v = d->base_low | ((unsigned long)d->base_mid << 16) |
 100                 ((unsigned long)d->base_high << 24);
 101 #ifdef CONFIG_X86_64
 102         if (d->system == 0 && (d->type == 2 || d->type == 9 || d->type == 11))
 103                 v |= ((unsigned long) \
 104                       ((struct segment_descriptor_64 *)d)->base_higher) << 32;
 105 #endif
 106         return v;
 107 }
 108 EXPORT_SYMBOL_GPL(segment_base);
 109
 110 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
 111 {
 112         if (irqchip_in_kernel(vcpu->kvm))
 113                 return vcpu->apic_base;
 114         else
 115                 return vcpu->apic_base;
 116 }
 117 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
 118
 119 void kvm_set_apic_base(struct kvm_vcpu *vcpu, u64 data)
 120 {
 121         /* TODO: reserve bits check */
 122         if (irqchip_in_kernel(vcpu->kvm))
 123                 kvm_lapic_set_base(vcpu, data);
 124         else
 125                 vcpu->apic_base = data;
 126 }
 127 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
 128
 129 static void inject_gp(struct kvm_vcpu *vcpu)
 130 {
 131         kvm_x86_ops->inject_gp(vcpu, 0);
 132 }
 133
 134 /*
 135  * Load the pae pdptrs.  Return true is they are all valid.
 136  */
 137 int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
 138 {
 139         gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
 140         unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
 141         int i;
 142         int ret;
 143         u64 pdpte[ARRAY_SIZE(vcpu->pdptrs)];
 144
 145         mutex_lock(&vcpu->kvm->lock);
 146         ret = kvm_read_guest_page(vcpu->kvm, pdpt_gfn, pdpte,
 147                                   offset * sizeof(u64), sizeof(pdpte));
 148         if (ret < 0) {
 149                 ret = 0;
 150                 goto out;
 151         }
 152         for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
 153                 if ((pdpte[i] & 1) && (pdpte[i] & 0xfffffff0000001e6ull)) {
 154                         ret = 0;
 155                         goto out;
 156                 }
 157         }
 158         ret = 1;
 159
 160         memcpy(vcpu->pdptrs, pdpte, sizeof(vcpu->pdptrs));
 161 out:
 162         mutex_unlock(&vcpu->kvm->lock);
 163
 164         return ret;
 165 }
 166
 167 void set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
 168 {
 169         if (cr0 & CR0_RESERVED_BITS) {
 170                 printk(KERN_DEBUG "set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
 171                        cr0, vcpu->cr0);
 172                 inject_gp(vcpu);
 173                 return;
 174         }
 175
 176         if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD)) {
 177                 printk(KERN_DEBUG "set_cr0: #GP, CD == 0 && NW == 1\n");
 178                 inject_gp(vcpu);
 179                 return;
 180         }
 181
 182         if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE)) {
 183                 printk(KERN_DEBUG "set_cr0: #GP, set PG flag "
 184                        "and a clear PE flag\n");
 185                 inject_gp(vcpu);
 186                 return;
 187         }
 188
 189         if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
 190 #ifdef CONFIG_X86_64
 191                 if ((vcpu->shadow_efer & EFER_LME)) {
 192                         int cs_db, cs_l;
 193
 194                         if (!is_pae(vcpu)) {
 195                                 printk(KERN_DEBUG "set_cr0: #GP, start paging "
 196                                        "in long mode while PAE is disabled\n");
 197                                 inject_gp(vcpu);
 198                                 return;
 199                         }
 200                         kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
 201                         if (cs_l) {
 202                                 printk(KERN_DEBUG "set_cr0: #GP, start paging "
 203                                        "in long mode while CS.L == 1\n");
 204                                 inject_gp(vcpu);
 205                                 return;
 206
 207                         }
 208                 } else
 209 #endif
 210                 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->cr3)) {
 211                         printk(KERN_DEBUG "set_cr0: #GP, pdptrs "
 212                                "reserved bits\n");
 213                         inject_gp(vcpu);
 214                         return;
 215                 }
 216
 217         }
 218
 219         kvm_x86_ops->set_cr0(vcpu, cr0);
 220         vcpu->cr0 = cr0;
 221
 222         mutex_lock(&vcpu->kvm->lock);
 223         kvm_mmu_reset_context(vcpu);
 224         mutex_unlock(&vcpu->kvm->lock);
 225         return;
 226 }
 227 EXPORT_SYMBOL_GPL(set_cr0);
 228
 229 void lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
 230 {
 231         set_cr0(vcpu, (vcpu->cr0 & ~0x0ful) | (msw & 0x0f));
 232 }
 233 EXPORT_SYMBOL_GPL(lmsw);
 234
 235 void set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
 236 {
 237         if (cr4 & CR4_RESERVED_BITS) {
 238                 printk(KERN_DEBUG "set_cr4: #GP, reserved bits\n");
 239                 inject_gp(vcpu);
 240                 return;
 241         }
 242
 243         if (is_long_mode(vcpu)) {
 244                 if (!(cr4 & X86_CR4_PAE)) {
 245                         printk(KERN_DEBUG "set_cr4: #GP, clearing PAE while "
 246                                "in long mode\n");
 247                         inject_gp(vcpu);
 248                         return;
 249                 }
 250         } else if (is_paging(vcpu) && !is_pae(vcpu) && (cr4 & X86_CR4_PAE)
 251                    && !load_pdptrs(vcpu, vcpu->cr3)) {
 252                 printk(KERN_DEBUG "set_cr4: #GP, pdptrs reserved bits\n");
 253                 inject_gp(vcpu);
 254                 return;
 255         }
 256
 257         if (cr4 & X86_CR4_VMXE) {
 258                 printk(KERN_DEBUG "set_cr4: #GP, setting VMXE\n");
 259                 inject_gp(vcpu);
 260                 return;
 261         }
 262         kvm_x86_ops->set_cr4(vcpu, cr4);
 263         vcpu->cr4 = cr4;
 264         mutex_lock(&vcpu->kvm->lock);
 265         kvm_mmu_reset_context(vcpu);
 266         mutex_unlock(&vcpu->kvm->lock);
 267 }
 268 EXPORT_SYMBOL_GPL(set_cr4);
 269
 270 void set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
 271 {
 272         if (is_long_mode(vcpu)) {
 273                 if (cr3 & CR3_L_MODE_RESERVED_BITS) {
 274                         printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
 275                         inject_gp(vcpu);
 276                         return;
 277                 }
 278         } else {
 279                 if (is_pae(vcpu)) {
 280                         if (cr3 & CR3_PAE_RESERVED_BITS) {
 281                                 printk(KERN_DEBUG
 282                                        "set_cr3: #GP, reserved bits\n");
 283                                 inject_gp(vcpu);
 284                                 return;
 285                         }
 286                         if (is_paging(vcpu) && !load_pdptrs(vcpu, cr3)) {
 287                                 printk(KERN_DEBUG "set_cr3: #GP, pdptrs "
 288                                        "reserved bits\n");
 289                                 inject_gp(vcpu);
 290                                 return;
 291                         }
 292                 }
 293                 /*
 294                  * We don't check reserved bits in nonpae mode, because
 295                  * this isn't enforced, and VMware depends on this.
 296                  */
 297         }
 298
 299         mutex_lock(&vcpu->kvm->lock);
 300         /*
 301          * Does the new cr3 value map to physical memory? (Note, we
 302          * catch an invalid cr3 even in real-mode, because it would
 303          * cause trouble later on when we turn on paging anyway.)
 304          *
 305          * A real CPU would silently accept an invalid cr3 and would
 306          * attempt to use it - with largely undefined (and often hard
 307          * to debug) behavior on the guest side.
 308          */
 309         if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
 310                 inject_gp(vcpu);
 311         else {
 312                 vcpu->cr3 = cr3;
 313                 vcpu->mmu.new_cr3(vcpu);
 314         }
 315         mutex_unlock(&vcpu->kvm->lock);
 316 }
 317 EXPORT_SYMBOL_GPL(set_cr3);
 318
 319 void set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
 320 {
 321         if (cr8 & CR8_RESERVED_BITS) {
 322                 printk(KERN_DEBUG "set_cr8: #GP, reserved bits 0x%lx\n", cr8);
 323                 inject_gp(vcpu);
 324                 return;
 325         }
 326         if (irqchip_in_kernel(vcpu->kvm))
 327                 kvm_lapic_set_tpr(vcpu, cr8);
 328         else
 329                 vcpu->cr8 = cr8;
 330 }
 331 EXPORT_SYMBOL_GPL(set_cr8);
 332
 333 unsigned long get_cr8(struct kvm_vcpu *vcpu)
 334 {
 335         if (irqchip_in_kernel(vcpu->kvm))
 336                 return kvm_lapic_get_cr8(vcpu);
 337         else
 338                 return vcpu->cr8;
 339 }
 340 EXPORT_SYMBOL_GPL(get_cr8);
 341
 342 /*
 343  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
 344  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
 345  *
 346  * This list is modified at module load time to reflect the
 347  * capabilities of the host cpu.
 348  */
 349 static u32 msrs_to_save[] = {
 350         MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
 351         MSR_K6_STAR,
 352 #ifdef CONFIG_X86_64
 353         MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
 354 #endif
 355         MSR_IA32_TIME_STAMP_COUNTER,
 356 };
 357
 358 static unsigned num_msrs_to_save;
 359
 360 static u32 emulated_msrs[] = {
 361         MSR_IA32_MISC_ENABLE,
 362 };
 363
 364 #ifdef CONFIG_X86_64
 365
 366 static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
 367 {
 368         if (efer & EFER_RESERVED_BITS) {
 369                 printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
 370                        efer);
 371                 inject_gp(vcpu);
 372                 return;
 373         }
 374
 375         if (is_paging(vcpu)
 376             && (vcpu->shadow_efer & EFER_LME) != (efer & EFER_LME)) {
 377                 printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
 378                 inject_gp(vcpu);
 379                 return;
 380         }
 381
 382         kvm_x86_ops->set_efer(vcpu, efer);
 383
 384         efer &= ~EFER_LMA;
 385         efer |= vcpu->shadow_efer & EFER_LMA;
 386
 387         vcpu->shadow_efer = efer;
 388 }
 389
 390 #endif
 391
 392 /*
 393  * Writes msr value into into the appropriate "register".
 394  * Returns 0 on success, non-0 otherwise.
 395  * Assumes vcpu_load() was already called.
 396  */
 397 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
 398 {
 399         return kvm_x86_ops->set_msr(vcpu, msr_index, data);
 400 }
 401
 402 /*
 403  * Adapt set_msr() to msr_io()'s calling convention
 404  */
 405 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
 406 {
 407         return kvm_set_msr(vcpu, index, *data);
 408 }
 409
 410
 411 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
 412 {
 413         switch (msr) {
 414 #ifdef CONFIG_X86_64
 415         case MSR_EFER:
 416                 set_efer(vcpu, data);
 417                 break;
 418 #endif
 419         case MSR_IA32_MC0_STATUS:
 420                 pr_unimpl(vcpu, "%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n",
 421                        __FUNCTION__, data);
 422                 break;
 423         case MSR_IA32_MCG_STATUS:
 424                 pr_unimpl(vcpu, "%s: MSR_IA32_MCG_STATUS 0x%llx, nop\n",
 425                         __FUNCTION__, data);
 426                 break;
 427         case MSR_IA32_UCODE_REV:
 428         case MSR_IA32_UCODE_WRITE:
 429         case 0x200 ... 0x2ff: /* MTRRs */
 430                 break;
 431         case MSR_IA32_APICBASE:
 432                 kvm_set_apic_base(vcpu, data);
 433                 break;
 434         case MSR_IA32_MISC_ENABLE:
 435                 vcpu->ia32_misc_enable_msr = data;
 436                 break;
 437         default:
 438                 pr_unimpl(vcpu, "unhandled wrmsr: 0x%x\n", msr);
 439                 return 1;
 440         }
 441         return 0;
 442 }
 443 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
 444
 445
 446 /*
 447  * Reads an msr value (of 'msr_index') into 'pdata'.
 448  * Returns 0 on success, non-0 otherwise.
 449  * Assumes vcpu_load() was already called.
 450  */
 451 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
 452 {
 453         return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
 454 }
 455
 456 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
 457 {
 458         u64 data;
 459
 460         switch (msr) {
 461         case 0xc0010010: /* SYSCFG */
 462         case 0xc0010015: /* HWCR */
 463         case MSR_IA32_PLATFORM_ID:
 464         case MSR_IA32_P5_MC_ADDR:
 465         case MSR_IA32_P5_MC_TYPE:
 466         case MSR_IA32_MC0_CTL:
 467         case MSR_IA32_MCG_STATUS:
 468         case MSR_IA32_MCG_CAP:
 469         case MSR_IA32_MC0_MISC:
 470         case MSR_IA32_MC0_MISC+4:
 471         case MSR_IA32_MC0_MISC+8:
 472         case MSR_IA32_MC0_MISC+12:
 473         case MSR_IA32_MC0_MISC+16:
 474         case MSR_IA32_UCODE_REV:
 475         case MSR_IA32_PERF_STATUS:
 476         case MSR_IA32_EBL_CR_POWERON:
 477                 /* MTRR registers */
 478         case 0xfe:
 479         case 0x200 ... 0x2ff:
 480                 data = 0;
 481                 break;
 482         case 0xcd: /* fsb frequency */
 483                 data = 3;
 484                 break;
 485         case MSR_IA32_APICBASE:
 486                 data = kvm_get_apic_base(vcpu);
 487                 break;
 488         case MSR_IA32_MISC_ENABLE:
 489                 data = vcpu->ia32_misc_enable_msr;
 490                 break;
 491 #ifdef CONFIG_X86_64
 492         case MSR_EFER:
 493                 data = vcpu->shadow_efer;
 494                 break;
 495 #endif
 496         default:
 497                 pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
 498                 return 1;
 499         }
 500         *pdata = data;
 501         return 0;
 502 }
 503 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
 504
 505 /*
 506  * Read or write a bunch of msrs. All parameters are kernel addresses.
 507  *
 508  * @return number of msrs set successfully.
 509  */
 510 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
 511                     struct kvm_msr_entry *entries,
 512                     int (*do_msr)(struct kvm_vcpu *vcpu,
 513                                   unsigned index, u64 *data))
 514 {
 515         int i;
 516
 517         vcpu_load(vcpu);
 518
 519         for (i = 0; i < msrs->nmsrs; ++i)
 520                 if (do_msr(vcpu, entries[i].index, &entries[i].data))
 521                         break;
 522
 523         vcpu_put(vcpu);
 524
 525         return i;
 526 }
 527
 528 /*
 529  * Read or write a bunch of msrs. Parameters are user addresses.
 530  *
 531  * @return number of msrs set successfully.
 532  */
 533 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
 534                   int (*do_msr)(struct kvm_vcpu *vcpu,
 535                                 unsigned index, u64 *data),
 536                   int writeback)
 537 {
 538         struct kvm_msrs msrs;
 539         struct kvm_msr_entry *entries;
 540         int r, n;
 541         unsigned size;
 542
 543         r = -EFAULT;
 544         if (copy_from_user(&msrs, user_msrs, sizeof msrs))
 545                 goto out;
 546
 547         r = -E2BIG;
 548         if (msrs.nmsrs >= MAX_IO_MSRS)
 549                 goto out;
 550
 551         r = -ENOMEM;
 552         size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
 553         entries = vmalloc(size);
 554         if (!entries)
 555                 goto out;
 556
 557         r = -EFAULT;
 558         if (copy_from_user(entries, user_msrs->entries, size))
 559                 goto out_free;
 560
 561         r = n = __msr_io(vcpu, &msrs, entries, do_msr);
 562         if (r < 0)
 563                 goto out_free;
 564
 565         r = -EFAULT;
 566         if (writeback && copy_to_user(user_msrs->entries, entries, size))
 567                 goto out_free;
 568
 569         r = n;
 570
 571 out_free:
 572         vfree(entries);
 573 out:
 574         return r;
 575 }
 576
 577 /*
 578  * Make sure that a cpu that is being hot-unplugged does not have any vcpus
 579  * cached on it.
 580  */
 581 void decache_vcpus_on_cpu(int cpu)
 582 {
 583         struct kvm *vm;
 584         struct kvm_vcpu *vcpu;
 585         int i;
 586
 587         spin_lock(&kvm_lock);
 588         list_for_each_entry(vm, &vm_list, vm_list)
 589                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 590                         vcpu = vm->vcpus[i];
 591                         if (!vcpu)
 592                                 continue;
 593                         /*
 594                          * If the vcpu is locked, then it is running on some
 595                          * other cpu and therefore it is not cached on the
 596                          * cpu in question.
 597                          *
 598                          * If it's not locked, check the last cpu it executed
 599                          * on.
 600                          */
 601                         if (mutex_trylock(&vcpu->mutex)) {
 602                                 if (vcpu->cpu == cpu) {
 603                                         kvm_x86_ops->vcpu_decache(vcpu);
 604                                         vcpu->cpu = -1;
 605                                 }
 606                                 mutex_unlock(&vcpu->mutex);
 607                         }
 608                 }
 609         spin_unlock(&kvm_lock);
 610 }
 611
 612 int kvm_dev_ioctl_check_extension(long ext)
 613 {
 614         int r;
 615
 616         switch (ext) {
 617         case KVM_CAP_IRQCHIP:
 618         case KVM_CAP_HLT:
 619         case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
 620         case KVM_CAP_USER_MEMORY:
 621         case KVM_CAP_SET_TSS_ADDR:
 622                 r = 1;
 623                 break;
 624         default:
 625                 r = 0;
 626                 break;
 627         }
 628         return r;
 629
 630 }
 631
 632 long kvm_arch_dev_ioctl(struct file *filp,
 633                         unsigned int ioctl, unsigned long arg)
 634 {
 635         void __user *argp = (void __user *)arg;
 636         long r;
 637
 638         switch (ioctl) {
 639         case KVM_GET_MSR_INDEX_LIST: {
 640                 struct kvm_msr_list __user *user_msr_list = argp;
 641                 struct kvm_msr_list msr_list;
 642                 unsigned n;
 643
 644                 r = -EFAULT;
 645                 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
 646                         goto out;
 647                 n = msr_list.nmsrs;
 648                 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
 649                 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
 650                         goto out;
 651                 r = -E2BIG;
 652                 if (n < num_msrs_to_save)
 653                         goto out;
 654                 r = -EFAULT;
 655                 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
 656                                  num_msrs_to_save * sizeof(u32)))
 657                         goto out;
 658                 if (copy_to_user(user_msr_list->indices
 659                                  + num_msrs_to_save * sizeof(u32),
 660                                  &emulated_msrs,
 661                                  ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
 662                         goto out;
 663                 r = 0;
 664                 break;
 665         }
 666         default:
 667                 r = -EINVAL;
 668         }
 669 out:
 670         return r;
 671 }
 672
 673 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
 674 {
 675         kvm_x86_ops->vcpu_load(vcpu, cpu);
 676 }
 677
 678 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
 679 {
 680         kvm_x86_ops->vcpu_put(vcpu);
 681         kvm_put_guest_fpu(vcpu);
 682 }
 683
 684 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
 685 {
 686         u64 efer;
 687         int i;
 688         struct kvm_cpuid_entry *e, *entry;
 689
 690         rdmsrl(MSR_EFER, efer);
 691         entry = NULL;
 692         for (i = 0; i < vcpu->cpuid_nent; ++i) {
 693                 e = &vcpu->cpuid_entries[i];
 694                 if (e->function == 0x80000001) {
 695                         entry = e;
 696                         break;
 697                 }
 698         }
 699         if (entry && (entry->edx & (1 << 20)) && !(efer & EFER_NX)) {
 700                 entry->edx &= ~(1 << 20);
 701                 printk(KERN_INFO "kvm: guest NX capability removed\n");
 702         }
 703 }
 704
 705 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
 706                                     struct kvm_cpuid *cpuid,
 707                                     struct kvm_cpuid_entry __user *entries)
 708 {
 709         int r;
 710
 711         r = -E2BIG;
 712         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
 713                 goto out;
 714         r = -EFAULT;
 715         if (copy_from_user(&vcpu->cpuid_entries, entries,
 716                            cpuid->nent * sizeof(struct kvm_cpuid_entry)))
 717                 goto out;
 718         vcpu->cpuid_nent = cpuid->nent;
 719         cpuid_fix_nx_cap(vcpu);
 720         return 0;
 721
 722 out:
 723         return r;
 724 }
 725
 726 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
 727                                     struct kvm_lapic_state *s)
 728 {
 729         vcpu_load(vcpu);
 730         memcpy(s->regs, vcpu->apic->regs, sizeof *s);
 731         vcpu_put(vcpu);
 732
 733         return 0;
 734 }
 735
 736 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
 737                                     struct kvm_lapic_state *s)
 738 {
 739         vcpu_load(vcpu);
 740         memcpy(vcpu->apic->regs, s->regs, sizeof *s);
 741         kvm_apic_post_state_restore(vcpu);
 742         vcpu_put(vcpu);
 743
 744         return 0;
 745 }
 746
 747 long kvm_arch_vcpu_ioctl(struct file *filp,
 748                          unsigned int ioctl, unsigned long arg)
 749 {
 750         struct kvm_vcpu *vcpu = filp->private_data;
 751         void __user *argp = (void __user *)arg;
 752         int r;
 753
 754         switch (ioctl) {
 755         case KVM_GET_LAPIC: {
 756                 struct kvm_lapic_state lapic;
 757
 758                 memset(&lapic, 0, sizeof lapic);
 759                 r = kvm_vcpu_ioctl_get_lapic(vcpu, &lapic);
 760                 if (r)
 761                         goto out;
 762                 r = -EFAULT;
 763                 if (copy_to_user(argp, &lapic, sizeof lapic))
 764                         goto out;
 765                 r = 0;
 766                 break;
 767         }
 768         case KVM_SET_LAPIC: {
 769                 struct kvm_lapic_state lapic;
 770
 771                 r = -EFAULT;
 772                 if (copy_from_user(&lapic, argp, sizeof lapic))
 773                         goto out;
 774                 r = kvm_vcpu_ioctl_set_lapic(vcpu, &lapic);;
 775                 if (r)
 776                         goto out;
 777                 r = 0;
 778                 break;
 779         }
 780         case KVM_SET_CPUID: {
 781                 struct kvm_cpuid __user *cpuid_arg = argp;
 782                 struct kvm_cpuid cpuid;
 783
 784                 r = -EFAULT;
 785                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
 786                         goto out;
 787                 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
 788                 if (r)
 789                         goto out;
 790                 break;
 791         }
 792         case KVM_GET_MSRS:
 793                 r = msr_io(vcpu, argp, kvm_get_msr, 1);
 794                 break;
 795         case KVM_SET_MSRS:
 796                 r = msr_io(vcpu, argp, do_set_msr, 0);
 797                 break;
 798         default:
 799                 r = -EINVAL;
 800         }
 801 out:
 802         return r;
 803 }
 804
 805 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
 806 {
 807         int ret;
 808
 809         if (addr > (unsigned int)(-3 * PAGE_SIZE))
 810                 return -1;
 811         ret = kvm_x86_ops->set_tss_addr(kvm, addr);
 812         return ret;
 813 }
 814
 815 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
 816                                           u32 kvm_nr_mmu_pages)
 817 {
 818         if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
 819                 return -EINVAL;
 820
 821         mutex_lock(&kvm->lock);
 822
 823         kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
 824         kvm->n_requested_mmu_pages = kvm_nr_mmu_pages;
 825
 826         mutex_unlock(&kvm->lock);
 827         return 0;
 828 }
 829
 830 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
 831 {
 832         return kvm->n_alloc_mmu_pages;
 833 }
 834
 835 /*
 836  * Set a new alias region.  Aliases map a portion of physical memory into
 837  * another portion.  This is useful for memory windows, for example the PC
 838  * VGA region.
 839  */
 840 static int kvm_vm_ioctl_set_memory_alias(struct kvm *kvm,
 841                                          struct kvm_memory_alias *alias)
 842 {
 843         int r, n;
 844         struct kvm_mem_alias *p;
 845
 846         r = -EINVAL;
 847         /* General sanity checks */
 848         if (alias->memory_size & (PAGE_SIZE - 1))
 849                 goto out;
 850         if (alias->guest_phys_addr & (PAGE_SIZE - 1))
 851                 goto out;
 852         if (alias->slot >= KVM_ALIAS_SLOTS)
 853                 goto out;
 854         if (alias->guest_phys_addr + alias->memory_size
 855             < alias->guest_phys_addr)
 856                 goto out;
 857         if (alias->target_phys_addr + alias->memory_size
 858             < alias->target_phys_addr)
 859                 goto out;
 860
 861         mutex_lock(&kvm->lock);
 862
 863         p = &kvm->aliases[alias->slot];
 864         p->base_gfn = alias->guest_phys_addr >> PAGE_SHIFT;
 865         p->npages = alias->memory_size >> PAGE_SHIFT;
 866         p->target_gfn = alias->target_phys_addr >> PAGE_SHIFT;
 867
 868         for (n = KVM_ALIAS_SLOTS; n > 0; --n)
 869                 if (kvm->aliases[n - 1].npages)
 870                         break;
 871         kvm->naliases = n;
 872
 873         kvm_mmu_zap_all(kvm);
 874
 875         mutex_unlock(&kvm->lock);
 876
 877         return 0;
 878
 879 out:
 880         return r;
 881 }
 882
 883 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
 884 {
 885         int r;
 886
 887         r = 0;
 888         switch (chip->chip_id) {
 889         case KVM_IRQCHIP_PIC_MASTER:
 890                 memcpy(&chip->chip.pic,
 891                         &pic_irqchip(kvm)->pics[0],
 892                         sizeof(struct kvm_pic_state));
 893                 break;
 894         case KVM_IRQCHIP_PIC_SLAVE:
 895                 memcpy(&chip->chip.pic,
 896                         &pic_irqchip(kvm)->pics[1],
 897                         sizeof(struct kvm_pic_state));
 898                 break;
 899         case KVM_IRQCHIP_IOAPIC:
 900                 memcpy(&chip->chip.ioapic,
 901                         ioapic_irqchip(kvm),
 902                         sizeof(struct kvm_ioapic_state));
 903                 break;
 904         default:
 905                 r = -EINVAL;
 906                 break;
 907         }
 908         return r;
 909 }
 910
 911 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
 912 {
 913         int r;
 914
 915         r = 0;
 916         switch (chip->chip_id) {
 917         case KVM_IRQCHIP_PIC_MASTER:
 918                 memcpy(&pic_irqchip(kvm)->pics[0],
 919                         &chip->chip.pic,
 920                         sizeof(struct kvm_pic_state));
 921                 break;
 922         case KVM_IRQCHIP_PIC_SLAVE:
 923                 memcpy(&pic_irqchip(kvm)->pics[1],
 924                         &chip->chip.pic,
 925                         sizeof(struct kvm_pic_state));
 926                 break;
 927         case KVM_IRQCHIP_IOAPIC:
 928                 memcpy(ioapic_irqchip(kvm),
 929                         &chip->chip.ioapic,
 930                         sizeof(struct kvm_ioapic_state));
 931                 break;
 932         default:
 933                 r = -EINVAL;
 934                 break;
 935         }
 936         kvm_pic_update_irq(pic_irqchip(kvm));
 937         return r;
 938 }
 939
 940 /*
 941  * Get (and clear) the dirty memory log for a memory slot.
 942  */
 943 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
 944                                       struct kvm_dirty_log *log)
 945 {
 946         int r;
 947         int n;
 948         struct kvm_memory_slot *memslot;
 949         int is_dirty = 0;
 950
 951         mutex_lock(&kvm->lock);
 952
 953         r = kvm_get_dirty_log(kvm, log, &is_dirty);
 954         if (r)
 955                 goto out;
 956
 957         /* If nothing is dirty, don't bother messing with page tables. */
 958         if (is_dirty) {
 959                 kvm_mmu_slot_remove_write_access(kvm, log->slot);
 960                 kvm_flush_remote_tlbs(kvm);
 961                 memslot = &kvm->memslots[log->slot];
 962                 n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
 963                 memset(memslot->dirty_bitmap, 0, n);
 964         }
 965         r = 0;
 966 out:
 967         mutex_unlock(&kvm->lock);
 968         return r;
 969 }
 970
 971 long kvm_arch_vm_ioctl(struct file *filp,
 972                        unsigned int ioctl, unsigned long arg)
 973 {
 974         struct kvm *kvm = filp->private_data;
 975         void __user *argp = (void __user *)arg;
 976         int r = -EINVAL;
 977
 978         switch (ioctl) {
 979         case KVM_SET_TSS_ADDR:
 980                 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
 981                 if (r < 0)
 982                         goto out;
 983                 break;
 984         case KVM_SET_MEMORY_REGION: {
 985                 struct kvm_memory_region kvm_mem;
 986                 struct kvm_userspace_memory_region kvm_userspace_mem;
 987
 988                 r = -EFAULT;
 989                 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
 990                         goto out;
 991                 kvm_userspace_mem.slot = kvm_mem.slot;
 992                 kvm_userspace_mem.flags = kvm_mem.flags;
 993                 kvm_userspace_mem.guest_phys_addr = kvm_mem.guest_phys_addr;
 994                 kvm_userspace_mem.memory_size = kvm_mem.memory_size;
 995                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 0);
 996                 if (r)
 997                         goto out;
 998                 break;
 999         }
1000         case KVM_SET_NR_MMU_PAGES:
1001                 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
1002                 if (r)
1003                         goto out;
1004                 break;
1005         case KVM_GET_NR_MMU_PAGES:
1006                 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
1007                 break;
1008         case KVM_SET_MEMORY_ALIAS: {
1009                 struct kvm_memory_alias alias;
1010
1011                 r = -EFAULT;
1012                 if (copy_from_user(&alias, argp, sizeof alias))
1013                         goto out;
1014                 r = kvm_vm_ioctl_set_memory_alias(kvm, &alias);
1015                 if (r)
1016                         goto out;
1017                 break;
1018         }
1019         case KVM_CREATE_IRQCHIP:
1020                 r = -ENOMEM;
1021                 kvm->vpic = kvm_create_pic(kvm);
1022                 if (kvm->vpic) {
1023                         r = kvm_ioapic_init(kvm);
1024                         if (r) {
1025                                 kfree(kvm->vpic);
1026                                 kvm->vpic = NULL;
1027                                 goto out;
1028                         }
1029                 } else
1030                         goto out;
1031                 break;
1032         case KVM_IRQ_LINE: {
1033                 struct kvm_irq_level irq_event;
1034
1035                 r = -EFAULT;
1036                 if (copy_from_user(&irq_event, argp, sizeof irq_event))
1037                         goto out;
1038                 if (irqchip_in_kernel(kvm)) {
1039                         mutex_lock(&kvm->lock);
1040                         if (irq_event.irq < 16)
1041                                 kvm_pic_set_irq(pic_irqchip(kvm),
1042                                         irq_event.irq,
1043                                         irq_event.level);
1044                         kvm_ioapic_set_irq(kvm->vioapic,
1045                                         irq_event.irq,
1046                                         irq_event.level);
1047                         mutex_unlock(&kvm->lock);
1048                         r = 0;
1049                 }
1050                 break;
1051         }
1052         case KVM_GET_IRQCHIP: {
1053                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
1054                 struct kvm_irqchip chip;
1055
1056                 r = -EFAULT;
1057                 if (copy_from_user(&chip, argp, sizeof chip))
1058                         goto out;
1059                 r = -ENXIO;
1060                 if (!irqchip_in_kernel(kvm))
1061                         goto out;
1062                 r = kvm_vm_ioctl_get_irqchip(kvm, &chip);
1063                 if (r)
1064                         goto out;
1065                 r = -EFAULT;
1066                 if (copy_to_user(argp, &chip, sizeof chip))
1067                         goto out;
1068                 r = 0;
1069                 break;
1070         }
1071         case KVM_SET_IRQCHIP: {
1072                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
1073                 struct kvm_irqchip chip;
1074
1075                 r = -EFAULT;
1076                 if (copy_from_user(&chip, argp, sizeof chip))
1077                         goto out;
1078                 r = -ENXIO;
1079                 if (!irqchip_in_kernel(kvm))
1080                         goto out;
1081                 r = kvm_vm_ioctl_set_irqchip(kvm, &chip);
1082                 if (r)
1083                         goto out;
1084                 r = 0;
1085                 break;
1086         }
1087         default:
1088                 ;
1089         }
1090 out:
1091         return r;
1092 }
1093
1094 static void kvm_init_msr_list(void)
1095 {
1096         u32 dummy[2];
1097         unsigned i, j;
1098
1099         for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
1100                 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
1101                         continue;
1102                 if (j < i)
1103                         msrs_to_save[j] = msrs_to_save[i];
1104                 j++;
1105         }
1106         num_msrs_to_save = j;
1107 }
1108
1109 /*
1110  * Only apic need an MMIO device hook, so shortcut now..
1111  */
1112 static struct kvm_io_device *vcpu_find_pervcpu_dev(struct kvm_vcpu *vcpu,
1113                                                 gpa_t addr)
1114 {
1115         struct kvm_io_device *dev;
1116
1117         if (vcpu->apic) {
1118                 dev = &vcpu->apic->dev;
1119                 if (dev->in_range(dev, addr))
1120                         return dev;
1121         }
1122         return NULL;
1123 }
1124
1125
1126 static struct kvm_io_device *vcpu_find_mmio_dev(struct kvm_vcpu *vcpu,
1127                                                 gpa_t addr)
1128 {
1129         struct kvm_io_device *dev;
1130
1131         dev = vcpu_find_pervcpu_dev(vcpu, addr);
1132         if (dev == NULL)
1133                 dev = kvm_io_bus_find_dev(&vcpu->kvm->mmio_bus, addr);
1134         return dev;
1135 }
1136
1137 int emulator_read_std(unsigned long addr,
1138                              void *val,
1139                              unsigned int bytes,
1140                              struct kvm_vcpu *vcpu)
1141 {
1142         void *data = val;
1143
1144         while (bytes) {
1145                 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1146                 unsigned offset = addr & (PAGE_SIZE-1);
1147                 unsigned tocopy = min(bytes, (unsigned)PAGE_SIZE - offset);
1148                 int ret;
1149
1150                 if (gpa == UNMAPPED_GVA)
1151                         return X86EMUL_PROPAGATE_FAULT;
1152                 ret = kvm_read_guest(vcpu->kvm, gpa, data, tocopy);
1153                 if (ret < 0)
1154                         return X86EMUL_UNHANDLEABLE;
1155
1156                 bytes -= tocopy;
1157                 data += tocopy;
1158                 addr += tocopy;
1159         }
1160
1161         return X86EMUL_CONTINUE;
1162 }
1163 EXPORT_SYMBOL_GPL(emulator_read_std);
1164
1165 static int emulator_write_std(unsigned long addr,
1166                               const void *val,
1167                               unsigned int bytes,
1168                               struct kvm_vcpu *vcpu)
1169 {
1170         pr_unimpl(vcpu, "emulator_write_std: addr %lx n %d\n", addr, bytes);
1171         return X86EMUL_UNHANDLEABLE;
1172 }
1173
1174 static int emulator_read_emulated(unsigned long addr,
1175                                   void *val,
1176                                   unsigned int bytes,
1177                                   struct kvm_vcpu *vcpu)
1178 {
1179         struct kvm_io_device *mmio_dev;
1180         gpa_t                 gpa;
1181
1182         if (vcpu->mmio_read_completed) {
1183                 memcpy(val, vcpu->mmio_data, bytes);
1184                 vcpu->mmio_read_completed = 0;
1185                 return X86EMUL_CONTINUE;
1186         }
1187
1188         gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1189
1190         /* For APIC access vmexit */
1191         if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
1192                 goto mmio;
1193
1194         if (emulator_read_std(addr, val, bytes, vcpu)
1195                         == X86EMUL_CONTINUE)
1196                 return X86EMUL_CONTINUE;
1197         if (gpa == UNMAPPED_GVA)
1198                 return X86EMUL_PROPAGATE_FAULT;
1199
1200 mmio:
1201         /*
1202          * Is this MMIO handled locally?
1203          */
1204         mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1205         if (mmio_dev) {
1206                 kvm_iodevice_read(mmio_dev, gpa, bytes, val);
1207                 return X86EMUL_CONTINUE;
1208         }
1209
1210         vcpu->mmio_needed = 1;
1211         vcpu->mmio_phys_addr = gpa;
1212         vcpu->mmio_size = bytes;
1213         vcpu->mmio_is_write = 0;
1214
1215         return X86EMUL_UNHANDLEABLE;
1216 }
1217
1218 static int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
1219                                const void *val, int bytes)
1220 {
1221         int ret;
1222
1223         ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
1224         if (ret < 0)
1225                 return 0;
1226         kvm_mmu_pte_write(vcpu, gpa, val, bytes);
1227         return 1;
1228 }
1229
1230 static int emulator_write_emulated_onepage(unsigned long addr,
1231                                            const void *val,
1232                                            unsigned int bytes,
1233                                            struct kvm_vcpu *vcpu)
1234 {
1235         struct kvm_io_device *mmio_dev;
1236         gpa_t                 gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1237
1238         if (gpa == UNMAPPED_GVA) {
1239                 kvm_x86_ops->inject_page_fault(vcpu, addr, 2);
1240                 return X86EMUL_PROPAGATE_FAULT;
1241         }
1242
1243         /* For APIC access vmexit */
1244         if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
1245                 goto mmio;
1246
1247         if (emulator_write_phys(vcpu, gpa, val, bytes))
1248                 return X86EMUL_CONTINUE;
1249
1250 mmio:
1251         /*
1252          * Is this MMIO handled locally?
1253          */
1254         mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1255         if (mmio_dev) {
1256                 kvm_iodevice_write(mmio_dev, gpa, bytes, val);
1257                 return X86EMUL_CONTINUE;
1258         }
1259
1260         vcpu->mmio_needed = 1;
1261         vcpu->mmio_phys_addr = gpa;
1262         vcpu->mmio_size = bytes;
1263         vcpu->mmio_is_write = 1;
1264         memcpy(vcpu->mmio_data, val, bytes);
1265
1266         return X86EMUL_CONTINUE;
1267 }
1268
1269 int emulator_write_emulated(unsigned long addr,
1270                                    const void *val,
1271                                    unsigned int bytes,
1272                                    struct kvm_vcpu *vcpu)
1273 {
1274         /* Crossing a page boundary? */
1275         if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
1276                 int rc, now;
1277
1278                 now = -addr & ~PAGE_MASK;
1279                 rc = emulator_write_emulated_onepage(addr, val, now, vcpu);
1280                 if (rc != X86EMUL_CONTINUE)
1281                         return rc;
1282                 addr += now;
1283                 val += now;
1284                 bytes -= now;
1285         }
1286         return emulator_write_emulated_onepage(addr, val, bytes, vcpu);
1287 }
1288 EXPORT_SYMBOL_GPL(emulator_write_emulated);
1289
1290 static int emulator_cmpxchg_emulated(unsigned long addr,
1291                                      const void *old,
1292                                      const void *new,
1293                                      unsigned int bytes,
1294                                      struct kvm_vcpu *vcpu)
1295 {
1296         static int reported;
1297
1298         if (!reported) {
1299                 reported = 1;
1300                 printk(KERN_WARNING "kvm: emulating exchange as write\n");
1301         }
1302         return emulator_write_emulated(addr, new, bytes, vcpu);
1303 }
1304
1305 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
1306 {
1307         return kvm_x86_ops->get_segment_base(vcpu, seg);
1308 }
1309
1310 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
1311 {
1312         return X86EMUL_CONTINUE;
1313 }
1314
1315 int emulate_clts(struct kvm_vcpu *vcpu)
1316 {
1317         kvm_x86_ops->set_cr0(vcpu, vcpu->cr0 & ~X86_CR0_TS);
1318         return X86EMUL_CONTINUE;
1319 }
1320
1321 int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long *dest)
1322 {
1323         struct kvm_vcpu *vcpu = ctxt->vcpu;
1324
1325         switch (dr) {
1326         case 0 ... 3:
1327                 *dest = kvm_x86_ops->get_dr(vcpu, dr);
1328                 return X86EMUL_CONTINUE;
1329         default:
1330                 pr_unimpl(vcpu, "%s: unexpected dr %u\n", __FUNCTION__, dr);
1331                 return X86EMUL_UNHANDLEABLE;
1332         }
1333 }
1334
1335 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
1336 {
1337         unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
1338         int exception;
1339
1340         kvm_x86_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
1341         if (exception) {
1342                 /* FIXME: better handling */
1343                 return X86EMUL_UNHANDLEABLE;
1344         }
1345         return X86EMUL_CONTINUE;
1346 }
1347
1348 void kvm_report_emulation_failure(struct kvm_vcpu *vcpu, const char *context)
1349 {
1350         static int reported;
1351         u8 opcodes[4];
1352         unsigned long rip = vcpu->rip;
1353         unsigned long rip_linear;
1354
1355         rip_linear = rip + get_segment_base(vcpu, VCPU_SREG_CS);
1356
1357         if (reported)
1358                 return;
1359
1360         emulator_read_std(rip_linear, (void *)opcodes, 4, vcpu);
1361
1362         printk(KERN_ERR "emulation failed (%s) rip %lx %02x %02x %02x %02x\n",
1363                context, rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
1364         reported = 1;
1365 }
1366 EXPORT_SYMBOL_GPL(kvm_report_emulation_failure);
1367
1368 struct x86_emulate_ops emulate_ops = {
1369         .read_std            = emulator_read_std,
1370         .write_std           = emulator_write_std,
1371         .read_emulated       = emulator_read_emulated,
1372         .write_emulated      = emulator_write_emulated,
1373         .cmpxchg_emulated    = emulator_cmpxchg_emulated,
1374 };
1375
1376 int emulate_instruction(struct kvm_vcpu *vcpu,
1377                         struct kvm_run *run,
1378                         unsigned long cr2,
1379                         u16 error_code,
1380                         int no_decode)
1381 {
1382         int r;
1383
1384         vcpu->mmio_fault_cr2 = cr2;
1385         kvm_x86_ops->cache_regs(vcpu);
1386
1387         vcpu->mmio_is_write = 0;
1388         vcpu->pio.string = 0;
1389
1390         if (!no_decode) {
1391                 int cs_db, cs_l;
1392                 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
1393
1394                 vcpu->emulate_ctxt.vcpu = vcpu;
1395                 vcpu->emulate_ctxt.eflags = kvm_x86_ops->get_rflags(vcpu);
1396                 vcpu->emulate_ctxt.cr2 = cr2;
1397                 vcpu->emulate_ctxt.mode =
1398                         (vcpu->emulate_ctxt.eflags & X86_EFLAGS_VM)
1399                         ? X86EMUL_MODE_REAL : cs_l
1400                         ? X86EMUL_MODE_PROT64 : cs_db
1401                         ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
1402
1403                 if (vcpu->emulate_ctxt.mode == X86EMUL_MODE_PROT64) {
1404                         vcpu->emulate_ctxt.cs_base = 0;
1405                         vcpu->emulate_ctxt.ds_base = 0;
1406                         vcpu->emulate_ctxt.es_base = 0;
1407                         vcpu->emulate_ctxt.ss_base = 0;
1408                 } else {
1409                         vcpu->emulate_ctxt.cs_base =
1410                                         get_segment_base(vcpu, VCPU_SREG_CS);
1411                         vcpu->emulate_ctxt.ds_base =
1412                                         get_segment_base(vcpu, VCPU_SREG_DS);
1413                         vcpu->emulate_ctxt.es_base =
1414                                         get_segment_base(vcpu, VCPU_SREG_ES);
1415                         vcpu->emulate_ctxt.ss_base =
1416                                         get_segment_base(vcpu, VCPU_SREG_SS);
1417                 }
1418
1419                 vcpu->emulate_ctxt.gs_base =
1420                                         get_segment_base(vcpu, VCPU_SREG_GS);
1421                 vcpu->emulate_ctxt.fs_base =
1422                                         get_segment_base(vcpu, VCPU_SREG_FS);
1423
1424                 r = x86_decode_insn(&vcpu->emulate_ctxt, &emulate_ops);
1425                 ++vcpu->stat.insn_emulation;
1426                 if (r)  {
1427                         ++vcpu->stat.insn_emulation_fail;
1428                         if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
1429                                 return EMULATE_DONE;
1430                         return EMULATE_FAIL;
1431                 }
1432         }
1433
1434         r = x86_emulate_insn(&vcpu->emulate_ctxt, &emulate_ops);
1435
1436         if (vcpu->pio.string)
1437                 return EMULATE_DO_MMIO;
1438
1439         if ((r || vcpu->mmio_is_write) && run) {
1440                 run->exit_reason = KVM_EXIT_MMIO;
1441                 run->mmio.phys_addr = vcpu->mmio_phys_addr;
1442                 memcpy(run->mmio.data, vcpu->mmio_data, 8);
1443                 run->mmio.len = vcpu->mmio_size;
1444                 run->mmio.is_write = vcpu->mmio_is_write;
1445         }
1446
1447         if (r) {
1448                 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
1449                         return EMULATE_DONE;
1450                 if (!vcpu->mmio_needed) {
1451                         kvm_report_emulation_failure(vcpu, "mmio");
1452                         return EMULATE_FAIL;
1453                 }
1454                 return EMULATE_DO_MMIO;
1455         }
1456
1457         kvm_x86_ops->decache_regs(vcpu);
1458         kvm_x86_ops->set_rflags(vcpu, vcpu->emulate_ctxt.eflags);
1459
1460         if (vcpu->mmio_is_write) {
1461                 vcpu->mmio_needed = 0;
1462                 return EMULATE_DO_MMIO;
1463         }
1464
1465         return EMULATE_DONE;
1466 }
1467 EXPORT_SYMBOL_GPL(emulate_instruction);
1468
1469 static void free_pio_guest_pages(struct kvm_vcpu *vcpu)
1470 {
1471         int i;
1472
1473         for (i = 0; i < ARRAY_SIZE(vcpu->pio.guest_pages); ++i)
1474                 if (vcpu->pio.guest_pages[i]) {
1475                         kvm_release_page(vcpu->pio.guest_pages[i]);
1476                         vcpu->pio.guest_pages[i] = NULL;
1477                 }
1478 }
1479
1480 static int pio_copy_data(struct kvm_vcpu *vcpu)
1481 {
1482         void *p = vcpu->pio_data;
1483         void *q;
1484         unsigned bytes;
1485         int nr_pages = vcpu->pio.guest_pages[1] ? 2 : 1;
1486
1487         q = vmap(vcpu->pio.guest_pages, nr_pages, VM_READ|VM_WRITE,
1488                  PAGE_KERNEL);
1489         if (!q) {
1490                 free_pio_guest_pages(vcpu);
1491                 return -ENOMEM;
1492         }
1493         q += vcpu->pio.guest_page_offset;
1494         bytes = vcpu->pio.size * vcpu->pio.cur_count;
1495         if (vcpu->pio.in)
1496                 memcpy(q, p, bytes);
1497         else
1498                 memcpy(p, q, bytes);
1499         q -= vcpu->pio.guest_page_offset;
1500         vunmap(q);
1501         free_pio_guest_pages(vcpu);
1502         return 0;
1503 }
1504
1505 int complete_pio(struct kvm_vcpu *vcpu)
1506 {
1507         struct kvm_pio_request *io = &vcpu->pio;
1508         long delta;
1509         int r;
1510
1511         kvm_x86_ops->cache_regs(vcpu);
1512
1513         if (!io->string) {
1514                 if (io->in)
1515                         memcpy(&vcpu->regs[VCPU_REGS_RAX], vcpu->pio_data,
1516                                io->size);
1517         } else {
1518                 if (io->in) {
1519                         r = pio_copy_data(vcpu);
1520                         if (r) {
1521                                 kvm_x86_ops->cache_regs(vcpu);
1522                                 return r;
1523                         }
1524                 }
1525
1526                 delta = 1;
1527                 if (io->rep) {
1528                         delta *= io->cur_count;
1529                         /*
1530                          * The size of the register should really depend on
1531                          * current address size.
1532                          */
1533                         vcpu->regs[VCPU_REGS_RCX] -= delta;
1534                 }
1535                 if (io->down)
1536                         delta = -delta;
1537                 delta *= io->size;
1538                 if (io->in)
1539                         vcpu->regs[VCPU_REGS_RDI] += delta;
1540                 else
1541                         vcpu->regs[VCPU_REGS_RSI] += delta;
1542         }
1543
1544         kvm_x86_ops->decache_regs(vcpu);
1545
1546         io->count -= io->cur_count;
1547         io->cur_count = 0;
1548
1549         return 0;
1550 }
1551
1552 static void kernel_pio(struct kvm_io_device *pio_dev,
1553                        struct kvm_vcpu *vcpu,
1554                        void *pd)
1555 {
1556         /* TODO: String I/O for in kernel device */
1557
1558         mutex_lock(&vcpu->kvm->lock);
1559         if (vcpu->pio.in)
1560                 kvm_iodevice_read(pio_dev, vcpu->pio.port,
1561                                   vcpu->pio.size,
1562                                   pd);
1563         else
1564                 kvm_iodevice_write(pio_dev, vcpu->pio.port,
1565                                    vcpu->pio.size,
1566                                    pd);
1567         mutex_unlock(&vcpu->kvm->lock);
1568 }
1569
1570 static void pio_string_write(struct kvm_io_device *pio_dev,
1571                              struct kvm_vcpu *vcpu)
1572 {
1573         struct kvm_pio_request *io = &vcpu->pio;
1574         void *pd = vcpu->pio_data;
1575         int i;
1576
1577         mutex_lock(&vcpu->kvm->lock);
1578         for (i = 0; i < io->cur_count; i++) {
1579                 kvm_iodevice_write(pio_dev, io->port,
1580                                    io->size,
1581                                    pd);
1582                 pd += io->size;
1583         }
1584         mutex_unlock(&vcpu->kvm->lock);
1585 }
1586
1587 static struct kvm_io_device *vcpu_find_pio_dev(struct kvm_vcpu *vcpu,
1588                                                gpa_t addr)
1589 {
1590         return kvm_io_bus_find_dev(&vcpu->kvm->pio_bus, addr);
1591 }
1592
1593 int kvm_emulate_pio(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
1594                   int size, unsigned port)
1595 {
1596         struct kvm_io_device *pio_dev;
1597
1598         vcpu->run->exit_reason = KVM_EXIT_IO;
1599         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
1600         vcpu->run->io.size = vcpu->pio.size = size;
1601         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
1602         vcpu->run->io.count = vcpu->pio.count = vcpu->pio.cur_count = 1;
1603         vcpu->run->io.port = vcpu->pio.port = port;
1604         vcpu->pio.in = in;
1605         vcpu->pio.string = 0;
1606         vcpu->pio.down = 0;
1607         vcpu->pio.guest_page_offset = 0;
1608         vcpu->pio.rep = 0;
1609
1610         kvm_x86_ops->cache_regs(vcpu);
1611         memcpy(vcpu->pio_data, &vcpu->regs[VCPU_REGS_RAX], 4);
1612         kvm_x86_ops->decache_regs(vcpu);
1613
1614         kvm_x86_ops->skip_emulated_instruction(vcpu);
1615
1616         pio_dev = vcpu_find_pio_dev(vcpu, port);
1617         if (pio_dev) {
1618                 kernel_pio(pio_dev, vcpu, vcpu->pio_data);
1619                 complete_pio(vcpu);
1620                 return 1;
1621         }
1622         return 0;
1623 }
1624 EXPORT_SYMBOL_GPL(kvm_emulate_pio);
1625
1626 int kvm_emulate_pio_string(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
1627                   int size, unsigned long count, int down,
1628                   gva_t address, int rep, unsigned port)
1629 {
1630         unsigned now, in_page;
1631         int i, ret = 0;
1632         int nr_pages = 1;
1633         struct page *page;
1634         struct kvm_io_device *pio_dev;
1635
1636         vcpu->run->exit_reason = KVM_EXIT_IO;
1637         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
1638         vcpu->run->io.size = vcpu->pio.size = size;
1639         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
1640         vcpu->run->io.count = vcpu->pio.count = vcpu->pio.cur_count = count;
1641         vcpu->run->io.port = vcpu->pio.port = port;
1642         vcpu->pio.in = in;
1643         vcpu->pio.string = 1;
1644         vcpu->pio.down = down;
1645         vcpu->pio.guest_page_offset = offset_in_page(address);
1646         vcpu->pio.rep = rep;
1647
1648         if (!count) {
1649                 kvm_x86_ops->skip_emulated_instruction(vcpu);
1650                 return 1;
1651         }
1652
1653         if (!down)
1654                 in_page = PAGE_SIZE - offset_in_page(address);
1655         else
1656                 in_page = offset_in_page(address) + size;
1657         now = min(count, (unsigned long)in_page / size);
1658         if (!now) {
1659                 /*
1660                  * String I/O straddles page boundary.  Pin two guest pages
1661                  * so that we satisfy atomicity constraints.  Do just one
1662                  * transaction to avoid complexity.
1663                  */
1664                 nr_pages = 2;
1665                 now = 1;
1666         }
1667         if (down) {
1668                 /*
1669                  * String I/O in reverse.  Yuck.  Kill the guest, fix later.
1670                  */
1671                 pr_unimpl(vcpu, "guest string pio down\n");
1672                 inject_gp(vcpu);
1673                 return 1;
1674         }
1675         vcpu->run->io.count = now;
1676         vcpu->pio.cur_count = now;
1677
1678         if (vcpu->pio.cur_count == vcpu->pio.count)
1679                 kvm_x86_ops->skip_emulated_instruction(vcpu);
1680
1681         for (i = 0; i < nr_pages; ++i) {
1682                 mutex_lock(&vcpu->kvm->lock);
1683                 page = gva_to_page(vcpu, address + i * PAGE_SIZE);
1684                 vcpu->pio.guest_pages[i] = page;
1685                 mutex_unlock(&vcpu->kvm->lock);
1686                 if (!page) {
1687                         inject_gp(vcpu);
1688                         free_pio_guest_pages(vcpu);
1689                         return 1;
1690                 }
1691         }
1692
1693         pio_dev = vcpu_find_pio_dev(vcpu, port);
1694         if (!vcpu->pio.in) {
1695                 /* string PIO write */
1696                 ret = pio_copy_data(vcpu);
1697                 if (ret >= 0 && pio_dev) {
1698                         pio_string_write(pio_dev, vcpu);
1699                         complete_pio(vcpu);
1700                         if (vcpu->pio.count == 0)
1701                                 ret = 1;
1702                 }
1703         } else if (pio_dev)
1704                 pr_unimpl(vcpu, "no string pio read support yet, "
1705                        "port %x size %d count %ld\n",
1706                         port, size, count);
1707
1708         return ret;
1709 }
1710 EXPORT_SYMBOL_GPL(kvm_emulate_pio_string);
1711
1712 int kvm_arch_init(void *opaque)
1713 {
1714         int r;
1715         struct kvm_x86_ops *ops = (struct kvm_x86_ops *)opaque;
1716
1717         r = kvm_mmu_module_init();
1718         if (r)
1719                 goto out_fail;
1720
1721         kvm_init_msr_list();
1722
1723         if (kvm_x86_ops) {
1724                 printk(KERN_ERR "kvm: already loaded the other module\n");
1725                 r = -EEXIST;
1726                 goto out;
1727         }
1728
1729         if (!ops->cpu_has_kvm_support()) {
1730                 printk(KERN_ERR "kvm: no hardware support\n");
1731                 r = -EOPNOTSUPP;
1732                 goto out;
1733         }
1734         if (ops->disabled_by_bios()) {
1735                 printk(KERN_ERR "kvm: disabled by bios\n");
1736                 r = -EOPNOTSUPP;
1737                 goto out;
1738         }
1739
1740         kvm_x86_ops = ops;
1741         kvm_mmu_set_nonpresent_ptes(0ull, 0ull);
1742         return 0;
1743
1744 out:
1745         kvm_mmu_module_exit();
1746 out_fail:
1747         return r;
1748 }
1749
1750 void kvm_arch_exit(void)
1751 {
1752         kvm_x86_ops = NULL;
1753         kvm_mmu_module_exit();
1754 }
1755
1756 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
1757 {
1758         ++vcpu->stat.halt_exits;
1759         if (irqchip_in_kernel(vcpu->kvm)) {
1760                 vcpu->mp_state = VCPU_MP_STATE_HALTED;
1761                 kvm_vcpu_block(vcpu);
1762                 if (vcpu->mp_state != VCPU_MP_STATE_RUNNABLE)
1763                         return -EINTR;
1764                 return 1;
1765         } else {
1766                 vcpu->run->exit_reason = KVM_EXIT_HLT;
1767                 return 0;
1768         }
1769 }
1770 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
1771
1772 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
1773 {
1774         unsigned long nr, a0, a1, a2, a3, ret;
1775
1776         kvm_x86_ops->cache_regs(vcpu);
1777
1778         nr = vcpu->regs[VCPU_REGS_RAX];
1779         a0 = vcpu->regs[VCPU_REGS_RBX];
1780         a1 = vcpu->regs[VCPU_REGS_RCX];
1781         a2 = vcpu->regs[VCPU_REGS_RDX];
1782         a3 = vcpu->regs[VCPU_REGS_RSI];
1783
1784         if (!is_long_mode(vcpu)) {
1785                 nr &= 0xFFFFFFFF;
1786                 a0 &= 0xFFFFFFFF;
1787                 a1 &= 0xFFFFFFFF;
1788                 a2 &= 0xFFFFFFFF;
1789                 a3 &= 0xFFFFFFFF;
1790         }
1791
1792         switch (nr) {
1793         default:
1794                 ret = -KVM_ENOSYS;
1795                 break;
1796         }
1797         vcpu->regs[VCPU_REGS_RAX] = ret;
1798         kvm_x86_ops->decache_regs(vcpu);
1799         return 0;
1800 }
1801 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
1802
1803 int kvm_fix_hypercall(struct kvm_vcpu *vcpu)
1804 {
1805         char instruction[3];
1806         int ret = 0;
1807
1808         mutex_lock(&vcpu->kvm->lock);
1809
1810         /*
1811          * Blow out the MMU to ensure that no other VCPU has an active mapping
1812          * to ensure that the updated hypercall appears atomically across all
1813          * VCPUs.
1814          */
1815         kvm_mmu_zap_all(vcpu->kvm);
1816
1817         kvm_x86_ops->cache_regs(vcpu);
1818         kvm_x86_ops->patch_hypercall(vcpu, instruction);
1819         if (emulator_write_emulated(vcpu->rip, instruction, 3, vcpu)
1820             != X86EMUL_CONTINUE)
1821                 ret = -EFAULT;
1822
1823         mutex_unlock(&vcpu->kvm->lock);
1824
1825         return ret;
1826 }
1827
1828 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
1829 {
1830         return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
1831 }
1832
1833 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1834 {
1835         struct descriptor_table dt = { limit, base };
1836
1837         kvm_x86_ops->set_gdt(vcpu, &dt);
1838 }
1839
1840 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1841 {
1842         struct descriptor_table dt = { limit, base };
1843
1844         kvm_x86_ops->set_idt(vcpu, &dt);
1845 }
1846
1847 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
1848                    unsigned long *rflags)
1849 {
1850         lmsw(vcpu, msw);
1851         *rflags = kvm_x86_ops->get_rflags(vcpu);
1852 }
1853
1854 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
1855 {
1856         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
1857         switch (cr) {
1858         case 0:
1859                 return vcpu->cr0;
1860         case 2:
1861                 return vcpu->cr2;
1862         case 3:
1863                 return vcpu->cr3;
1864         case 4:
1865                 return vcpu->cr4;
1866         default:
1867                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1868                 return 0;
1869         }
1870 }
1871
1872 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
1873                      unsigned long *rflags)
1874 {
1875         switch (cr) {
1876         case 0:
1877                 set_cr0(vcpu, mk_cr_64(vcpu->cr0, val));
1878                 *rflags = kvm_x86_ops->get_rflags(vcpu);
1879                 break;
1880         case 2:
1881                 vcpu->cr2 = val;
1882                 break;
1883         case 3:
1884                 set_cr3(vcpu, val);
1885                 break;
1886         case 4:
1887                 set_cr4(vcpu, mk_cr_64(vcpu->cr4, val));
1888                 break;
1889         default:
1890                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1891         }
1892 }
1893
1894 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
1895 {
1896         int i;
1897         u32 function;
1898         struct kvm_cpuid_entry *e, *best;
1899
1900         kvm_x86_ops->cache_regs(vcpu);
1901         function = vcpu->regs[VCPU_REGS_RAX];
1902         vcpu->regs[VCPU_REGS_RAX] = 0;
1903         vcpu->regs[VCPU_REGS_RBX] = 0;
1904         vcpu->regs[VCPU_REGS_RCX] = 0;
1905         vcpu->regs[VCPU_REGS_RDX] = 0;
1906         best = NULL;
1907         for (i = 0; i < vcpu->cpuid_nent; ++i) {
1908                 e = &vcpu->cpuid_entries[i];
1909                 if (e->function == function) {
1910                         best = e;
1911                         break;
1912                 }
1913                 /*
1914                  * Both basic or both extended?
1915                  */
1916                 if (((e->function ^ function) & 0x80000000) == 0)
1917                         if (!best || e->function > best->function)
1918                                 best = e;
1919         }
1920         if (best) {
1921                 vcpu->regs[VCPU_REGS_RAX] = best->eax;
1922                 vcpu->regs[VCPU_REGS_RBX] = best->ebx;
1923                 vcpu->regs[VCPU_REGS_RCX] = best->ecx;
1924                 vcpu->regs[VCPU_REGS_RDX] = best->edx;
1925         }
1926         kvm_x86_ops->decache_regs(vcpu);
1927         kvm_x86_ops->skip_emulated_instruction(vcpu);
1928 }
1929 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
1930
1931 /*
1932  * Check if userspace requested an interrupt window, and that the
1933  * interrupt window is open.
1934  *
1935  * No need to exit to userspace if we already have an interrupt queued.
1936  */
1937 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu,
1938                                           struct kvm_run *kvm_run)
1939 {
1940         return (!vcpu->irq_summary &&
1941                 kvm_run->request_interrupt_window &&
1942                 vcpu->interrupt_window_open &&
1943                 (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF));
1944 }
1945
1946 static void post_kvm_run_save(struct kvm_vcpu *vcpu,
1947                               struct kvm_run *kvm_run)
1948 {
1949         kvm_run->if_flag = (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
1950         kvm_run->cr8 = get_cr8(vcpu);
1951         kvm_run->apic_base = kvm_get_apic_base(vcpu);
1952         if (irqchip_in_kernel(vcpu->kvm))
1953                 kvm_run->ready_for_interrupt_injection = 1;
1954         else
1955                 kvm_run->ready_for_interrupt_injection =
1956                                         (vcpu->interrupt_window_open &&
1957                                          vcpu->irq_summary == 0);
1958 }
1959
1960 static int __vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1961 {
1962         int r;
1963
1964         if (unlikely(vcpu->mp_state == VCPU_MP_STATE_SIPI_RECEIVED)) {
1965                 pr_debug("vcpu %d received sipi with vector # %x\n",
1966                        vcpu->vcpu_id, vcpu->sipi_vector);
1967                 kvm_lapic_reset(vcpu);
1968                 r = kvm_x86_ops->vcpu_reset(vcpu);
1969                 if (r)
1970                         return r;
1971                 vcpu->mp_state = VCPU_MP_STATE_RUNNABLE;
1972         }
1973
1974 preempted:
1975         if (vcpu->guest_debug.enabled)
1976                 kvm_x86_ops->guest_debug_pre(vcpu);
1977
1978 again:
1979         r = kvm_mmu_reload(vcpu);
1980         if (unlikely(r))
1981                 goto out;
1982
1983         kvm_inject_pending_timer_irqs(vcpu);
1984
1985         preempt_disable();
1986
1987         kvm_x86_ops->prepare_guest_switch(vcpu);
1988         kvm_load_guest_fpu(vcpu);
1989
1990         local_irq_disable();
1991
1992         if (signal_pending(current)) {
1993                 local_irq_enable();
1994                 preempt_enable();
1995                 r = -EINTR;
1996                 kvm_run->exit_reason = KVM_EXIT_INTR;
1997                 ++vcpu->stat.signal_exits;
1998                 goto out;
1999         }
2000
2001         if (irqchip_in_kernel(vcpu->kvm))
2002                 kvm_x86_ops->inject_pending_irq(vcpu);
2003         else if (!vcpu->mmio_read_completed)
2004                 kvm_x86_ops->inject_pending_vectors(vcpu, kvm_run);
2005
2006         vcpu->guest_mode = 1;
2007         kvm_guest_enter();
2008
2009         if (vcpu->requests)
2010                 if (test_and_clear_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
2011                         kvm_x86_ops->tlb_flush(vcpu);
2012
2013         kvm_x86_ops->run(vcpu, kvm_run);
2014
2015         vcpu->guest_mode = 0;
2016         local_irq_enable();
2017
2018         ++vcpu->stat.exits;
2019
2020         /*
2021          * We must have an instruction between local_irq_enable() and
2022          * kvm_guest_exit(), so the timer interrupt isn't delayed by
2023          * the interrupt shadow.  The stat.exits increment will do nicely.
2024          * But we need to prevent reordering, hence this barrier():
2025          */
2026         barrier();
2027
2028         kvm_guest_exit();
2029
2030         preempt_enable();
2031
2032         /*
2033          * Profile KVM exit RIPs:
2034          */
2035         if (unlikely(prof_on == KVM_PROFILING)) {
2036                 kvm_x86_ops->cache_regs(vcpu);
2037                 profile_hit(KVM_PROFILING, (void *)vcpu->rip);
2038         }
2039
2040         r = kvm_x86_ops->handle_exit(kvm_run, vcpu);
2041
2042         if (r > 0) {
2043                 if (dm_request_for_irq_injection(vcpu, kvm_run)) {
2044                         r = -EINTR;
2045                         kvm_run->exit_reason = KVM_EXIT_INTR;
2046                         ++vcpu->stat.request_irq_exits;
2047                         goto out;
2048                 }
2049                 if (!need_resched())
2050                         goto again;
2051         }
2052
2053 out:
2054         if (r > 0) {
2055                 kvm_resched(vcpu);
2056                 goto preempted;
2057         }
2058
2059         post_kvm_run_save(vcpu, kvm_run);
2060
2061         return r;
2062 }
2063
2064 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2065 {
2066         int r;
2067         sigset_t sigsaved;
2068
2069         vcpu_load(vcpu);
2070
2071         if (unlikely(vcpu->mp_state == VCPU_MP_STATE_UNINITIALIZED)) {
2072                 kvm_vcpu_block(vcpu);
2073                 vcpu_put(vcpu);
2074                 return -EAGAIN;
2075         }
2076
2077         if (vcpu->sigset_active)
2078                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
2079
2080         /* re-sync apic's tpr */
2081         if (!irqchip_in_kernel(vcpu->kvm))
2082                 set_cr8(vcpu, kvm_run->cr8);
2083
2084         if (vcpu->pio.cur_count) {
2085                 r = complete_pio(vcpu);
2086                 if (r)
2087                         goto out;
2088         }
2089 #if CONFIG_HAS_IOMEM
2090         if (vcpu->mmio_needed) {
2091                 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
2092                 vcpu->mmio_read_completed = 1;
2093                 vcpu->mmio_needed = 0;
2094                 r = emulate_instruction(vcpu, kvm_run,
2095                                         vcpu->mmio_fault_cr2, 0, 1);
2096                 if (r == EMULATE_DO_MMIO) {
2097                         /*
2098                          * Read-modify-write.  Back to userspace.
2099                          */
2100                         r = 0;
2101                         goto out;
2102                 }
2103         }
2104 #endif
2105         if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL) {
2106                 kvm_x86_ops->cache_regs(vcpu);
2107                 vcpu->regs[VCPU_REGS_RAX] = kvm_run->hypercall.ret;
2108                 kvm_x86_ops->decache_regs(vcpu);
2109         }
2110
2111         r = __vcpu_run(vcpu, kvm_run);
2112
2113 out:
2114         if (vcpu->sigset_active)
2115                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
2116
2117         vcpu_put(vcpu);
2118         return r;
2119 }
2120
2121 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
2122 {
2123         vcpu_load(vcpu);
2124
2125         kvm_x86_ops->cache_regs(vcpu);
2126
2127         regs->rax = vcpu->regs[VCPU_REGS_RAX];
2128         regs->rbx = vcpu->regs[VCPU_REGS_RBX];
2129         regs->rcx = vcpu->regs[VCPU_REGS_RCX];
2130         regs->rdx = vcpu->regs[VCPU_REGS_RDX];
2131         regs->rsi = vcpu->regs[VCPU_REGS_RSI];
2132         regs->rdi = vcpu->regs[VCPU_REGS_RDI];
2133         regs->rsp = vcpu->regs[VCPU_REGS_RSP];
2134         regs->rbp = vcpu->regs[VCPU_REGS_RBP];
2135 #ifdef CONFIG_X86_64
2136         regs->r8 = vcpu->regs[VCPU_REGS_R8];
2137         regs->r9 = vcpu->regs[VCPU_REGS_R9];
2138         regs->r10 = vcpu->regs[VCPU_REGS_R10];
2139         regs->r11 = vcpu->regs[VCPU_REGS_R11];
2140         regs->r12 = vcpu->regs[VCPU_REGS_R12];
2141         regs->r13 = vcpu->regs[VCPU_REGS_R13];
2142         regs->r14 = vcpu->regs[VCPU_REGS_R14];
2143         regs->r15 = vcpu->regs[VCPU_REGS_R15];
2144 #endif
2145
2146         regs->rip = vcpu->rip;
2147         regs->rflags = kvm_x86_ops->get_rflags(vcpu);
2148
2149         /*
2150          * Don't leak debug flags in case they were set for guest debugging
2151          */
2152         if (vcpu->guest_debug.enabled && vcpu->guest_debug.singlestep)
2153                 regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
2154
2155         vcpu_put(vcpu);
2156
2157         return 0;
2158 }
2159
2160 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
2161 {
2162         vcpu_load(vcpu);
2163
2164         vcpu->regs[VCPU_REGS_RAX] = regs->rax;
2165         vcpu->regs[VCPU_REGS_RBX] = regs->rbx;
2166         vcpu->regs[VCPU_REGS_RCX] = regs->rcx;
2167         vcpu->regs[VCPU_REGS_RDX] = regs->rdx;
2168         vcpu->regs[VCPU_REGS_RSI] = regs->rsi;
2169         vcpu->regs[VCPU_REGS_RDI] = regs->rdi;
2170         vcpu->regs[VCPU_REGS_RSP] = regs->rsp;
2171         vcpu->regs[VCPU_REGS_RBP] = regs->rbp;
2172 #ifdef CONFIG_X86_64
2173         vcpu->regs[VCPU_REGS_R8] = regs->r8;
2174         vcpu->regs[VCPU_REGS_R9] = regs->r9;
2175         vcpu->regs[VCPU_REGS_R10] = regs->r10;
2176         vcpu->regs[VCPU_REGS_R11] = regs->r11;
2177         vcpu->regs[VCPU_REGS_R12] = regs->r12;
2178         vcpu->regs[VCPU_REGS_R13] = regs->r13;
2179         vcpu->regs[VCPU_REGS_R14] = regs->r14;
2180         vcpu->regs[VCPU_REGS_R15] = regs->r15;
2181 #endif
2182
2183         vcpu->rip = regs->rip;
2184         kvm_x86_ops->set_rflags(vcpu, regs->rflags);
2185
2186         kvm_x86_ops->decache_regs(vcpu);
2187
2188         vcpu_put(vcpu);
2189
2190         return 0;
2191 }
2192
2193 static void get_segment(struct kvm_vcpu *vcpu,
2194                         struct kvm_segment *var, int seg)
2195 {
2196         return kvm_x86_ops->get_segment(vcpu, var, seg);
2197 }
2198
2199 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
2200 {
2201         struct kvm_segment cs;
2202
2203         get_segment(vcpu, &cs, VCPU_SREG_CS);
2204         *db = cs.db;
2205         *l = cs.l;
2206 }
2207 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
2208
2209 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
2210                                   struct kvm_sregs *sregs)
2211 {
2212         struct descriptor_table dt;
2213         int pending_vec;
2214
2215         vcpu_load(vcpu);
2216
2217         get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
2218         get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
2219         get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
2220         get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
2221         get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
2222         get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
2223
2224         get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
2225         get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2226
2227         kvm_x86_ops->get_idt(vcpu, &dt);
2228         sregs->idt.limit = dt.limit;
2229         sregs->idt.base = dt.base;
2230         kvm_x86_ops->get_gdt(vcpu, &dt);
2231         sregs->gdt.limit = dt.limit;
2232         sregs->gdt.base = dt.base;
2233
2234         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
2235         sregs->cr0 = vcpu->cr0;
2236         sregs->cr2 = vcpu->cr2;
2237         sregs->cr3 = vcpu->cr3;
2238         sregs->cr4 = vcpu->cr4;
2239         sregs->cr8 = get_cr8(vcpu);
2240         sregs->efer = vcpu->shadow_efer;
2241         sregs->apic_base = kvm_get_apic_base(vcpu);
2242
2243         if (irqchip_in_kernel(vcpu->kvm)) {
2244                 memset(sregs->interrupt_bitmap, 0,
2245                        sizeof sregs->interrupt_bitmap);
2246                 pending_vec = kvm_x86_ops->get_irq(vcpu);
2247                 if (pending_vec >= 0)
2248                         set_bit(pending_vec,
2249                                 (unsigned long *)sregs->interrupt_bitmap);
2250         } else
2251                 memcpy(sregs->interrupt_bitmap, vcpu->irq_pending,
2252                        sizeof sregs->interrupt_bitmap);
2253
2254         vcpu_put(vcpu);
2255
2256         return 0;
2257 }
2258
2259 static void set_segment(struct kvm_vcpu *vcpu,
2260                         struct kvm_segment *var, int seg)
2261 {
2262         return kvm_x86_ops->set_segment(vcpu, var, seg);
2263 }
2264
2265 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
2266                                   struct kvm_sregs *sregs)
2267 {
2268         int mmu_reset_needed = 0;
2269         int i, pending_vec, max_bits;
2270         struct descriptor_table dt;
2271
2272         vcpu_load(vcpu);
2273
2274         dt.limit = sregs->idt.limit;
2275         dt.base = sregs->idt.base;
2276         kvm_x86_ops->set_idt(vcpu, &dt);
2277         dt.limit = sregs->gdt.limit;
2278         dt.base = sregs->gdt.base;
2279         kvm_x86_ops->set_gdt(vcpu, &dt);
2280
2281         vcpu->cr2 = sregs->cr2;
2282         mmu_reset_needed |= vcpu->cr3 != sregs->cr3;
2283         vcpu->cr3 = sregs->cr3;
2284
2285         set_cr8(vcpu, sregs->cr8);
2286
2287         mmu_reset_needed |= vcpu->shadow_efer != sregs->efer;
2288 #ifdef CONFIG_X86_64
2289         kvm_x86_ops->set_efer(vcpu, sregs->efer);
2290 #endif
2291         kvm_set_apic_base(vcpu, sregs->apic_base);
2292
2293         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
2294
2295         mmu_reset_needed |= vcpu->cr0 != sregs->cr0;
2296         vcpu->cr0 = sregs->cr0;
2297         kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
2298
2299         mmu_reset_needed |= vcpu->cr4 != sregs->cr4;
2300         kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
2301         if (!is_long_mode(vcpu) && is_pae(vcpu))
2302                 load_pdptrs(vcpu, vcpu->cr3);
2303
2304         if (mmu_reset_needed)
2305                 kvm_mmu_reset_context(vcpu);
2306
2307         if (!irqchip_in_kernel(vcpu->kvm)) {
2308                 memcpy(vcpu->irq_pending, sregs->interrupt_bitmap,
2309                        sizeof vcpu->irq_pending);
2310                 vcpu->irq_summary = 0;
2311                 for (i = 0; i < ARRAY_SIZE(vcpu->irq_pending); ++i)
2312                         if (vcpu->irq_pending[i])
2313                                 __set_bit(i, &vcpu->irq_summary);
2314         } else {
2315                 max_bits = (sizeof sregs->interrupt_bitmap) << 3;
2316                 pending_vec = find_first_bit(
2317                         (const unsigned long *)sregs->interrupt_bitmap,
2318                         max_bits);
2319                 /* Only pending external irq is handled here */
2320                 if (pending_vec < max_bits) {
2321                         kvm_x86_ops->set_irq(vcpu, pending_vec);
2322                         pr_debug("Set back pending irq %d\n",
2323                                  pending_vec);
2324                 }
2325         }
2326
2327         set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
2328         set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
2329         set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
2330         set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
2331         set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
2332         set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
2333
2334         set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
2335         set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2336
2337         vcpu_put(vcpu);
2338
2339         return 0;
2340 }
2341
2342 int kvm_arch_vcpu_ioctl_debug_guest(struct kvm_vcpu *vcpu,
2343                                     struct kvm_debug_guest *dbg)
2344 {
2345         int r;
2346
2347         vcpu_load(vcpu);
2348
2349         r = kvm_x86_ops->set_guest_debug(vcpu, dbg);
2350
2351         vcpu_put(vcpu);
2352
2353         return r;
2354 }
2355
2356 /*
2357  * fxsave fpu state.  Taken from x86_64/processor.h.  To be killed when
2358  * we have asm/x86/processor.h
2359  */
2360 struct fxsave {
2361         u16     cwd;
2362         u16     swd;
2363         u16     twd;
2364         u16     fop;
2365         u64     rip;
2366         u64     rdp;
2367         u32     mxcsr;
2368         u32     mxcsr_mask;
2369         u32     st_space[32];   /* 8*16 bytes for each FP-reg = 128 bytes */
2370 #ifdef CONFIG_X86_64
2371         u32     xmm_space[64];  /* 16*16 bytes for each XMM-reg = 256 bytes */
2372 #else
2373         u32     xmm_space[32];  /* 8*16 bytes for each XMM-reg = 128 bytes */
2374 #endif
2375 };
2376
2377 /*
2378  * Translate a guest virtual address to a guest physical address.
2379  */
2380 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
2381                                     struct kvm_translation *tr)
2382 {
2383         unsigned long vaddr = tr->linear_address;
2384         gpa_t gpa;
2385
2386         vcpu_load(vcpu);
2387         mutex_lock(&vcpu->kvm->lock);
2388         gpa = vcpu->mmu.gva_to_gpa(vcpu, vaddr);
2389         tr->physical_address = gpa;
2390         tr->valid = gpa != UNMAPPED_GVA;
2391         tr->writeable = 1;
2392         tr->usermode = 0;
2393         mutex_unlock(&vcpu->kvm->lock);
2394         vcpu_put(vcpu);
2395
2396         return 0;
2397 }
2398
2399 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2400 {
2401         struct fxsave *fxsave = (struct fxsave *)&vcpu->guest_fx_image;
2402
2403         vcpu_load(vcpu);
2404
2405         memcpy(fpu->fpr, fxsave->st_space, 128);
2406         fpu->fcw = fxsave->cwd;
2407         fpu->fsw = fxsave->swd;
2408         fpu->ftwx = fxsave->twd;
2409         fpu->last_opcode = fxsave->fop;
2410         fpu->last_ip = fxsave->rip;
2411         fpu->last_dp = fxsave->rdp;
2412         memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
2413
2414         vcpu_put(vcpu);
2415
2416         return 0;
2417 }
2418
2419 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2420 {
2421         struct fxsave *fxsave = (struct fxsave *)&vcpu->guest_fx_image;
2422
2423         vcpu_load(vcpu);
2424
2425         memcpy(fxsave->st_space, fpu->fpr, 128);
2426         fxsave->cwd = fpu->fcw;
2427         fxsave->swd = fpu->fsw;
2428         fxsave->twd = fpu->ftwx;
2429         fxsave->fop = fpu->last_opcode;
2430         fxsave->rip = fpu->last_ip;
2431         fxsave->rdp = fpu->last_dp;
2432         memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
2433
2434         vcpu_put(vcpu);
2435
2436         return 0;
2437 }
2438
2439 void fx_init(struct kvm_vcpu *vcpu)
2440 {
2441         unsigned after_mxcsr_mask;
2442
2443         /* Initialize guest FPU by resetting ours and saving into guest's */
2444         preempt_disable();
2445         fx_save(&vcpu->host_fx_image);
2446         fpu_init();
2447         fx_save(&vcpu->guest_fx_image);
2448         fx_restore(&vcpu->host_fx_image);
2449         preempt_enable();
2450
2451         vcpu->cr0 |= X86_CR0_ET;
2452         after_mxcsr_mask = offsetof(struct i387_fxsave_struct, st_space);
2453         vcpu->guest_fx_image.mxcsr = 0x1f80;
2454         memset((void *)&vcpu->guest_fx_image + after_mxcsr_mask,
2455                0, sizeof(struct i387_fxsave_struct) - after_mxcsr_mask);
2456 }
2457 EXPORT_SYMBOL_GPL(fx_init);
2458
2459 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
2460 {
2461         if (!vcpu->fpu_active || vcpu->guest_fpu_loaded)
2462                 return;
2463
2464         vcpu->guest_fpu_loaded = 1;
2465         fx_save(&vcpu->host_fx_image);
2466         fx_restore(&vcpu->guest_fx_image);
2467 }
2468 EXPORT_SYMBOL_GPL(kvm_load_guest_fpu);
2469
2470 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
2471 {
2472         if (!vcpu->guest_fpu_loaded)
2473                 return;
2474
2475         vcpu->guest_fpu_loaded = 0;
2476         fx_save(&vcpu->guest_fx_image);
2477         fx_restore(&vcpu->host_fx_image);
2478         ++vcpu->stat.fpu_reload;
2479 }
2480 EXPORT_SYMBOL_GPL(kvm_put_guest_fpu);
2481
2482 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
2483 {
2484         kvm_x86_ops->vcpu_free(vcpu);
2485 }
2486
2487 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
2488                                                 unsigned int id)
2489 {
2490         int r;
2491         struct kvm_vcpu *vcpu = kvm_x86_ops->vcpu_create(kvm, id);
2492
2493         if (IS_ERR(vcpu)) {
2494                 r = -ENOMEM;
2495                 goto fail;
2496         }
2497
2498         /* We do fxsave: this must be aligned. */
2499         BUG_ON((unsigned long)&vcpu->host_fx_image & 0xF);
2500
2501         vcpu_load(vcpu);
2502         r = kvm_arch_vcpu_reset(vcpu);
2503         if (r == 0)
2504                 r = kvm_mmu_setup(vcpu);
2505         vcpu_put(vcpu);
2506         if (r < 0)
2507                 goto free_vcpu;
2508
2509         return vcpu;
2510 free_vcpu:
2511         kvm_x86_ops->vcpu_free(vcpu);
2512 fail:
2513         return ERR_PTR(r);
2514 }
2515
2516 void kvm_arch_vcpu_destory(struct kvm_vcpu *vcpu)
2517 {
2518         vcpu_load(vcpu);
2519         kvm_mmu_unload(vcpu);
2520         vcpu_put(vcpu);
2521
2522         kvm_x86_ops->vcpu_free(vcpu);
2523 }
2524
2525 int kvm_arch_vcpu_reset(struct kvm_vcpu *vcpu)
2526 {
2527         return kvm_x86_ops->vcpu_reset(vcpu);
2528 }
2529
2530 void kvm_arch_hardware_enable(void *garbage)
2531 {
2532         kvm_x86_ops->hardware_enable(garbage);
2533 }
2534
2535 void kvm_arch_hardware_disable(void *garbage)
2536 {
2537         kvm_x86_ops->hardware_disable(garbage);
2538 }
2539
2540 int kvm_arch_hardware_setup(void)
2541 {
2542         return kvm_x86_ops->hardware_setup();
2543 }
2544
2545 void kvm_arch_hardware_unsetup(void)
2546 {
2547         kvm_x86_ops->hardware_unsetup();
2548 }
2549
2550 void kvm_arch_check_processor_compat(void *rtn)
2551 {
2552         kvm_x86_ops->check_processor_compatibility(rtn);
2553 }
2554
2555 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
2556 {
2557         struct page *page;
2558         struct kvm *kvm;
2559         int r;
2560
2561         BUG_ON(vcpu->kvm == NULL);
2562         kvm = vcpu->kvm;
2563
2564         vcpu->mmu.root_hpa = INVALID_PAGE;
2565         if (!irqchip_in_kernel(kvm) || vcpu->vcpu_id == 0)
2566                 vcpu->mp_state = VCPU_MP_STATE_RUNNABLE;
2567         else
2568                 vcpu->mp_state = VCPU_MP_STATE_UNINITIALIZED;
2569
2570         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2571         if (!page) {
2572                 r = -ENOMEM;
2573                 goto fail;
2574         }
2575         vcpu->pio_data = page_address(page);
2576
2577         r = kvm_mmu_create(vcpu);
2578         if (r < 0)
2579                 goto fail_free_pio_data;
2580
2581         if (irqchip_in_kernel(kvm)) {
2582                 r = kvm_create_lapic(vcpu);
2583                 if (r < 0)
2584                         goto fail_mmu_destroy;
2585         }
2586
2587         return 0;
2588
2589 fail_mmu_destroy:
2590         kvm_mmu_destroy(vcpu);
2591 fail_free_pio_data:
2592         free_page((unsigned long)vcpu->pio_data);
2593 fail:
2594         return r;
2595 }
2596
2597 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
2598 {
2599         kvm_free_lapic(vcpu);
2600         kvm_mmu_destroy(vcpu);
2601         free_page((unsigned long)vcpu->pio_data);
2602 }
2603
2604 struct  kvm *kvm_arch_create_vm(void)
2605 {
2606         struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
2607
2608         if (!kvm)
2609                 return ERR_PTR(-ENOMEM);
2610
2611         INIT_LIST_HEAD(&kvm->active_mmu_pages);
2612
2613         return kvm;
2614 }
2615
2616 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
2617 {
2618         vcpu_load(vcpu);
2619         kvm_mmu_unload(vcpu);
2620         vcpu_put(vcpu);
2621 }
2622
2623 static void kvm_free_vcpus(struct kvm *kvm)
2624 {
2625         unsigned int i;
2626
2627         /*
2628          * Unpin any mmu pages first.
2629          */
2630         for (i = 0; i < KVM_MAX_VCPUS; ++i)
2631                 if (kvm->vcpus[i])
2632                         kvm_unload_vcpu_mmu(kvm->vcpus[i]);
2633         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
2634                 if (kvm->vcpus[i]) {
2635                         kvm_arch_vcpu_free(kvm->vcpus[i]);
2636                         kvm->vcpus[i] = NULL;
2637                 }
2638         }
2639
2640 }
2641
2642 void kvm_arch_destroy_vm(struct kvm *kvm)
2643 {
2644         kfree(kvm->vpic);
2645         kfree(kvm->vioapic);
2646         kvm_free_vcpus(kvm);
2647         kvm_free_physmem(kvm);
2648         kfree(kvm);
2649 }