drivers/kvm/vmx.c

   1 /*
   2  * Kernel-based Virtual Machine driver for Linux
   3  *
   4  * This module enables machines with Intel VT-x extensions to run virtual
   5  * machines without emulation or binary translation.
   6  *
   7  * Copyright (C) 2006 Qumranet, Inc.
   8  *
   9  * Authors:
  10  *   Avi Kivity   <avi@qumranet.com>
  11  *   Yaniv Kamay  <yaniv@qumranet.com>
  12  *
  13  * This work is licensed under the terms of the GNU GPL, version 2.  See
  14  * the COPYING file in the top-level directory.
  15  *
  16  */
  17
  18 #include "kvm.h"
  19 #include "vmx.h"
  20 #include "segment_descriptor.h"
  21
  22 #include <linux/module.h>
  23 #include <linux/kernel.h>
  24 #include <linux/mm.h>
  25 #include <linux/highmem.h>
  26 #include <linux/profile.h>
  27 #include <linux/sched.h>
  28
  29 #include <asm/io.h>
  30 #include <asm/desc.h>
  31
  32 MODULE_AUTHOR("Qumranet");
  33 MODULE_LICENSE("GPL");
  34
  35 static int init_rmode_tss(struct kvm *kvm);
  36
  37 static DEFINE_PER_CPU(struct vmcs *, vmxarea);
  38 static DEFINE_PER_CPU(struct vmcs *, current_vmcs);
  39
  40 static struct page *vmx_io_bitmap_a;
  41 static struct page *vmx_io_bitmap_b;
  42
  43 #ifdef CONFIG_X86_64
  44 #define HOST_IS_64 1
  45 #else
  46 #define HOST_IS_64 0
  47 #endif
  48 #define EFER_SAVE_RESTORE_BITS ((u64)EFER_SCE)
  49
  50 static struct vmcs_descriptor {
  51         int size;
  52         int order;
  53         u32 revision_id;
  54 } vmcs_descriptor;
  55
  56 #define VMX_SEGMENT_FIELD(seg)                                  \
  57         [VCPU_SREG_##seg] = {                                   \
  58                 .selector = GUEST_##seg##_SELECTOR,             \
  59                 .base = GUEST_##seg##_BASE,                     \
  60                 .limit = GUEST_##seg##_LIMIT,                   \
  61                 .ar_bytes = GUEST_##seg##_AR_BYTES,             \
  62         }
  63
  64 static struct kvm_vmx_segment_field {
  65         unsigned selector;
  66         unsigned base;
  67         unsigned limit;
  68         unsigned ar_bytes;
  69 } kvm_vmx_segment_fields[] = {
  70         VMX_SEGMENT_FIELD(CS),
  71         VMX_SEGMENT_FIELD(DS),
  72         VMX_SEGMENT_FIELD(ES),
  73         VMX_SEGMENT_FIELD(FS),
  74         VMX_SEGMENT_FIELD(GS),
  75         VMX_SEGMENT_FIELD(SS),
  76         VMX_SEGMENT_FIELD(TR),
  77         VMX_SEGMENT_FIELD(LDTR),
  78 };
  79
  80 /*
  81  * Keep MSR_K6_STAR at the end, as setup_msrs() will try to optimize it
  82  * away by decrementing the array size.
  83  */
  84 static const u32 vmx_msr_index[] = {
  85 #ifdef CONFIG_X86_64
  86         MSR_SYSCALL_MASK, MSR_LSTAR, MSR_CSTAR, MSR_KERNEL_GS_BASE,
  87 #endif
  88         MSR_EFER, MSR_K6_STAR,
  89 };
  90 #define NR_VMX_MSR ARRAY_SIZE(vmx_msr_index)
  91
  92 static inline u64 msr_efer_save_restore_bits(struct vmx_msr_entry msr)
  93 {
  94         return (u64)msr.data & EFER_SAVE_RESTORE_BITS;
  95 }
  96
  97 static inline int msr_efer_need_save_restore(struct kvm_vcpu *vcpu)
  98 {
  99         int efer_offset = vcpu->msr_offset_efer;
 100         return msr_efer_save_restore_bits(vcpu->host_msrs[efer_offset]) !=
 101                 msr_efer_save_restore_bits(vcpu->guest_msrs[efer_offset]);
 102 }
 103
 104 static inline int is_page_fault(u32 intr_info)
 105 {
 106         return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
 107                              INTR_INFO_VALID_MASK)) ==
 108                 (INTR_TYPE_EXCEPTION | PF_VECTOR | INTR_INFO_VALID_MASK);
 109 }
 110
 111 static inline int is_no_device(u32 intr_info)
 112 {
 113         return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
 114                              INTR_INFO_VALID_MASK)) ==
 115                 (INTR_TYPE_EXCEPTION | NM_VECTOR | INTR_INFO_VALID_MASK);
 116 }
 117
 118 static inline int is_external_interrupt(u32 intr_info)
 119 {
 120         return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VALID_MASK))
 121                 == (INTR_TYPE_EXT_INTR | INTR_INFO_VALID_MASK);
 122 }
 123
 124 static int __find_msr_index(struct kvm_vcpu *vcpu, u32 msr)
 125 {
 126         int i;
 127
 128         for (i = 0; i < vcpu->nmsrs; ++i)
 129                 if (vcpu->guest_msrs[i].index == msr)
 130                         return i;
 131         return -1;
 132 }
 133
 134 static struct vmx_msr_entry *find_msr_entry(struct kvm_vcpu *vcpu, u32 msr)
 135 {
 136         int i;
 137
 138         i = __find_msr_index(vcpu, msr);
 139         if (i >= 0)
 140                 return &vcpu->guest_msrs[i];
 141         return NULL;
 142 }
 143
 144 static void vmcs_clear(struct vmcs *vmcs)
 145 {
 146         u64 phys_addr = __pa(vmcs);
 147         u8 error;
 148
 149         asm volatile (ASM_VMX_VMCLEAR_RAX "; setna %0"
 150                       : "=g"(error) : "a"(&phys_addr), "m"(phys_addr)
 151                       : "cc", "memory");
 152         if (error)
 153                 printk(KERN_ERR "kvm: vmclear fail: %p/%llx\n",
 154                        vmcs, phys_addr);
 155 }
 156
 157 static void __vcpu_clear(void *arg)
 158 {
 159         struct kvm_vcpu *vcpu = arg;
 160         int cpu = raw_smp_processor_id();
 161
 162         if (vcpu->cpu == cpu)
 163                 vmcs_clear(vcpu->vmcs);
 164         if (per_cpu(current_vmcs, cpu) == vcpu->vmcs)
 165                 per_cpu(current_vmcs, cpu) = NULL;
 166         rdtscll(vcpu->host_tsc);
 167 }
 168
 169 static void vcpu_clear(struct kvm_vcpu *vcpu)
 170 {
 171         if (vcpu->cpu != raw_smp_processor_id() && vcpu->cpu != -1)
 172                 smp_call_function_single(vcpu->cpu, __vcpu_clear, vcpu, 0, 1);
 173         else
 174                 __vcpu_clear(vcpu);
 175         vcpu->launched = 0;
 176 }
 177
 178 static unsigned long vmcs_readl(unsigned long field)
 179 {
 180         unsigned long value;
 181
 182         asm volatile (ASM_VMX_VMREAD_RDX_RAX
 183                       : "=a"(value) : "d"(field) : "cc");
 184         return value;
 185 }
 186
 187 static u16 vmcs_read16(unsigned long field)
 188 {
 189         return vmcs_readl(field);
 190 }
 191
 192 static u32 vmcs_read32(unsigned long field)
 193 {
 194         return vmcs_readl(field);
 195 }
 196
 197 static u64 vmcs_read64(unsigned long field)
 198 {
 199 #ifdef CONFIG_X86_64
 200         return vmcs_readl(field);
 201 #else
 202         return vmcs_readl(field) | ((u64)vmcs_readl(field+1) << 32);
 203 #endif
 204 }
 205
 206 static noinline void vmwrite_error(unsigned long field, unsigned long value)
 207 {
 208         printk(KERN_ERR "vmwrite error: reg %lx value %lx (err %d)\n",
 209                field, value, vmcs_read32(VM_INSTRUCTION_ERROR));
 210         dump_stack();
 211 }
 212
 213 static void vmcs_writel(unsigned long field, unsigned long value)
 214 {
 215         u8 error;
 216
 217         asm volatile (ASM_VMX_VMWRITE_RAX_RDX "; setna %0"
 218                        : "=q"(error) : "a"(value), "d"(field) : "cc" );
 219         if (unlikely(error))
 220                 vmwrite_error(field, value);
 221 }
 222
 223 static void vmcs_write16(unsigned long field, u16 value)
 224 {
 225         vmcs_writel(field, value);
 226 }
 227
 228 static void vmcs_write32(unsigned long field, u32 value)
 229 {
 230         vmcs_writel(field, value);
 231 }
 232
 233 static void vmcs_write64(unsigned long field, u64 value)
 234 {
 235 #ifdef CONFIG_X86_64
 236         vmcs_writel(field, value);
 237 #else
 238         vmcs_writel(field, value);
 239         asm volatile ("");
 240         vmcs_writel(field+1, value >> 32);
 241 #endif
 242 }
 243
 244 static void vmcs_clear_bits(unsigned long field, u32 mask)
 245 {
 246         vmcs_writel(field, vmcs_readl(field) & ~mask);
 247 }
 248
 249 static void vmcs_set_bits(unsigned long field, u32 mask)
 250 {
 251         vmcs_writel(field, vmcs_readl(field) | mask);
 252 }
 253
 254 static void update_exception_bitmap(struct kvm_vcpu *vcpu)
 255 {
 256         u32 eb;
 257
 258         eb = 1u << PF_VECTOR;
 259         if (!vcpu->fpu_active)
 260                 eb |= 1u << NM_VECTOR;
 261         if (vcpu->guest_debug.enabled)
 262                 eb |= 1u << 1;
 263         if (vcpu->rmode.active)
 264                 eb = ~0;
 265         vmcs_write32(EXCEPTION_BITMAP, eb);
 266 }
 267
 268 static void reload_tss(void)
 269 {
 270 #ifndef CONFIG_X86_64
 271
 272         /*
 273          * VT restores TR but not its size.  Useless.
 274          */
 275         struct descriptor_table gdt;
 276         struct segment_descriptor *descs;
 277
 278         get_gdt(&gdt);
 279         descs = (void *)gdt.base;
 280         descs[GDT_ENTRY_TSS].type = 9; /* available TSS */
 281         load_TR_desc();
 282 #endif
 283 }
 284
 285 static void load_transition_efer(struct kvm_vcpu *vcpu)
 286 {
 287         u64 trans_efer;
 288         int efer_offset = vcpu->msr_offset_efer;
 289
 290         trans_efer = vcpu->host_msrs[efer_offset].data;
 291         trans_efer &= ~EFER_SAVE_RESTORE_BITS;
 292         trans_efer |= msr_efer_save_restore_bits(
 293                                 vcpu->guest_msrs[efer_offset]);
 294         wrmsrl(MSR_EFER, trans_efer);
 295         vcpu->stat.efer_reload++;
 296 }
 297
 298 static void vmx_save_host_state(struct kvm_vcpu *vcpu)
 299 {
 300         struct vmx_host_state *hs = &vcpu->vmx_host_state;
 301
 302         if (hs->loaded)
 303                 return;
 304
 305         hs->loaded = 1;
 306         /*
 307          * Set host fs and gs selectors.  Unfortunately, 22.2.3 does not
 308          * allow segment selectors with cpl > 0 or ti == 1.
 309          */
 310         hs->ldt_sel = read_ldt();
 311         hs->fs_gs_ldt_reload_needed = hs->ldt_sel;
 312         hs->fs_sel = read_fs();
 313         if (!(hs->fs_sel & 7))
 314                 vmcs_write16(HOST_FS_SELECTOR, hs->fs_sel);
 315         else {
 316                 vmcs_write16(HOST_FS_SELECTOR, 0);
 317                 hs->fs_gs_ldt_reload_needed = 1;
 318         }
 319         hs->gs_sel = read_gs();
 320         if (!(hs->gs_sel & 7))
 321                 vmcs_write16(HOST_GS_SELECTOR, hs->gs_sel);
 322         else {
 323                 vmcs_write16(HOST_GS_SELECTOR, 0);
 324                 hs->fs_gs_ldt_reload_needed = 1;
 325         }
 326
 327 #ifdef CONFIG_X86_64
 328         vmcs_writel(HOST_FS_BASE, read_msr(MSR_FS_BASE));
 329         vmcs_writel(HOST_GS_BASE, read_msr(MSR_GS_BASE));
 330 #else
 331         vmcs_writel(HOST_FS_BASE, segment_base(hs->fs_sel));
 332         vmcs_writel(HOST_GS_BASE, segment_base(hs->gs_sel));
 333 #endif
 334
 335 #ifdef CONFIG_X86_64
 336         if (is_long_mode(vcpu)) {
 337                 save_msrs(vcpu->host_msrs + vcpu->msr_offset_kernel_gs_base, 1);
 338         }
 339 #endif
 340         load_msrs(vcpu->guest_msrs, vcpu->save_nmsrs);
 341         if (msr_efer_need_save_restore(vcpu))
 342                 load_transition_efer(vcpu);
 343 }
 344
 345 static void vmx_load_host_state(struct kvm_vcpu *vcpu)
 346 {
 347         struct vmx_host_state *hs = &vcpu->vmx_host_state;
 348
 349         if (!hs->loaded)
 350                 return;
 351
 352         hs->loaded = 0;
 353         if (hs->fs_gs_ldt_reload_needed) {
 354                 load_ldt(hs->ldt_sel);
 355                 load_fs(hs->fs_sel);
 356                 /*
 357                  * If we have to reload gs, we must take care to
 358                  * preserve our gs base.
 359                  */
 360                 local_irq_disable();
 361                 load_gs(hs->gs_sel);
 362 #ifdef CONFIG_X86_64
 363                 wrmsrl(MSR_GS_BASE, vmcs_readl(HOST_GS_BASE));
 364 #endif
 365                 local_irq_enable();
 366
 367                 reload_tss();
 368         }
 369         save_msrs(vcpu->guest_msrs, vcpu->save_nmsrs);
 370         load_msrs(vcpu->host_msrs, vcpu->save_nmsrs);
 371         if (msr_efer_need_save_restore(vcpu))
 372                 load_msrs(vcpu->host_msrs + vcpu->msr_offset_efer, 1);
 373 }
 374
 375 /*
 376  * Switches to specified vcpu, until a matching vcpu_put(), but assumes
 377  * vcpu mutex is already taken.
 378  */
 379 static void vmx_vcpu_load(struct kvm_vcpu *vcpu)
 380 {
 381         u64 phys_addr = __pa(vcpu->vmcs);
 382         int cpu;
 383         u64 tsc_this, delta;
 384
 385         cpu = get_cpu();
 386
 387         if (vcpu->cpu != cpu)
 388                 vcpu_clear(vcpu);
 389
 390         if (per_cpu(current_vmcs, cpu) != vcpu->vmcs) {
 391                 u8 error;
 392
 393                 per_cpu(current_vmcs, cpu) = vcpu->vmcs;
 394                 asm volatile (ASM_VMX_VMPTRLD_RAX "; setna %0"
 395                               : "=g"(error) : "a"(&phys_addr), "m"(phys_addr)
 396                               : "cc");
 397                 if (error)
 398                         printk(KERN_ERR "kvm: vmptrld %p/%llx fail\n",
 399                                vcpu->vmcs, phys_addr);
 400         }
 401
 402         if (vcpu->cpu != cpu) {
 403                 struct descriptor_table dt;
 404                 unsigned long sysenter_esp;
 405
 406                 vcpu->cpu = cpu;
 407                 /*
 408                  * Linux uses per-cpu TSS and GDT, so set these when switching
 409                  * processors.
 410                  */
 411                 vmcs_writel(HOST_TR_BASE, read_tr_base()); /* 22.2.4 */
 412                 get_gdt(&dt);
 413                 vmcs_writel(HOST_GDTR_BASE, dt.base);   /* 22.2.4 */
 414
 415                 rdmsrl(MSR_IA32_SYSENTER_ESP, sysenter_esp);
 416                 vmcs_writel(HOST_IA32_SYSENTER_ESP, sysenter_esp); /* 22.2.3 */
 417
 418                 /*
 419                  * Make sure the time stamp counter is monotonous.
 420                  */
 421                 rdtscll(tsc_this);
 422                 delta = vcpu->host_tsc - tsc_this;
 423                 vmcs_write64(TSC_OFFSET, vmcs_read64(TSC_OFFSET) + delta);
 424         }
 425 }
 426
 427 static void vmx_vcpu_put(struct kvm_vcpu *vcpu)
 428 {
 429         vmx_load_host_state(vcpu);
 430         kvm_put_guest_fpu(vcpu);
 431         put_cpu();
 432 }
 433
 434 static void vmx_fpu_activate(struct kvm_vcpu *vcpu)
 435 {
 436         if (vcpu->fpu_active)
 437                 return;
 438         vcpu->fpu_active = 1;
 439         vmcs_clear_bits(GUEST_CR0, CR0_TS_MASK);
 440         if (vcpu->cr0 & CR0_TS_MASK)
 441                 vmcs_set_bits(GUEST_CR0, CR0_TS_MASK);
 442         update_exception_bitmap(vcpu);
 443 }
 444
 445 static void vmx_fpu_deactivate(struct kvm_vcpu *vcpu)
 446 {
 447         if (!vcpu->fpu_active)
 448                 return;
 449         vcpu->fpu_active = 0;
 450         vmcs_set_bits(GUEST_CR0, CR0_TS_MASK);
 451         update_exception_bitmap(vcpu);
 452 }
 453
 454 static void vmx_vcpu_decache(struct kvm_vcpu *vcpu)
 455 {
 456         vcpu_clear(vcpu);
 457 }
 458
 459 static unsigned long vmx_get_rflags(struct kvm_vcpu *vcpu)
 460 {
 461         return vmcs_readl(GUEST_RFLAGS);
 462 }
 463
 464 static void vmx_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
 465 {
 466         vmcs_writel(GUEST_RFLAGS, rflags);
 467 }
 468
 469 static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
 470 {
 471         unsigned long rip;
 472         u32 interruptibility;
 473
 474         rip = vmcs_readl(GUEST_RIP);
 475         rip += vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
 476         vmcs_writel(GUEST_RIP, rip);
 477
 478         /*
 479          * We emulated an instruction, so temporary interrupt blocking
 480          * should be removed, if set.
 481          */
 482         interruptibility = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
 483         if (interruptibility & 3)
 484                 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO,
 485                              interruptibility & ~3);
 486         vcpu->interrupt_window_open = 1;
 487 }
 488
 489 static void vmx_inject_gp(struct kvm_vcpu *vcpu, unsigned error_code)
 490 {
 491         printk(KERN_DEBUG "inject_general_protection: rip 0x%lx\n",
 492                vmcs_readl(GUEST_RIP));
 493         vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE, error_code);
 494         vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
 495                      GP_VECTOR |
 496                      INTR_TYPE_EXCEPTION |
 497                      INTR_INFO_DELIEVER_CODE_MASK |
 498                      INTR_INFO_VALID_MASK);
 499 }
 500
 501 /*
 502  * Swap MSR entry in host/guest MSR entry array.
 503  */
 504 void move_msr_up(struct kvm_vcpu *vcpu, int from, int to)
 505 {
 506         struct vmx_msr_entry tmp;
 507         tmp = vcpu->guest_msrs[to];
 508         vcpu->guest_msrs[to] = vcpu->guest_msrs[from];
 509         vcpu->guest_msrs[from] = tmp;
 510         tmp = vcpu->host_msrs[to];
 511         vcpu->host_msrs[to] = vcpu->host_msrs[from];
 512         vcpu->host_msrs[from] = tmp;
 513 }
 514
 515 /*
 516  * Set up the vmcs to automatically save and restore system
 517  * msrs.  Don't touch the 64-bit msrs if the guest is in legacy
 518  * mode, as fiddling with msrs is very expensive.
 519  */
 520 static void setup_msrs(struct kvm_vcpu *vcpu)
 521 {
 522         int save_nmsrs;
 523
 524         save_nmsrs = 0;
 525 #ifdef CONFIG_X86_64
 526         if (is_long_mode(vcpu)) {
 527                 int index;
 528
 529                 index = __find_msr_index(vcpu, MSR_SYSCALL_MASK);
 530                 if (index >= 0)
 531                         move_msr_up(vcpu, index, save_nmsrs++);
 532                 index = __find_msr_index(vcpu, MSR_LSTAR);
 533                 if (index >= 0)
 534                         move_msr_up(vcpu, index, save_nmsrs++);
 535                 index = __find_msr_index(vcpu, MSR_CSTAR);
 536                 if (index >= 0)
 537                         move_msr_up(vcpu, index, save_nmsrs++);
 538                 index = __find_msr_index(vcpu, MSR_KERNEL_GS_BASE);
 539                 if (index >= 0)
 540                         move_msr_up(vcpu, index, save_nmsrs++);
 541                 /*
 542                  * MSR_K6_STAR is only needed on long mode guests, and only
 543                  * if efer.sce is enabled.
 544                  */
 545                 index = __find_msr_index(vcpu, MSR_K6_STAR);
 546                 if ((index >= 0) && (vcpu->shadow_efer & EFER_SCE))
 547                         move_msr_up(vcpu, index, save_nmsrs++);
 548         }
 549 #endif
 550         vcpu->save_nmsrs = save_nmsrs;
 551
 552 #ifdef CONFIG_X86_64
 553         vcpu->msr_offset_kernel_gs_base =
 554                 __find_msr_index(vcpu, MSR_KERNEL_GS_BASE);
 555 #endif
 556         vcpu->msr_offset_efer = __find_msr_index(vcpu, MSR_EFER);
 557 }
 558
 559 /*
 560  * reads and returns guest's timestamp counter "register"
 561  * guest_tsc = host_tsc + tsc_offset    -- 21.3
 562  */
 563 static u64 guest_read_tsc(void)
 564 {
 565         u64 host_tsc, tsc_offset;
 566
 567         rdtscll(host_tsc);
 568         tsc_offset = vmcs_read64(TSC_OFFSET);
 569         return host_tsc + tsc_offset;
 570 }
 571
 572 /*
 573  * writes 'guest_tsc' into guest's timestamp counter "register"
 574  * guest_tsc = host_tsc + tsc_offset ==> tsc_offset = guest_tsc - host_tsc
 575  */
 576 static void guest_write_tsc(u64 guest_tsc)
 577 {
 578         u64 host_tsc;
 579
 580         rdtscll(host_tsc);
 581         vmcs_write64(TSC_OFFSET, guest_tsc - host_tsc);
 582 }
 583
 584 /*
 585  * Reads an msr value (of 'msr_index') into 'pdata'.
 586  * Returns 0 on success, non-0 otherwise.
 587  * Assumes vcpu_load() was already called.
 588  */
 589 static int vmx_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
 590 {
 591         u64 data;
 592         struct vmx_msr_entry *msr;
 593
 594         if (!pdata) {
 595                 printk(KERN_ERR "BUG: get_msr called with NULL pdata\n");
 596                 return -EINVAL;
 597         }
 598
 599         switch (msr_index) {
 600 #ifdef CONFIG_X86_64
 601         case MSR_FS_BASE:
 602                 data = vmcs_readl(GUEST_FS_BASE);
 603                 break;
 604         case MSR_GS_BASE:
 605                 data = vmcs_readl(GUEST_GS_BASE);
 606                 break;
 607         case MSR_EFER:
 608                 return kvm_get_msr_common(vcpu, msr_index, pdata);
 609 #endif
 610         case MSR_IA32_TIME_STAMP_COUNTER:
 611                 data = guest_read_tsc();
 612                 break;
 613         case MSR_IA32_SYSENTER_CS:
 614                 data = vmcs_read32(GUEST_SYSENTER_CS);
 615                 break;
 616         case MSR_IA32_SYSENTER_EIP:
 617                 data = vmcs_readl(GUEST_SYSENTER_EIP);
 618                 break;
 619         case MSR_IA32_SYSENTER_ESP:
 620                 data = vmcs_readl(GUEST_SYSENTER_ESP);
 621                 break;
 622         default:
 623                 msr = find_msr_entry(vcpu, msr_index);
 624                 if (msr) {
 625                         data = msr->data;
 626                         break;
 627                 }
 628                 return kvm_get_msr_common(vcpu, msr_index, pdata);
 629         }
 630
 631         *pdata = data;
 632         return 0;
 633 }
 634
 635 /*
 636  * Writes msr value into into the appropriate "register".
 637  * Returns 0 on success, non-0 otherwise.
 638  * Assumes vcpu_load() was already called.
 639  */
 640 static int vmx_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
 641 {
 642         struct vmx_msr_entry *msr;
 643         int ret = 0;
 644
 645         switch (msr_index) {
 646 #ifdef CONFIG_X86_64
 647         case MSR_EFER:
 648                 ret = kvm_set_msr_common(vcpu, msr_index, data);
 649                 if (vcpu->vmx_host_state.loaded)
 650                         load_transition_efer(vcpu);
 651                 break;
 652         case MSR_FS_BASE:
 653                 vmcs_writel(GUEST_FS_BASE, data);
 654                 break;
 655         case MSR_GS_BASE:
 656                 vmcs_writel(GUEST_GS_BASE, data);
 657                 break;
 658 #endif
 659         case MSR_IA32_SYSENTER_CS:
 660                 vmcs_write32(GUEST_SYSENTER_CS, data);
 661                 break;
 662         case MSR_IA32_SYSENTER_EIP:
 663                 vmcs_writel(GUEST_SYSENTER_EIP, data);
 664                 break;
 665         case MSR_IA32_SYSENTER_ESP:
 666                 vmcs_writel(GUEST_SYSENTER_ESP, data);
 667                 break;
 668         case MSR_IA32_TIME_STAMP_COUNTER:
 669                 guest_write_tsc(data);
 670                 break;
 671         default:
 672                 msr = find_msr_entry(vcpu, msr_index);
 673                 if (msr) {
 674                         msr->data = data;
 675                         if (vcpu->vmx_host_state.loaded)
 676                                 load_msrs(vcpu->guest_msrs, vcpu->save_nmsrs);
 677                         break;
 678                 }
 679                 ret = kvm_set_msr_common(vcpu, msr_index, data);
 680         }
 681
 682         return ret;
 683 }
 684
 685 /*
 686  * Sync the rsp and rip registers into the vcpu structure.  This allows
 687  * registers to be accessed by indexing vcpu->regs.
 688  */
 689 static void vcpu_load_rsp_rip(struct kvm_vcpu *vcpu)
 690 {
 691         vcpu->regs[VCPU_REGS_RSP] = vmcs_readl(GUEST_RSP);
 692         vcpu->rip = vmcs_readl(GUEST_RIP);
 693 }
 694
 695 /*
 696  * Syncs rsp and rip back into the vmcs.  Should be called after possible
 697  * modification.
 698  */
 699 static void vcpu_put_rsp_rip(struct kvm_vcpu *vcpu)
 700 {
 701         vmcs_writel(GUEST_RSP, vcpu->regs[VCPU_REGS_RSP]);
 702         vmcs_writel(GUEST_RIP, vcpu->rip);
 703 }
 704
 705 static int set_guest_debug(struct kvm_vcpu *vcpu, struct kvm_debug_guest *dbg)
 706 {
 707         unsigned long dr7 = 0x400;
 708         int old_singlestep;
 709
 710         old_singlestep = vcpu->guest_debug.singlestep;
 711
 712         vcpu->guest_debug.enabled = dbg->enabled;
 713         if (vcpu->guest_debug.enabled) {
 714                 int i;
 715
 716                 dr7 |= 0x200;  /* exact */
 717                 for (i = 0; i < 4; ++i) {
 718                         if (!dbg->breakpoints[i].enabled)
 719                                 continue;
 720                         vcpu->guest_debug.bp[i] = dbg->breakpoints[i].address;
 721                         dr7 |= 2 << (i*2);    /* global enable */
 722                         dr7 |= 0 << (i*4+16); /* execution breakpoint */
 723                 }
 724
 725                 vcpu->guest_debug.singlestep = dbg->singlestep;
 726         } else
 727                 vcpu->guest_debug.singlestep = 0;
 728
 729         if (old_singlestep && !vcpu->guest_debug.singlestep) {
 730                 unsigned long flags;
 731
 732                 flags = vmcs_readl(GUEST_RFLAGS);
 733                 flags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
 734                 vmcs_writel(GUEST_RFLAGS, flags);
 735         }
 736
 737         update_exception_bitmap(vcpu);
 738         vmcs_writel(GUEST_DR7, dr7);
 739
 740         return 0;
 741 }
 742
 743 static __init int cpu_has_kvm_support(void)
 744 {
 745         unsigned long ecx = cpuid_ecx(1);
 746         return test_bit(5, &ecx); /* CPUID.1:ECX.VMX[bit 5] -> VT */
 747 }
 748
 749 static __init int vmx_disabled_by_bios(void)
 750 {
 751         u64 msr;
 752
 753         rdmsrl(MSR_IA32_FEATURE_CONTROL, msr);
 754         return (msr & 5) == 1; /* locked but not enabled */
 755 }
 756
 757 static void hardware_enable(void *garbage)
 758 {
 759         int cpu = raw_smp_processor_id();
 760         u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
 761         u64 old;
 762
 763         rdmsrl(MSR_IA32_FEATURE_CONTROL, old);
 764         if ((old & 5) != 5)
 765                 /* enable and lock */
 766                 wrmsrl(MSR_IA32_FEATURE_CONTROL, old | 5);
 767         write_cr4(read_cr4() | CR4_VMXE); /* FIXME: not cpu hotplug safe */
 768         asm volatile (ASM_VMX_VMXON_RAX : : "a"(&phys_addr), "m"(phys_addr)
 769                       : "memory", "cc");
 770 }
 771
 772 static void hardware_disable(void *garbage)
 773 {
 774         asm volatile (ASM_VMX_VMXOFF : : : "cc");
 775 }
 776
 777 static __init void setup_vmcs_descriptor(void)
 778 {
 779         u32 vmx_msr_low, vmx_msr_high;
 780
 781         rdmsr(MSR_IA32_VMX_BASIC, vmx_msr_low, vmx_msr_high);
 782         vmcs_descriptor.size = vmx_msr_high & 0x1fff;
 783         vmcs_descriptor.order = get_order(vmcs_descriptor.size);
 784         vmcs_descriptor.revision_id = vmx_msr_low;
 785 }
 786
 787 static struct vmcs *alloc_vmcs_cpu(int cpu)
 788 {
 789         int node = cpu_to_node(cpu);
 790         struct page *pages;
 791         struct vmcs *vmcs;
 792
 793         pages = alloc_pages_node(node, GFP_KERNEL, vmcs_descriptor.order);
 794         if (!pages)
 795                 return NULL;
 796         vmcs = page_address(pages);
 797         memset(vmcs, 0, vmcs_descriptor.size);
 798         vmcs->revision_id = vmcs_descriptor.revision_id; /* vmcs revision id */
 799         return vmcs;
 800 }
 801
 802 static struct vmcs *alloc_vmcs(void)
 803 {
 804         return alloc_vmcs_cpu(raw_smp_processor_id());
 805 }
 806
 807 static void free_vmcs(struct vmcs *vmcs)
 808 {
 809         free_pages((unsigned long)vmcs, vmcs_descriptor.order);
 810 }
 811
 812 static void free_kvm_area(void)
 813 {
 814         int cpu;
 815
 816         for_each_online_cpu(cpu)
 817                 free_vmcs(per_cpu(vmxarea, cpu));
 818 }
 819
 820 extern struct vmcs *alloc_vmcs_cpu(int cpu);
 821
 822 static __init int alloc_kvm_area(void)
 823 {
 824         int cpu;
 825
 826         for_each_online_cpu(cpu) {
 827                 struct vmcs *vmcs;
 828
 829                 vmcs = alloc_vmcs_cpu(cpu);
 830                 if (!vmcs) {
 831                         free_kvm_area();
 832                         return -ENOMEM;
 833                 }
 834
 835                 per_cpu(vmxarea, cpu) = vmcs;
 836         }
 837         return 0;
 838 }
 839
 840 static __init int hardware_setup(void)
 841 {
 842         setup_vmcs_descriptor();
 843         return alloc_kvm_area();
 844 }
 845
 846 static __exit void hardware_unsetup(void)
 847 {
 848         free_kvm_area();
 849 }
 850
 851 static void fix_pmode_dataseg(int seg, struct kvm_save_segment *save)
 852 {
 853         struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
 854
 855         if (vmcs_readl(sf->base) == save->base && (save->base & AR_S_MASK)) {
 856                 vmcs_write16(sf->selector, save->selector);
 857                 vmcs_writel(sf->base, save->base);
 858                 vmcs_write32(sf->limit, save->limit);
 859                 vmcs_write32(sf->ar_bytes, save->ar);
 860         } else {
 861                 u32 dpl = (vmcs_read16(sf->selector) & SELECTOR_RPL_MASK)
 862                         << AR_DPL_SHIFT;
 863                 vmcs_write32(sf->ar_bytes, 0x93 | dpl);
 864         }
 865 }
 866
 867 static void enter_pmode(struct kvm_vcpu *vcpu)
 868 {
 869         unsigned long flags;
 870
 871         vcpu->rmode.active = 0;
 872
 873         vmcs_writel(GUEST_TR_BASE, vcpu->rmode.tr.base);
 874         vmcs_write32(GUEST_TR_LIMIT, vcpu->rmode.tr.limit);
 875         vmcs_write32(GUEST_TR_AR_BYTES, vcpu->rmode.tr.ar);
 876
 877         flags = vmcs_readl(GUEST_RFLAGS);
 878         flags &= ~(IOPL_MASK | X86_EFLAGS_VM);
 879         flags |= (vcpu->rmode.save_iopl << IOPL_SHIFT);
 880         vmcs_writel(GUEST_RFLAGS, flags);
 881
 882         vmcs_writel(GUEST_CR4, (vmcs_readl(GUEST_CR4) & ~CR4_VME_MASK) |
 883                         (vmcs_readl(CR4_READ_SHADOW) & CR4_VME_MASK));
 884
 885         update_exception_bitmap(vcpu);
 886
 887         fix_pmode_dataseg(VCPU_SREG_ES, &vcpu->rmode.es);
 888         fix_pmode_dataseg(VCPU_SREG_DS, &vcpu->rmode.ds);
 889         fix_pmode_dataseg(VCPU_SREG_GS, &vcpu->rmode.gs);
 890         fix_pmode_dataseg(VCPU_SREG_FS, &vcpu->rmode.fs);
 891
 892         vmcs_write16(GUEST_SS_SELECTOR, 0);
 893         vmcs_write32(GUEST_SS_AR_BYTES, 0x93);
 894
 895         vmcs_write16(GUEST_CS_SELECTOR,
 896                      vmcs_read16(GUEST_CS_SELECTOR) & ~SELECTOR_RPL_MASK);
 897         vmcs_write32(GUEST_CS_AR_BYTES, 0x9b);
 898 }
 899
 900 static int rmode_tss_base(struct kvm* kvm)
 901 {
 902         gfn_t base_gfn = kvm->memslots[0].base_gfn + kvm->memslots[0].npages - 3;
 903         return base_gfn << PAGE_SHIFT;
 904 }
 905
 906 static void fix_rmode_seg(int seg, struct kvm_save_segment *save)
 907 {
 908         struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
 909
 910         save->selector = vmcs_read16(sf->selector);
 911         save->base = vmcs_readl(sf->base);
 912         save->limit = vmcs_read32(sf->limit);
 913         save->ar = vmcs_read32(sf->ar_bytes);
 914         vmcs_write16(sf->selector, vmcs_readl(sf->base) >> 4);
 915         vmcs_write32(sf->limit, 0xffff);
 916         vmcs_write32(sf->ar_bytes, 0xf3);
 917 }
 918
 919 static void enter_rmode(struct kvm_vcpu *vcpu)
 920 {
 921         unsigned long flags;
 922
 923         vcpu->rmode.active = 1;
 924
 925         vcpu->rmode.tr.base = vmcs_readl(GUEST_TR_BASE);
 926         vmcs_writel(GUEST_TR_BASE, rmode_tss_base(vcpu->kvm));
 927
 928         vcpu->rmode.tr.limit = vmcs_read32(GUEST_TR_LIMIT);
 929         vmcs_write32(GUEST_TR_LIMIT, RMODE_TSS_SIZE - 1);
 930
 931         vcpu->rmode.tr.ar = vmcs_read32(GUEST_TR_AR_BYTES);
 932         vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
 933
 934         flags = vmcs_readl(GUEST_RFLAGS);
 935         vcpu->rmode.save_iopl = (flags & IOPL_MASK) >> IOPL_SHIFT;
 936
 937         flags |= IOPL_MASK | X86_EFLAGS_VM;
 938
 939         vmcs_writel(GUEST_RFLAGS, flags);
 940         vmcs_writel(GUEST_CR4, vmcs_readl(GUEST_CR4) | CR4_VME_MASK);
 941         update_exception_bitmap(vcpu);
 942
 943         vmcs_write16(GUEST_SS_SELECTOR, vmcs_readl(GUEST_SS_BASE) >> 4);
 944         vmcs_write32(GUEST_SS_LIMIT, 0xffff);
 945         vmcs_write32(GUEST_SS_AR_BYTES, 0xf3);
 946
 947         vmcs_write32(GUEST_CS_AR_BYTES, 0xf3);
 948         vmcs_write32(GUEST_CS_LIMIT, 0xffff);
 949         if (vmcs_readl(GUEST_CS_BASE) == 0xffff0000)
 950                 vmcs_writel(GUEST_CS_BASE, 0xf0000);
 951         vmcs_write16(GUEST_CS_SELECTOR, vmcs_readl(GUEST_CS_BASE) >> 4);
 952
 953         fix_rmode_seg(VCPU_SREG_ES, &vcpu->rmode.es);
 954         fix_rmode_seg(VCPU_SREG_DS, &vcpu->rmode.ds);
 955         fix_rmode_seg(VCPU_SREG_GS, &vcpu->rmode.gs);
 956         fix_rmode_seg(VCPU_SREG_FS, &vcpu->rmode.fs);
 957
 958         init_rmode_tss(vcpu->kvm);
 959 }
 960
 961 #ifdef CONFIG_X86_64
 962
 963 static void enter_lmode(struct kvm_vcpu *vcpu)
 964 {
 965         u32 guest_tr_ar;
 966
 967         guest_tr_ar = vmcs_read32(GUEST_TR_AR_BYTES);
 968         if ((guest_tr_ar & AR_TYPE_MASK) != AR_TYPE_BUSY_64_TSS) {
 969                 printk(KERN_DEBUG "%s: tss fixup for long mode. \n",
 970                        __FUNCTION__);
 971                 vmcs_write32(GUEST_TR_AR_BYTES,
 972                              (guest_tr_ar & ~AR_TYPE_MASK)
 973                              | AR_TYPE_BUSY_64_TSS);
 974         }
 975
 976         vcpu->shadow_efer |= EFER_LMA;
 977
 978         find_msr_entry(vcpu, MSR_EFER)->data |= EFER_LMA | EFER_LME;
 979         vmcs_write32(VM_ENTRY_CONTROLS,
 980                      vmcs_read32(VM_ENTRY_CONTROLS)
 981                      | VM_ENTRY_CONTROLS_IA32E_MASK);
 982 }
 983
 984 static void exit_lmode(struct kvm_vcpu *vcpu)
 985 {
 986         vcpu->shadow_efer &= ~EFER_LMA;
 987
 988         vmcs_write32(VM_ENTRY_CONTROLS,
 989                      vmcs_read32(VM_ENTRY_CONTROLS)
 990                      & ~VM_ENTRY_CONTROLS_IA32E_MASK);
 991 }
 992
 993 #endif
 994
 995 static void vmx_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
 996 {
 997         vcpu->cr4 &= KVM_GUEST_CR4_MASK;
 998         vcpu->cr4 |= vmcs_readl(GUEST_CR4) & ~KVM_GUEST_CR4_MASK;
 999 }
1000
1001 static void vmx_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
1002 {
1003         vmx_fpu_deactivate(vcpu);
1004
1005         if (vcpu->rmode.active && (cr0 & CR0_PE_MASK))
1006                 enter_pmode(vcpu);
1007
1008         if (!vcpu->rmode.active && !(cr0 & CR0_PE_MASK))
1009                 enter_rmode(vcpu);
1010
1011 #ifdef CONFIG_X86_64
1012         if (vcpu->shadow_efer & EFER_LME) {
1013                 if (!is_paging(vcpu) && (cr0 & CR0_PG_MASK))
1014                         enter_lmode(vcpu);
1015                 if (is_paging(vcpu) && !(cr0 & CR0_PG_MASK))
1016                         exit_lmode(vcpu);
1017         }
1018 #endif
1019
1020         vmcs_writel(CR0_READ_SHADOW, cr0);
1021         vmcs_writel(GUEST_CR0,
1022                     (cr0 & ~KVM_GUEST_CR0_MASK) | KVM_VM_CR0_ALWAYS_ON);
1023         vcpu->cr0 = cr0;
1024
1025         if (!(cr0 & CR0_TS_MASK) || !(cr0 & CR0_PE_MASK))
1026                 vmx_fpu_activate(vcpu);
1027 }
1028
1029 static void vmx_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
1030 {
1031         vmcs_writel(GUEST_CR3, cr3);
1032         if (vcpu->cr0 & CR0_PE_MASK)
1033                 vmx_fpu_deactivate(vcpu);
1034 }
1035
1036 static void vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
1037 {
1038         vmcs_writel(CR4_READ_SHADOW, cr4);
1039         vmcs_writel(GUEST_CR4, cr4 | (vcpu->rmode.active ?
1040                     KVM_RMODE_VM_CR4_ALWAYS_ON : KVM_PMODE_VM_CR4_ALWAYS_ON));
1041         vcpu->cr4 = cr4;
1042 }
1043
1044 #ifdef CONFIG_X86_64
1045
1046 static void vmx_set_efer(struct kvm_vcpu *vcpu, u64 efer)
1047 {
1048         struct vmx_msr_entry *msr = find_msr_entry(vcpu, MSR_EFER);
1049
1050         vcpu->shadow_efer = efer;
1051         if (efer & EFER_LMA) {
1052                 vmcs_write32(VM_ENTRY_CONTROLS,
1053                                      vmcs_read32(VM_ENTRY_CONTROLS) |
1054                                      VM_ENTRY_CONTROLS_IA32E_MASK);
1055                 msr->data = efer;
1056
1057         } else {
1058                 vmcs_write32(VM_ENTRY_CONTROLS,
1059                                      vmcs_read32(VM_ENTRY_CONTROLS) &
1060                                      ~VM_ENTRY_CONTROLS_IA32E_MASK);
1061
1062                 msr->data = efer & ~EFER_LME;
1063         }
1064         setup_msrs(vcpu);
1065 }
1066
1067 #endif
1068
1069 static u64 vmx_get_segment_base(struct kvm_vcpu *vcpu, int seg)
1070 {
1071         struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1072
1073         return vmcs_readl(sf->base);
1074 }
1075
1076 static void vmx_get_segment(struct kvm_vcpu *vcpu,
1077                             struct kvm_segment *var, int seg)
1078 {
1079         struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1080         u32 ar;
1081
1082         var->base = vmcs_readl(sf->base);
1083         var->limit = vmcs_read32(sf->limit);
1084         var->selector = vmcs_read16(sf->selector);
1085         ar = vmcs_read32(sf->ar_bytes);
1086         if (ar & AR_UNUSABLE_MASK)
1087                 ar = 0;
1088         var->type = ar & 15;
1089         var->s = (ar >> 4) & 1;
1090         var->dpl = (ar >> 5) & 3;
1091         var->present = (ar >> 7) & 1;
1092         var->avl = (ar >> 12) & 1;
1093         var->l = (ar >> 13) & 1;
1094         var->db = (ar >> 14) & 1;
1095         var->g = (ar >> 15) & 1;
1096         var->unusable = (ar >> 16) & 1;
1097 }
1098
1099 static u32 vmx_segment_access_rights(struct kvm_segment *var)
1100 {
1101         u32 ar;
1102
1103         if (var->unusable)
1104                 ar = 1 << 16;
1105         else {
1106                 ar = var->type & 15;
1107                 ar |= (var->s & 1) << 4;
1108                 ar |= (var->dpl & 3) << 5;
1109                 ar |= (var->present & 1) << 7;
1110                 ar |= (var->avl & 1) << 12;
1111                 ar |= (var->l & 1) << 13;
1112                 ar |= (var->db & 1) << 14;
1113                 ar |= (var->g & 1) << 15;
1114         }
1115         if (ar == 0) /* a 0 value means unusable */
1116                 ar = AR_UNUSABLE_MASK;
1117
1118         return ar;
1119 }
1120
1121 static void vmx_set_segment(struct kvm_vcpu *vcpu,
1122                             struct kvm_segment *var, int seg)
1123 {
1124         struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1125         u32 ar;
1126
1127         if (vcpu->rmode.active && seg == VCPU_SREG_TR) {
1128                 vcpu->rmode.tr.selector = var->selector;
1129                 vcpu->rmode.tr.base = var->base;
1130                 vcpu->rmode.tr.limit = var->limit;
1131                 vcpu->rmode.tr.ar = vmx_segment_access_rights(var);
1132                 return;
1133         }
1134         vmcs_writel(sf->base, var->base);
1135         vmcs_write32(sf->limit, var->limit);
1136         vmcs_write16(sf->selector, var->selector);
1137         if (vcpu->rmode.active && var->s) {
1138                 /*
1139                  * Hack real-mode segments into vm86 compatibility.
1140                  */
1141                 if (var->base == 0xffff0000 && var->selector == 0xf000)
1142                         vmcs_writel(sf->base, 0xf0000);
1143                 ar = 0xf3;
1144         } else
1145                 ar = vmx_segment_access_rights(var);
1146         vmcs_write32(sf->ar_bytes, ar);
1147 }
1148
1149 static void vmx_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
1150 {
1151         u32 ar = vmcs_read32(GUEST_CS_AR_BYTES);
1152
1153         *db = (ar >> 14) & 1;
1154         *l = (ar >> 13) & 1;
1155 }
1156
1157 static void vmx_get_idt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
1158 {
1159         dt->limit = vmcs_read32(GUEST_IDTR_LIMIT);
1160         dt->base = vmcs_readl(GUEST_IDTR_BASE);
1161 }
1162
1163 static void vmx_set_idt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
1164 {
1165         vmcs_write32(GUEST_IDTR_LIMIT, dt->limit);
1166         vmcs_writel(GUEST_IDTR_BASE, dt->base);
1167 }
1168
1169 static void vmx_get_gdt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
1170 {
1171         dt->limit = vmcs_read32(GUEST_GDTR_LIMIT);
1172         dt->base = vmcs_readl(GUEST_GDTR_BASE);
1173 }
1174
1175 static void vmx_set_gdt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
1176 {
1177         vmcs_write32(GUEST_GDTR_LIMIT, dt->limit);
1178         vmcs_writel(GUEST_GDTR_BASE, dt->base);
1179 }
1180
1181 static int init_rmode_tss(struct kvm* kvm)
1182 {
1183         struct page *p1, *p2, *p3;
1184         gfn_t fn = rmode_tss_base(kvm) >> PAGE_SHIFT;
1185         char *page;
1186
1187         p1 = gfn_to_page(kvm, fn++);
1188         p2 = gfn_to_page(kvm, fn++);
1189         p3 = gfn_to_page(kvm, fn);
1190
1191         if (!p1 || !p2 || !p3) {
1192                 kvm_printf(kvm,"%s: gfn_to_page failed\n", __FUNCTION__);
1193                 return 0;
1194         }
1195
1196         page = kmap_atomic(p1, KM_USER0);
1197         clear_page(page);
1198         *(u16*)(page + 0x66) = TSS_BASE_SIZE + TSS_REDIRECTION_SIZE;
1199         kunmap_atomic(page, KM_USER0);
1200
1201         page = kmap_atomic(p2, KM_USER0);
1202         clear_page(page);
1203         kunmap_atomic(page, KM_USER0);
1204
1205         page = kmap_atomic(p3, KM_USER0);
1206         clear_page(page);
1207         *(page + RMODE_TSS_SIZE - 2 * PAGE_SIZE - 1) = ~0;
1208         kunmap_atomic(page, KM_USER0);
1209
1210         return 1;
1211 }
1212
1213 static void vmcs_write32_fixedbits(u32 msr, u32 vmcs_field, u32 val)
1214 {
1215         u32 msr_high, msr_low;
1216
1217         rdmsr(msr, msr_low, msr_high);
1218
1219         val &= msr_high;
1220         val |= msr_low;
1221         vmcs_write32(vmcs_field, val);
1222 }
1223
1224 static void seg_setup(int seg)
1225 {
1226         struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1227
1228         vmcs_write16(sf->selector, 0);
1229         vmcs_writel(sf->base, 0);
1230         vmcs_write32(sf->limit, 0xffff);
1231         vmcs_write32(sf->ar_bytes, 0x93);
1232 }
1233
1234 /*
1235  * Sets up the vmcs for emulated real mode.
1236  */
1237 static int vmx_vcpu_setup(struct kvm_vcpu *vcpu)
1238 {
1239         u32 host_sysenter_cs;
1240         u32 junk;
1241         unsigned long a;
1242         struct descriptor_table dt;
1243         int i;
1244         int ret = 0;
1245         unsigned long kvm_vmx_return;
1246
1247         if (!init_rmode_tss(vcpu->kvm)) {
1248                 ret = -ENOMEM;
1249                 goto out;
1250         }
1251
1252         memset(vcpu->regs, 0, sizeof(vcpu->regs));
1253         vcpu->regs[VCPU_REGS_RDX] = get_rdx_init_val();
1254         vcpu->cr8 = 0;
1255         vcpu->apic_base = 0xfee00000 | MSR_IA32_APICBASE_ENABLE;
1256         if (vcpu->vcpu_id == 0)
1257                 vcpu->apic_base |= MSR_IA32_APICBASE_BSP;
1258
1259         fx_init(vcpu);
1260
1261         /*
1262          * GUEST_CS_BASE should really be 0xffff0000, but VT vm86 mode
1263          * insists on having GUEST_CS_BASE == GUEST_CS_SELECTOR << 4.  Sigh.
1264          */
1265         vmcs_write16(GUEST_CS_SELECTOR, 0xf000);
1266         vmcs_writel(GUEST_CS_BASE, 0x000f0000);
1267         vmcs_write32(GUEST_CS_LIMIT, 0xffff);
1268         vmcs_write32(GUEST_CS_AR_BYTES, 0x9b);
1269
1270         seg_setup(VCPU_SREG_DS);
1271         seg_setup(VCPU_SREG_ES);
1272         seg_setup(VCPU_SREG_FS);
1273         seg_setup(VCPU_SREG_GS);
1274         seg_setup(VCPU_SREG_SS);
1275
1276         vmcs_write16(GUEST_TR_SELECTOR, 0);
1277         vmcs_writel(GUEST_TR_BASE, 0);
1278         vmcs_write32(GUEST_TR_LIMIT, 0xffff);
1279         vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
1280
1281         vmcs_write16(GUEST_LDTR_SELECTOR, 0);
1282         vmcs_writel(GUEST_LDTR_BASE, 0);
1283         vmcs_write32(GUEST_LDTR_LIMIT, 0xffff);
1284         vmcs_write32(GUEST_LDTR_AR_BYTES, 0x00082);
1285
1286         vmcs_write32(GUEST_SYSENTER_CS, 0);
1287         vmcs_writel(GUEST_SYSENTER_ESP, 0);
1288         vmcs_writel(GUEST_SYSENTER_EIP, 0);
1289
1290         vmcs_writel(GUEST_RFLAGS, 0x02);
1291         vmcs_writel(GUEST_RIP, 0xfff0);
1292         vmcs_writel(GUEST_RSP, 0);
1293
1294         //todo: dr0 = dr1 = dr2 = dr3 = 0; dr6 = 0xffff0ff0
1295         vmcs_writel(GUEST_DR7, 0x400);
1296
1297         vmcs_writel(GUEST_GDTR_BASE, 0);
1298         vmcs_write32(GUEST_GDTR_LIMIT, 0xffff);
1299
1300         vmcs_writel(GUEST_IDTR_BASE, 0);
1301         vmcs_write32(GUEST_IDTR_LIMIT, 0xffff);
1302
1303         vmcs_write32(GUEST_ACTIVITY_STATE, 0);
1304         vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, 0);
1305         vmcs_write32(GUEST_PENDING_DBG_EXCEPTIONS, 0);
1306
1307         /* I/O */
1308         vmcs_write64(IO_BITMAP_A, page_to_phys(vmx_io_bitmap_a));
1309         vmcs_write64(IO_BITMAP_B, page_to_phys(vmx_io_bitmap_b));
1310
1311         guest_write_tsc(0);
1312
1313         vmcs_write64(VMCS_LINK_POINTER, -1ull); /* 22.3.1.5 */
1314
1315         /* Special registers */
1316         vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
1317
1318         /* Control */
1319         vmcs_write32_fixedbits(MSR_IA32_VMX_PINBASED_CTLS,
1320                                PIN_BASED_VM_EXEC_CONTROL,
1321                                PIN_BASED_EXT_INTR_MASK   /* 20.6.1 */
1322                                | PIN_BASED_NMI_EXITING   /* 20.6.1 */
1323                         );
1324         vmcs_write32_fixedbits(MSR_IA32_VMX_PROCBASED_CTLS,
1325                                CPU_BASED_VM_EXEC_CONTROL,
1326                                CPU_BASED_HLT_EXITING         /* 20.6.2 */
1327                                | CPU_BASED_CR8_LOAD_EXITING    /* 20.6.2 */
1328                                | CPU_BASED_CR8_STORE_EXITING   /* 20.6.2 */
1329                                | CPU_BASED_ACTIVATE_IO_BITMAP  /* 20.6.2 */
1330                                | CPU_BASED_MOV_DR_EXITING
1331                                | CPU_BASED_USE_TSC_OFFSETING   /* 21.3 */
1332                         );
1333
1334         vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, 0);
1335         vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, 0);
1336         vmcs_write32(CR3_TARGET_COUNT, 0);           /* 22.2.1 */
1337
1338         vmcs_writel(HOST_CR0, read_cr0());  /* 22.2.3 */
1339         vmcs_writel(HOST_CR4, read_cr4());  /* 22.2.3, 22.2.5 */
1340         vmcs_writel(HOST_CR3, read_cr3());  /* 22.2.3  FIXME: shadow tables */
1341
1342         vmcs_write16(HOST_CS_SELECTOR, __KERNEL_CS);  /* 22.2.4 */
1343         vmcs_write16(HOST_DS_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
1344         vmcs_write16(HOST_ES_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
1345         vmcs_write16(HOST_FS_SELECTOR, read_fs());    /* 22.2.4 */
1346         vmcs_write16(HOST_GS_SELECTOR, read_gs());    /* 22.2.4 */
1347         vmcs_write16(HOST_SS_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
1348 #ifdef CONFIG_X86_64
1349         rdmsrl(MSR_FS_BASE, a);
1350         vmcs_writel(HOST_FS_BASE, a); /* 22.2.4 */
1351         rdmsrl(MSR_GS_BASE, a);
1352         vmcs_writel(HOST_GS_BASE, a); /* 22.2.4 */
1353 #else
1354         vmcs_writel(HOST_FS_BASE, 0); /* 22.2.4 */
1355         vmcs_writel(HOST_GS_BASE, 0); /* 22.2.4 */
1356 #endif
1357
1358         vmcs_write16(HOST_TR_SELECTOR, GDT_ENTRY_TSS*8);  /* 22.2.4 */
1359
1360         get_idt(&dt);
1361         vmcs_writel(HOST_IDTR_BASE, dt.base);   /* 22.2.4 */
1362
1363         asm ("mov $.Lkvm_vmx_return, %0" : "=r"(kvm_vmx_return));
1364         vmcs_writel(HOST_RIP, kvm_vmx_return); /* 22.2.5 */
1365         vmcs_write32(VM_EXIT_MSR_STORE_COUNT, 0);
1366         vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, 0);
1367         vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, 0);
1368
1369         rdmsr(MSR_IA32_SYSENTER_CS, host_sysenter_cs, junk);
1370         vmcs_write32(HOST_IA32_SYSENTER_CS, host_sysenter_cs);
1371         rdmsrl(MSR_IA32_SYSENTER_ESP, a);
1372         vmcs_writel(HOST_IA32_SYSENTER_ESP, a);   /* 22.2.3 */
1373         rdmsrl(MSR_IA32_SYSENTER_EIP, a);
1374         vmcs_writel(HOST_IA32_SYSENTER_EIP, a);   /* 22.2.3 */
1375
1376         for (i = 0; i < NR_VMX_MSR; ++i) {
1377                 u32 index = vmx_msr_index[i];
1378                 u32 data_low, data_high;
1379                 u64 data;
1380                 int j = vcpu->nmsrs;
1381
1382                 if (rdmsr_safe(index, &data_low, &data_high) < 0)
1383                         continue;
1384                 if (wrmsr_safe(index, data_low, data_high) < 0)
1385                         continue;
1386                 data = data_low | ((u64)data_high << 32);
1387                 vcpu->host_msrs[j].index = index;
1388                 vcpu->host_msrs[j].reserved = 0;
1389                 vcpu->host_msrs[j].data = data;
1390                 vcpu->guest_msrs[j] = vcpu->host_msrs[j];
1391                 ++vcpu->nmsrs;
1392         }
1393
1394         setup_msrs(vcpu);
1395
1396         vmcs_write32_fixedbits(MSR_IA32_VMX_EXIT_CTLS, VM_EXIT_CONTROLS,
1397                                (HOST_IS_64 << 9));  /* 22.2,1, 20.7.1 */
1398
1399         /* 22.2.1, 20.8.1 */
1400         vmcs_write32_fixedbits(MSR_IA32_VMX_ENTRY_CTLS,
1401                                VM_ENTRY_CONTROLS, 0);
1402         vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);  /* 22.2.1 */
1403
1404 #ifdef CONFIG_X86_64
1405         vmcs_writel(VIRTUAL_APIC_PAGE_ADDR, 0);
1406         vmcs_writel(TPR_THRESHOLD, 0);
1407 #endif
1408
1409         vmcs_writel(CR0_GUEST_HOST_MASK, ~0UL);
1410         vmcs_writel(CR4_GUEST_HOST_MASK, KVM_GUEST_CR4_MASK);
1411
1412         vcpu->cr0 = 0x60000010;
1413         vmx_set_cr0(vcpu, vcpu->cr0); // enter rmode
1414         vmx_set_cr4(vcpu, 0);
1415 #ifdef CONFIG_X86_64
1416         vmx_set_efer(vcpu, 0);
1417 #endif
1418         vmx_fpu_activate(vcpu);
1419         update_exception_bitmap(vcpu);
1420
1421         return 0;
1422
1423 out:
1424         return ret;
1425 }
1426
1427 static void inject_rmode_irq(struct kvm_vcpu *vcpu, int irq)
1428 {
1429         u16 ent[2];
1430         u16 cs;
1431         u16 ip;
1432         unsigned long flags;
1433         unsigned long ss_base = vmcs_readl(GUEST_SS_BASE);
1434         u16 sp =  vmcs_readl(GUEST_RSP);
1435         u32 ss_limit = vmcs_read32(GUEST_SS_LIMIT);
1436
1437         if (sp > ss_limit || sp < 6 ) {
1438                 vcpu_printf(vcpu, "%s: #SS, rsp 0x%lx ss 0x%lx limit 0x%x\n",
1439                             __FUNCTION__,
1440                             vmcs_readl(GUEST_RSP),
1441                             vmcs_readl(GUEST_SS_BASE),
1442                             vmcs_read32(GUEST_SS_LIMIT));
1443                 return;
1444         }
1445
1446         if (kvm_read_guest(vcpu, irq * sizeof(ent), sizeof(ent), &ent) !=
1447                                                                 sizeof(ent)) {
1448                 vcpu_printf(vcpu, "%s: read guest err\n", __FUNCTION__);
1449                 return;
1450         }
1451
1452         flags =  vmcs_readl(GUEST_RFLAGS);
1453         cs =  vmcs_readl(GUEST_CS_BASE) >> 4;
1454         ip =  vmcs_readl(GUEST_RIP);
1455
1456
1457         if (kvm_write_guest(vcpu, ss_base + sp - 2, 2, &flags) != 2 ||
1458             kvm_write_guest(vcpu, ss_base + sp - 4, 2, &cs) != 2 ||
1459             kvm_write_guest(vcpu, ss_base + sp - 6, 2, &ip) != 2) {
1460                 vcpu_printf(vcpu, "%s: write guest err\n", __FUNCTION__);
1461                 return;
1462         }
1463
1464         vmcs_writel(GUEST_RFLAGS, flags &
1465                     ~( X86_EFLAGS_IF | X86_EFLAGS_AC | X86_EFLAGS_TF));
1466         vmcs_write16(GUEST_CS_SELECTOR, ent[1]) ;
1467         vmcs_writel(GUEST_CS_BASE, ent[1] << 4);
1468         vmcs_writel(GUEST_RIP, ent[0]);
1469         vmcs_writel(GUEST_RSP, (vmcs_readl(GUEST_RSP) & ~0xffff) | (sp - 6));
1470 }
1471
1472 static void kvm_do_inject_irq(struct kvm_vcpu *vcpu)
1473 {
1474         int word_index = __ffs(vcpu->irq_summary);
1475         int bit_index = __ffs(vcpu->irq_pending[word_index]);
1476         int irq = word_index * BITS_PER_LONG + bit_index;
1477
1478         clear_bit(bit_index, &vcpu->irq_pending[word_index]);
1479         if (!vcpu->irq_pending[word_index])
1480                 clear_bit(word_index, &vcpu->irq_summary);
1481
1482         if (vcpu->rmode.active) {
1483                 inject_rmode_irq(vcpu, irq);
1484                 return;
1485         }
1486         vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
1487                         irq | INTR_TYPE_EXT_INTR | INTR_INFO_VALID_MASK);
1488 }
1489
1490
1491 static void do_interrupt_requests(struct kvm_vcpu *vcpu,
1492                                        struct kvm_run *kvm_run)
1493 {
1494         u32 cpu_based_vm_exec_control;
1495
1496         vcpu->interrupt_window_open =
1497                 ((vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_IF) &&
1498                  (vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & 3) == 0);
1499
1500         if (vcpu->interrupt_window_open &&
1501             vcpu->irq_summary &&
1502             !(vmcs_read32(VM_ENTRY_INTR_INFO_FIELD) & INTR_INFO_VALID_MASK))
1503                 /*
1504                  * If interrupts enabled, and not blocked by sti or mov ss. Good.
1505                  */
1506                 kvm_do_inject_irq(vcpu);
1507
1508         cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
1509         if (!vcpu->interrupt_window_open &&
1510             (vcpu->irq_summary || kvm_run->request_interrupt_window))
1511                 /*
1512                  * Interrupts blocked.  Wait for unblock.
1513                  */
1514                 cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_INTR_PENDING;
1515         else
1516                 cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_INTR_PENDING;
1517         vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
1518 }
1519
1520 static void kvm_guest_debug_pre(struct kvm_vcpu *vcpu)
1521 {
1522         struct kvm_guest_debug *dbg = &vcpu->guest_debug;
1523
1524         set_debugreg(dbg->bp[0], 0);
1525         set_debugreg(dbg->bp[1], 1);
1526         set_debugreg(dbg->bp[2], 2);
1527         set_debugreg(dbg->bp[3], 3);
1528
1529         if (dbg->singlestep) {
1530                 unsigned long flags;
1531
1532                 flags = vmcs_readl(GUEST_RFLAGS);
1533                 flags |= X86_EFLAGS_TF | X86_EFLAGS_RF;
1534                 vmcs_writel(GUEST_RFLAGS, flags);
1535         }
1536 }
1537
1538 static int handle_rmode_exception(struct kvm_vcpu *vcpu,
1539                                   int vec, u32 err_code)
1540 {
1541         if (!vcpu->rmode.active)
1542                 return 0;
1543
1544         /*
1545          * Instruction with address size override prefix opcode 0x67
1546          * Cause the #SS fault with 0 error code in VM86 mode.
1547          */
1548         if (((vec == GP_VECTOR) || (vec == SS_VECTOR)) && err_code == 0)
1549                 if (emulate_instruction(vcpu, NULL, 0, 0) == EMULATE_DONE)
1550                         return 1;
1551         return 0;
1552 }
1553
1554 static int handle_exception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1555 {
1556         u32 intr_info, error_code;
1557         unsigned long cr2, rip;
1558         u32 vect_info;
1559         enum emulation_result er;
1560         int r;
1561
1562         vect_info = vmcs_read32(IDT_VECTORING_INFO_FIELD);
1563         intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
1564
1565         if ((vect_info & VECTORING_INFO_VALID_MASK) &&
1566                                                 !is_page_fault(intr_info)) {
1567                 printk(KERN_ERR "%s: unexpected, vectoring info 0x%x "
1568                        "intr info 0x%x\n", __FUNCTION__, vect_info, intr_info);
1569         }
1570
1571         if (is_external_interrupt(vect_info)) {
1572                 int irq = vect_info & VECTORING_INFO_VECTOR_MASK;
1573                 set_bit(irq, vcpu->irq_pending);
1574                 set_bit(irq / BITS_PER_LONG, &vcpu->irq_summary);
1575         }
1576
1577         if ((intr_info & INTR_INFO_INTR_TYPE_MASK) == 0x200) { /* nmi */
1578                 asm ("int $2");
1579                 return 1;
1580         }
1581
1582         if (is_no_device(intr_info)) {
1583                 vmx_fpu_activate(vcpu);
1584                 return 1;
1585         }
1586
1587         error_code = 0;
1588         rip = vmcs_readl(GUEST_RIP);
1589         if (intr_info & INTR_INFO_DELIEVER_CODE_MASK)
1590                 error_code = vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
1591         if (is_page_fault(intr_info)) {
1592                 cr2 = vmcs_readl(EXIT_QUALIFICATION);
1593
1594                 spin_lock(&vcpu->kvm->lock);
1595                 r = kvm_mmu_page_fault(vcpu, cr2, error_code);
1596                 if (r < 0) {
1597                         spin_unlock(&vcpu->kvm->lock);
1598                         return r;
1599                 }
1600                 if (!r) {
1601                         spin_unlock(&vcpu->kvm->lock);
1602                         return 1;
1603                 }
1604
1605                 er = emulate_instruction(vcpu, kvm_run, cr2, error_code);
1606                 spin_unlock(&vcpu->kvm->lock);
1607
1608                 switch (er) {
1609                 case EMULATE_DONE:
1610                         return 1;
1611                 case EMULATE_DO_MMIO:
1612                         ++vcpu->stat.mmio_exits;
1613                         kvm_run->exit_reason = KVM_EXIT_MMIO;
1614                         return 0;
1615                  case EMULATE_FAIL:
1616                         vcpu_printf(vcpu, "%s: emulate fail\n", __FUNCTION__);
1617                         break;
1618                 default:
1619                         BUG();
1620                 }
1621         }
1622
1623         if (vcpu->rmode.active &&
1624             handle_rmode_exception(vcpu, intr_info & INTR_INFO_VECTOR_MASK,
1625                                                                 error_code)) {
1626                 if (vcpu->halt_request) {
1627                         vcpu->halt_request = 0;
1628                         return kvm_emulate_halt(vcpu);
1629                 }
1630                 return 1;
1631         }
1632
1633         if ((intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK)) == (INTR_TYPE_EXCEPTION | 1)) {
1634                 kvm_run->exit_reason = KVM_EXIT_DEBUG;
1635                 return 0;
1636         }
1637         kvm_run->exit_reason = KVM_EXIT_EXCEPTION;
1638         kvm_run->ex.exception = intr_info & INTR_INFO_VECTOR_MASK;
1639         kvm_run->ex.error_code = error_code;
1640         return 0;
1641 }
1642
1643 static int handle_external_interrupt(struct kvm_vcpu *vcpu,
1644                                      struct kvm_run *kvm_run)
1645 {
1646         ++vcpu->stat.irq_exits;
1647         return 1;
1648 }
1649
1650 static int handle_triple_fault(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1651 {
1652         kvm_run->exit_reason = KVM_EXIT_SHUTDOWN;
1653         return 0;
1654 }
1655
1656 static int get_io_count(struct kvm_vcpu *vcpu, unsigned long *count)
1657 {
1658         u64 inst;
1659         gva_t rip;
1660         int countr_size;
1661         int i, n;
1662
1663         if ((vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_VM)) {
1664                 countr_size = 2;
1665         } else {
1666                 u32 cs_ar = vmcs_read32(GUEST_CS_AR_BYTES);
1667
1668                 countr_size = (cs_ar & AR_L_MASK) ? 8:
1669                               (cs_ar & AR_DB_MASK) ? 4: 2;
1670         }
1671
1672         rip =  vmcs_readl(GUEST_RIP);
1673         if (countr_size != 8)
1674                 rip += vmcs_readl(GUEST_CS_BASE);
1675
1676         n = kvm_read_guest(vcpu, rip, sizeof(inst), &inst);
1677
1678         for (i = 0; i < n; i++) {
1679                 switch (((u8*)&inst)[i]) {
1680                 case 0xf0:
1681                 case 0xf2:
1682                 case 0xf3:
1683                 case 0x2e:
1684                 case 0x36:
1685                 case 0x3e:
1686                 case 0x26:
1687                 case 0x64:
1688                 case 0x65:
1689                 case 0x66:
1690                         break;
1691                 case 0x67:
1692                         countr_size = (countr_size == 2) ? 4: (countr_size >> 1);
1693                 default:
1694                         goto done;
1695                 }
1696         }
1697         return 0;
1698 done:
1699         countr_size *= 8;
1700         *count = vcpu->regs[VCPU_REGS_RCX] & (~0ULL >> (64 - countr_size));
1701         //printk("cx: %lx\n", vcpu->regs[VCPU_REGS_RCX]);
1702         return 1;
1703 }
1704
1705 static int handle_io(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1706 {
1707         u64 exit_qualification;
1708         int size, down, in, string, rep;
1709         unsigned port;
1710         unsigned long count;
1711         gva_t address;
1712
1713         ++vcpu->stat.io_exits;
1714         exit_qualification = vmcs_read64(EXIT_QUALIFICATION);
1715         in = (exit_qualification & 8) != 0;
1716         size = (exit_qualification & 7) + 1;
1717         string = (exit_qualification & 16) != 0;
1718         down = (vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_DF) != 0;
1719         count = 1;
1720         rep = (exit_qualification & 32) != 0;
1721         port = exit_qualification >> 16;
1722         address = 0;
1723         if (string) {
1724                 if (rep && !get_io_count(vcpu, &count))
1725                         return 1;
1726                 address = vmcs_readl(GUEST_LINEAR_ADDRESS);
1727         }
1728         return kvm_setup_pio(vcpu, kvm_run, in, size, count, string, down,
1729                              address, rep, port);
1730 }
1731
1732 static void
1733 vmx_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
1734 {
1735         /*
1736          * Patch in the VMCALL instruction:
1737          */
1738         hypercall[0] = 0x0f;
1739         hypercall[1] = 0x01;
1740         hypercall[2] = 0xc1;
1741         hypercall[3] = 0xc3;
1742 }
1743
1744 static int handle_cr(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1745 {
1746         u64 exit_qualification;
1747         int cr;
1748         int reg;
1749
1750         exit_qualification = vmcs_read64(EXIT_QUALIFICATION);
1751         cr = exit_qualification & 15;
1752         reg = (exit_qualification >> 8) & 15;
1753         switch ((exit_qualification >> 4) & 3) {
1754         case 0: /* mov to cr */
1755                 switch (cr) {
1756                 case 0:
1757                         vcpu_load_rsp_rip(vcpu);
1758                         set_cr0(vcpu, vcpu->regs[reg]);
1759                         skip_emulated_instruction(vcpu);
1760                         return 1;
1761                 case 3:
1762                         vcpu_load_rsp_rip(vcpu);
1763                         set_cr3(vcpu, vcpu->regs[reg]);
1764                         skip_emulated_instruction(vcpu);
1765                         return 1;
1766                 case 4:
1767                         vcpu_load_rsp_rip(vcpu);
1768                         set_cr4(vcpu, vcpu->regs[reg]);
1769                         skip_emulated_instruction(vcpu);
1770                         return 1;
1771                 case 8:
1772                         vcpu_load_rsp_rip(vcpu);
1773                         set_cr8(vcpu, vcpu->regs[reg]);
1774                         skip_emulated_instruction(vcpu);
1775                         return 1;
1776                 };
1777                 break;
1778         case 2: /* clts */
1779                 vcpu_load_rsp_rip(vcpu);
1780                 vmx_fpu_deactivate(vcpu);
1781                 vcpu->cr0 &= ~CR0_TS_MASK;
1782                 vmcs_writel(CR0_READ_SHADOW, vcpu->cr0);
1783                 vmx_fpu_activate(vcpu);
1784                 skip_emulated_instruction(vcpu);
1785                 return 1;
1786         case 1: /*mov from cr*/
1787                 switch (cr) {
1788                 case 3:
1789                         vcpu_load_rsp_rip(vcpu);
1790                         vcpu->regs[reg] = vcpu->cr3;
1791                         vcpu_put_rsp_rip(vcpu);
1792                         skip_emulated_instruction(vcpu);
1793                         return 1;
1794                 case 8:
1795                         vcpu_load_rsp_rip(vcpu);
1796                         vcpu->regs[reg] = vcpu->cr8;
1797                         vcpu_put_rsp_rip(vcpu);
1798                         skip_emulated_instruction(vcpu);
1799                         return 1;
1800                 }
1801                 break;
1802         case 3: /* lmsw */
1803                 lmsw(vcpu, (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f);
1804
1805                 skip_emulated_instruction(vcpu);
1806                 return 1;
1807         default:
1808                 break;
1809         }
1810         kvm_run->exit_reason = 0;
1811         printk(KERN_ERR "kvm: unhandled control register: op %d cr %d\n",
1812                (int)(exit_qualification >> 4) & 3, cr);
1813         return 0;
1814 }
1815
1816 static int handle_dr(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1817 {
1818         u64 exit_qualification;
1819         unsigned long val;
1820         int dr, reg;
1821
1822         /*
1823          * FIXME: this code assumes the host is debugging the guest.
1824          *        need to deal with guest debugging itself too.
1825          */
1826         exit_qualification = vmcs_read64(EXIT_QUALIFICATION);
1827         dr = exit_qualification & 7;
1828         reg = (exit_qualification >> 8) & 15;
1829         vcpu_load_rsp_rip(vcpu);
1830         if (exit_qualification & 16) {
1831                 /* mov from dr */
1832                 switch (dr) {
1833                 case 6:
1834                         val = 0xffff0ff0;
1835                         break;
1836                 case 7:
1837                         val = 0x400;
1838                         break;
1839                 default:
1840                         val = 0;
1841                 }
1842                 vcpu->regs[reg] = val;
1843         } else {
1844                 /* mov to dr */
1845         }
1846         vcpu_put_rsp_rip(vcpu);
1847         skip_emulated_instruction(vcpu);
1848         return 1;
1849 }
1850
1851 static int handle_cpuid(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1852 {
1853         kvm_emulate_cpuid(vcpu);
1854         return 1;
1855 }
1856
1857 static int handle_rdmsr(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1858 {
1859         u32 ecx = vcpu->regs[VCPU_REGS_RCX];
1860         u64 data;
1861
1862         if (vmx_get_msr(vcpu, ecx, &data)) {
1863                 vmx_inject_gp(vcpu, 0);
1864                 return 1;
1865         }
1866
1867         /* FIXME: handling of bits 32:63 of rax, rdx */
1868         vcpu->regs[VCPU_REGS_RAX] = data & -1u;
1869         vcpu->regs[VCPU_REGS_RDX] = (data >> 32) & -1u;
1870         skip_emulated_instruction(vcpu);
1871         return 1;
1872 }
1873
1874 static int handle_wrmsr(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1875 {
1876         u32 ecx = vcpu->regs[VCPU_REGS_RCX];
1877         u64 data = (vcpu->regs[VCPU_REGS_RAX] & -1u)
1878                 | ((u64)(vcpu->regs[VCPU_REGS_RDX] & -1u) << 32);
1879
1880         if (vmx_set_msr(vcpu, ecx, data) != 0) {
1881                 vmx_inject_gp(vcpu, 0);
1882                 return 1;
1883         }
1884
1885         skip_emulated_instruction(vcpu);
1886         return 1;
1887 }
1888
1889 static void post_kvm_run_save(struct kvm_vcpu *vcpu,
1890                               struct kvm_run *kvm_run)
1891 {
1892         kvm_run->if_flag = (vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_IF) != 0;
1893         kvm_run->cr8 = vcpu->cr8;
1894         kvm_run->apic_base = vcpu->apic_base;
1895         kvm_run->ready_for_interrupt_injection = (vcpu->interrupt_window_open &&
1896                                                   vcpu->irq_summary == 0);
1897 }
1898
1899 static int handle_interrupt_window(struct kvm_vcpu *vcpu,
1900                                    struct kvm_run *kvm_run)
1901 {
1902         /*
1903          * If the user space waits to inject interrupts, exit as soon as
1904          * possible
1905          */
1906         if (kvm_run->request_interrupt_window &&
1907             !vcpu->irq_summary) {
1908                 kvm_run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
1909                 ++vcpu->stat.irq_window_exits;
1910                 return 0;
1911         }
1912         return 1;
1913 }
1914
1915 static int handle_halt(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1916 {
1917         skip_emulated_instruction(vcpu);
1918         return kvm_emulate_halt(vcpu);
1919 }
1920
1921 static int handle_vmcall(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1922 {
1923         skip_emulated_instruction(vcpu);
1924         return kvm_hypercall(vcpu, kvm_run);
1925 }
1926
1927 /*
1928  * The exit handlers return 1 if the exit was handled fully and guest execution
1929  * may resume.  Otherwise they set the kvm_run parameter to indicate what needs
1930  * to be done to userspace and return 0.
1931  */
1932 static int (*kvm_vmx_exit_handlers[])(struct kvm_vcpu *vcpu,
1933                                       struct kvm_run *kvm_run) = {
1934         [EXIT_REASON_EXCEPTION_NMI]           = handle_exception,
1935         [EXIT_REASON_EXTERNAL_INTERRUPT]      = handle_external_interrupt,
1936         [EXIT_REASON_TRIPLE_FAULT]            = handle_triple_fault,
1937         [EXIT_REASON_IO_INSTRUCTION]          = handle_io,
1938         [EXIT_REASON_CR_ACCESS]               = handle_cr,
1939         [EXIT_REASON_DR_ACCESS]               = handle_dr,
1940         [EXIT_REASON_CPUID]                   = handle_cpuid,
1941         [EXIT_REASON_MSR_READ]                = handle_rdmsr,
1942         [EXIT_REASON_MSR_WRITE]               = handle_wrmsr,
1943         [EXIT_REASON_PENDING_INTERRUPT]       = handle_interrupt_window,
1944         [EXIT_REASON_HLT]                     = handle_halt,
1945         [EXIT_REASON_VMCALL]                  = handle_vmcall,
1946 };
1947
1948 static const int kvm_vmx_max_exit_handlers =
1949         ARRAY_SIZE(kvm_vmx_exit_handlers);
1950
1951 /*
1952  * The guest has exited.  See if we can fix it or if we need userspace
1953  * assistance.
1954  */
1955 static int kvm_handle_exit(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
1956 {
1957         u32 vectoring_info = vmcs_read32(IDT_VECTORING_INFO_FIELD);
1958         u32 exit_reason = vmcs_read32(VM_EXIT_REASON);
1959
1960         if ( (vectoring_info & VECTORING_INFO_VALID_MASK) &&
1961                                 exit_reason != EXIT_REASON_EXCEPTION_NMI )
1962                 printk(KERN_WARNING "%s: unexpected, valid vectoring info and "
1963                        "exit reason is 0x%x\n", __FUNCTION__, exit_reason);
1964         if (exit_reason < kvm_vmx_max_exit_handlers
1965             && kvm_vmx_exit_handlers[exit_reason])
1966                 return kvm_vmx_exit_handlers[exit_reason](vcpu, kvm_run);
1967         else {
1968                 kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
1969                 kvm_run->hw.hardware_exit_reason = exit_reason;
1970         }
1971         return 0;
1972 }
1973
1974 /*
1975  * Check if userspace requested an interrupt window, and that the
1976  * interrupt window is open.
1977  *
1978  * No need to exit to userspace if we already have an interrupt queued.
1979  */
1980 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu,
1981                                           struct kvm_run *kvm_run)
1982 {
1983         return (!vcpu->irq_summary &&
1984                 kvm_run->request_interrupt_window &&
1985                 vcpu->interrupt_window_open &&
1986                 (vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_IF));
1987 }
1988
1989 static void vmx_flush_tlb(struct kvm_vcpu *vcpu)
1990 {
1991 }
1992
1993 static int vmx_vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1994 {
1995         u8 fail;
1996         int r;
1997
1998 preempted:
1999         if (vcpu->guest_debug.enabled)
2000                 kvm_guest_debug_pre(vcpu);
2001
2002 again:
2003         if (!vcpu->mmio_read_completed)
2004                 do_interrupt_requests(vcpu, kvm_run);
2005
2006         vmx_save_host_state(vcpu);
2007         kvm_load_guest_fpu(vcpu);
2008
2009         r = kvm_mmu_reload(vcpu);
2010         if (unlikely(r))
2011                 goto out;
2012
2013         /*
2014          * Loading guest fpu may have cleared host cr0.ts
2015          */
2016         vmcs_writel(HOST_CR0, read_cr0());
2017
2018         local_irq_disable();
2019
2020         vcpu->guest_mode = 1;
2021         if (vcpu->requests)
2022                 if (test_and_clear_bit(KVM_TLB_FLUSH, &vcpu->requests))
2023                     vmx_flush_tlb(vcpu);
2024
2025         asm (
2026                 /* Store host registers */
2027 #ifdef CONFIG_X86_64
2028                 "push %%rax; push %%rbx; push %%rdx;"
2029                 "push %%rsi; push %%rdi; push %%rbp;"
2030                 "push %%r8;  push %%r9;  push %%r10; push %%r11;"
2031                 "push %%r12; push %%r13; push %%r14; push %%r15;"
2032                 "push %%rcx \n\t"
2033                 ASM_VMX_VMWRITE_RSP_RDX "\n\t"
2034 #else
2035                 "pusha; push %%ecx \n\t"
2036                 ASM_VMX_VMWRITE_RSP_RDX "\n\t"
2037 #endif
2038                 /* Check if vmlaunch of vmresume is needed */
2039                 "cmp $0, %1 \n\t"
2040                 /* Load guest registers.  Don't clobber flags. */
2041 #ifdef CONFIG_X86_64
2042                 "mov %c[cr2](%3), %%rax \n\t"
2043                 "mov %%rax, %%cr2 \n\t"
2044                 "mov %c[rax](%3), %%rax \n\t"
2045                 "mov %c[rbx](%3), %%rbx \n\t"
2046                 "mov %c[rdx](%3), %%rdx \n\t"
2047                 "mov %c[rsi](%3), %%rsi \n\t"
2048                 "mov %c[rdi](%3), %%rdi \n\t"
2049                 "mov %c[rbp](%3), %%rbp \n\t"
2050                 "mov %c[r8](%3),  %%r8  \n\t"
2051                 "mov %c[r9](%3),  %%r9  \n\t"
2052                 "mov %c[r10](%3), %%r10 \n\t"
2053                 "mov %c[r11](%3), %%r11 \n\t"
2054                 "mov %c[r12](%3), %%r12 \n\t"
2055                 "mov %c[r13](%3), %%r13 \n\t"
2056                 "mov %c[r14](%3), %%r14 \n\t"
2057                 "mov %c[r15](%3), %%r15 \n\t"
2058                 "mov %c[rcx](%3), %%rcx \n\t" /* kills %3 (rcx) */
2059 #else
2060                 "mov %c[cr2](%3), %%eax \n\t"
2061                 "mov %%eax,   %%cr2 \n\t"
2062                 "mov %c[rax](%3), %%eax \n\t"
2063                 "mov %c[rbx](%3), %%ebx \n\t"
2064                 "mov %c[rdx](%3), %%edx \n\t"
2065                 "mov %c[rsi](%3), %%esi \n\t"
2066                 "mov %c[rdi](%3), %%edi \n\t"
2067                 "mov %c[rbp](%3), %%ebp \n\t"
2068                 "mov %c[rcx](%3), %%ecx \n\t" /* kills %3 (ecx) */
2069 #endif
2070                 /* Enter guest mode */
2071                 "jne .Llaunched \n\t"
2072                 ASM_VMX_VMLAUNCH "\n\t"
2073                 "jmp .Lkvm_vmx_return \n\t"
2074                 ".Llaunched: " ASM_VMX_VMRESUME "\n\t"
2075                 ".Lkvm_vmx_return: "
2076                 /* Save guest registers, load host registers, keep flags */
2077 #ifdef CONFIG_X86_64
2078                 "xchg %3,     (%%rsp) \n\t"
2079                 "mov %%rax, %c[rax](%3) \n\t"
2080                 "mov %%rbx, %c[rbx](%3) \n\t"
2081                 "pushq (%%rsp); popq %c[rcx](%3) \n\t"
2082                 "mov %%rdx, %c[rdx](%3) \n\t"
2083                 "mov %%rsi, %c[rsi](%3) \n\t"
2084                 "mov %%rdi, %c[rdi](%3) \n\t"
2085                 "mov %%rbp, %c[rbp](%3) \n\t"
2086                 "mov %%r8,  %c[r8](%3) \n\t"
2087                 "mov %%r9,  %c[r9](%3) \n\t"
2088                 "mov %%r10, %c[r10](%3) \n\t"
2089                 "mov %%r11, %c[r11](%3) \n\t"
2090                 "mov %%r12, %c[r12](%3) \n\t"
2091                 "mov %%r13, %c[r13](%3) \n\t"
2092                 "mov %%r14, %c[r14](%3) \n\t"
2093                 "mov %%r15, %c[r15](%3) \n\t"
2094                 "mov %%cr2, %%rax   \n\t"
2095                 "mov %%rax, %c[cr2](%3) \n\t"
2096                 "mov (%%rsp), %3 \n\t"
2097
2098                 "pop  %%rcx; pop  %%r15; pop  %%r14; pop  %%r13; pop  %%r12;"
2099                 "pop  %%r11; pop  %%r10; pop  %%r9;  pop  %%r8;"
2100                 "pop  %%rbp; pop  %%rdi; pop  %%rsi;"
2101                 "pop  %%rdx; pop  %%rbx; pop  %%rax \n\t"
2102 #else
2103                 "xchg %3, (%%esp) \n\t"
2104                 "mov %%eax, %c[rax](%3) \n\t"
2105                 "mov %%ebx, %c[rbx](%3) \n\t"
2106                 "pushl (%%esp); popl %c[rcx](%3) \n\t"
2107                 "mov %%edx, %c[rdx](%3) \n\t"
2108                 "mov %%esi, %c[rsi](%3) \n\t"
2109                 "mov %%edi, %c[rdi](%3) \n\t"
2110                 "mov %%ebp, %c[rbp](%3) \n\t"
2111                 "mov %%cr2, %%eax  \n\t"
2112                 "mov %%eax, %c[cr2](%3) \n\t"
2113                 "mov (%%esp), %3 \n\t"
2114
2115                 "pop %%ecx; popa \n\t"
2116 #endif
2117                 "setbe %0 \n\t"
2118               : "=q" (fail)
2119               : "r"(vcpu->launched), "d"((unsigned long)HOST_RSP),
2120                 "c"(vcpu),
2121                 [rax]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_RAX])),
2122                 [rbx]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_RBX])),
2123                 [rcx]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_RCX])),
2124                 [rdx]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_RDX])),
2125                 [rsi]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_RSI])),
2126                 [rdi]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_RDI])),
2127                 [rbp]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_RBP])),
2128 #ifdef CONFIG_X86_64
2129                 [r8 ]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_R8 ])),
2130                 [r9 ]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_R9 ])),
2131                 [r10]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_R10])),
2132                 [r11]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_R11])),
2133                 [r12]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_R12])),
2134                 [r13]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_R13])),
2135                 [r14]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_R14])),
2136                 [r15]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_R15])),
2137 #endif
2138                 [cr2]"i"(offsetof(struct kvm_vcpu, cr2))
2139               : "cc", "memory" );
2140
2141         vcpu->guest_mode = 0;
2142         local_irq_enable();
2143
2144         ++vcpu->stat.exits;
2145
2146         vcpu->interrupt_window_open = (vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & 3) == 0;
2147
2148         asm ("mov %0, %%ds; mov %0, %%es" : : "r"(__USER_DS));
2149
2150         if (unlikely(fail)) {
2151                 kvm_run->exit_reason = KVM_EXIT_FAIL_ENTRY;
2152                 kvm_run->fail_entry.hardware_entry_failure_reason
2153                         = vmcs_read32(VM_INSTRUCTION_ERROR);
2154                 r = 0;
2155                 goto out;
2156         }
2157         /*
2158          * Profile KVM exit RIPs:
2159          */
2160         if (unlikely(prof_on == KVM_PROFILING))
2161                 profile_hit(KVM_PROFILING, (void *)vmcs_readl(GUEST_RIP));
2162
2163         vcpu->launched = 1;
2164         r = kvm_handle_exit(kvm_run, vcpu);
2165         if (r > 0) {
2166                 /* Give scheduler a change to reschedule. */
2167                 if (signal_pending(current)) {
2168                         r = -EINTR;
2169                         kvm_run->exit_reason = KVM_EXIT_INTR;
2170                         ++vcpu->stat.signal_exits;
2171                         goto out;
2172                 }
2173
2174                 if (dm_request_for_irq_injection(vcpu, kvm_run)) {
2175                         r = -EINTR;
2176                         kvm_run->exit_reason = KVM_EXIT_INTR;
2177                         ++vcpu->stat.request_irq_exits;
2178                         goto out;
2179                 }
2180                 if (!need_resched()) {
2181                         ++vcpu->stat.light_exits;
2182                         goto again;
2183                 }
2184         }
2185
2186 out:
2187         if (r > 0) {
2188                 kvm_resched(vcpu);
2189                 goto preempted;
2190         }
2191
2192         post_kvm_run_save(vcpu, kvm_run);
2193         return r;
2194 }
2195
2196 static void vmx_inject_page_fault(struct kvm_vcpu *vcpu,
2197                                   unsigned long addr,
2198                                   u32 err_code)
2199 {
2200         u32 vect_info = vmcs_read32(IDT_VECTORING_INFO_FIELD);
2201
2202         ++vcpu->stat.pf_guest;
2203
2204         if (is_page_fault(vect_info)) {
2205                 printk(KERN_DEBUG "inject_page_fault: "
2206                        "double fault 0x%lx @ 0x%lx\n",
2207                        addr, vmcs_readl(GUEST_RIP));
2208                 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE, 0);
2209                 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
2210                              DF_VECTOR |
2211                              INTR_TYPE_EXCEPTION |
2212                              INTR_INFO_DELIEVER_CODE_MASK |
2213                              INTR_INFO_VALID_MASK);
2214                 return;
2215         }
2216         vcpu->cr2 = addr;
2217         vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE, err_code);
2218         vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
2219                      PF_VECTOR |
2220                      INTR_TYPE_EXCEPTION |
2221                      INTR_INFO_DELIEVER_CODE_MASK |
2222                      INTR_INFO_VALID_MASK);
2223
2224 }
2225
2226 static void vmx_free_vmcs(struct kvm_vcpu *vcpu)
2227 {
2228         if (vcpu->vmcs) {
2229                 on_each_cpu(__vcpu_clear, vcpu, 0, 1);
2230                 free_vmcs(vcpu->vmcs);
2231                 vcpu->vmcs = NULL;
2232         }
2233 }
2234
2235 static void vmx_free_vcpu(struct kvm_vcpu *vcpu)
2236 {
2237         vmx_free_vmcs(vcpu);
2238 }
2239
2240 static int vmx_create_vcpu(struct kvm_vcpu *vcpu)
2241 {
2242         struct vmcs *vmcs;
2243
2244         vcpu->guest_msrs = kmalloc(PAGE_SIZE, GFP_KERNEL);
2245         if (!vcpu->guest_msrs)
2246                 return -ENOMEM;
2247
2248         vcpu->host_msrs = kmalloc(PAGE_SIZE, GFP_KERNEL);
2249         if (!vcpu->host_msrs)
2250                 goto out_free_guest_msrs;
2251
2252         vmcs = alloc_vmcs();
2253         if (!vmcs)
2254                 goto out_free_msrs;
2255
2256         vmcs_clear(vmcs);
2257         vcpu->vmcs = vmcs;
2258         vcpu->launched = 0;
2259
2260         return 0;
2261
2262 out_free_msrs:
2263         kfree(vcpu->host_msrs);
2264         vcpu->host_msrs = NULL;
2265
2266 out_free_guest_msrs:
2267         kfree(vcpu->guest_msrs);
2268         vcpu->guest_msrs = NULL;
2269
2270         return -ENOMEM;
2271 }
2272
2273 static struct kvm_arch_ops vmx_arch_ops = {
2274         .cpu_has_kvm_support = cpu_has_kvm_support,
2275         .disabled_by_bios = vmx_disabled_by_bios,
2276         .hardware_setup = hardware_setup,
2277         .hardware_unsetup = hardware_unsetup,
2278         .hardware_enable = hardware_enable,
2279         .hardware_disable = hardware_disable,
2280
2281         .vcpu_create = vmx_create_vcpu,
2282         .vcpu_free = vmx_free_vcpu,
2283
2284         .vcpu_load = vmx_vcpu_load,
2285         .vcpu_put = vmx_vcpu_put,
2286         .vcpu_decache = vmx_vcpu_decache,
2287
2288         .set_guest_debug = set_guest_debug,
2289         .get_msr = vmx_get_msr,
2290         .set_msr = vmx_set_msr,
2291         .get_segment_base = vmx_get_segment_base,
2292         .get_segment = vmx_get_segment,
2293         .set_segment = vmx_set_segment,
2294         .get_cs_db_l_bits = vmx_get_cs_db_l_bits,
2295         .decache_cr4_guest_bits = vmx_decache_cr4_guest_bits,
2296         .set_cr0 = vmx_set_cr0,
2297         .set_cr3 = vmx_set_cr3,
2298         .set_cr4 = vmx_set_cr4,
2299 #ifdef CONFIG_X86_64
2300         .set_efer = vmx_set_efer,
2301 #endif
2302         .get_idt = vmx_get_idt,
2303         .set_idt = vmx_set_idt,
2304         .get_gdt = vmx_get_gdt,
2305         .set_gdt = vmx_set_gdt,
2306         .cache_regs = vcpu_load_rsp_rip,
2307         .decache_regs = vcpu_put_rsp_rip,
2308         .get_rflags = vmx_get_rflags,
2309         .set_rflags = vmx_set_rflags,
2310
2311         .tlb_flush = vmx_flush_tlb,
2312         .inject_page_fault = vmx_inject_page_fault,
2313
2314         .inject_gp = vmx_inject_gp,
2315
2316         .run = vmx_vcpu_run,
2317         .skip_emulated_instruction = skip_emulated_instruction,
2318         .vcpu_setup = vmx_vcpu_setup,
2319         .patch_hypercall = vmx_patch_hypercall,
2320 };
2321
2322 static int __init vmx_init(void)
2323 {
2324         void *iova;
2325         int r;
2326
2327         vmx_io_bitmap_a = alloc_page(GFP_KERNEL | __GFP_HIGHMEM);
2328         if (!vmx_io_bitmap_a)
2329                 return -ENOMEM;
2330
2331         vmx_io_bitmap_b = alloc_page(GFP_KERNEL | __GFP_HIGHMEM);
2332         if (!vmx_io_bitmap_b) {
2333                 r = -ENOMEM;
2334                 goto out;
2335         }
2336
2337         /*
2338          * Allow direct access to the PC debug port (it is often used for I/O
2339          * delays, but the vmexits simply slow things down).
2340          */
2341         iova = kmap(vmx_io_bitmap_a);
2342         memset(iova, 0xff, PAGE_SIZE);
2343         clear_bit(0x80, iova);
2344         kunmap(vmx_io_bitmap_a);
2345
2346         iova = kmap(vmx_io_bitmap_b);
2347         memset(iova, 0xff, PAGE_SIZE);
2348         kunmap(vmx_io_bitmap_b);
2349
2350         r = kvm_init_arch(&vmx_arch_ops, THIS_MODULE);
2351         if (r)
2352                 goto out1;
2353
2354         return 0;
2355
2356 out1:
2357         __free_page(vmx_io_bitmap_b);
2358 out:
2359         __free_page(vmx_io_bitmap_a);
2360         return r;
2361 }
2362
2363 static void __exit vmx_exit(void)
2364 {
2365         __free_page(vmx_io_bitmap_b);
2366         __free_page(vmx_io_bitmap_a);
2367
2368         kvm_exit_arch();
2369 }
2370
2371 module_init(vmx_init)
2372 module_exit(vmx_exit)