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 "kvm_vmx.h"
  21 #include <linux/module.h>
  22 #include <linux/mm.h>
  23 #include <linux/highmem.h>
  24 #include <linux/profile.h>
  25 #include <asm/io.h>
  26 #include <asm/desc.h>
  27
  28 #include "segment_descriptor.h"
  29
  30
  31 MODULE_AUTHOR("Qumranet");
  32 MODULE_LICENSE("GPL");
  33
  34 static DEFINE_PER_CPU(struct vmcs *, vmxarea);
  35 static DEFINE_PER_CPU(struct vmcs *, current_vmcs);
  36
  37 #ifdef CONFIG_X86_64
  38 #define HOST_IS_64 1
  39 #else
  40 #define HOST_IS_64 0
  41 #endif
  42
  43 static struct vmcs_descriptor {
  44         int size;
  45         int order;
  46         u32 revision_id;
  47 } vmcs_descriptor;
  48
  49 #define VMX_SEGMENT_FIELD(seg)                                  \
  50         [VCPU_SREG_##seg] = {                                   \
  51                 .selector = GUEST_##seg##_SELECTOR,             \
  52                 .base = GUEST_##seg##_BASE,                     \
  53                 .limit = GUEST_##seg##_LIMIT,                   \
  54                 .ar_bytes = GUEST_##seg##_AR_BYTES,             \
  55         }
  56
  57 static struct kvm_vmx_segment_field {
  58         unsigned selector;
  59         unsigned base;
  60         unsigned limit;
  61         unsigned ar_bytes;
  62 } kvm_vmx_segment_fields[] = {
  63         VMX_SEGMENT_FIELD(CS),
  64         VMX_SEGMENT_FIELD(DS),
  65         VMX_SEGMENT_FIELD(ES),
  66         VMX_SEGMENT_FIELD(FS),
  67         VMX_SEGMENT_FIELD(GS),
  68         VMX_SEGMENT_FIELD(SS),
  69         VMX_SEGMENT_FIELD(TR),
  70         VMX_SEGMENT_FIELD(LDTR),
  71 };
  72
  73 static const u32 vmx_msr_index[] = {
  74 #ifdef CONFIG_X86_64
  75         MSR_SYSCALL_MASK, MSR_LSTAR, MSR_CSTAR, MSR_KERNEL_GS_BASE,
  76 #endif
  77         MSR_EFER, MSR_K6_STAR,
  78 };
  79 #define NR_VMX_MSR (sizeof(vmx_msr_index) / sizeof(*vmx_msr_index))
  80
  81 static inline int is_page_fault(u32 intr_info)
  82 {
  83         return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
  84                              INTR_INFO_VALID_MASK)) ==
  85                 (INTR_TYPE_EXCEPTION | PF_VECTOR | INTR_INFO_VALID_MASK);
  86 }
  87
  88 static inline int is_external_interrupt(u32 intr_info)
  89 {
  90         return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VALID_MASK))
  91                 == (INTR_TYPE_EXT_INTR | INTR_INFO_VALID_MASK);
  92 }
  93
  94 static struct vmx_msr_entry *find_msr_entry(struct kvm_vcpu *vcpu, u32 msr)
  95 {
  96         int i;
  97
  98         for (i = 0; i < vcpu->nmsrs; ++i)
  99                 if (vcpu->guest_msrs[i].index == msr)
 100                         return &vcpu->guest_msrs[i];
 101         return NULL;
 102 }
 103
 104 static void vmcs_clear(struct vmcs *vmcs)
 105 {
 106         u64 phys_addr = __pa(vmcs);
 107         u8 error;
 108
 109         asm volatile (ASM_VMX_VMCLEAR_RAX "; setna %0"
 110                       : "=g"(error) : "a"(&phys_addr), "m"(phys_addr)
 111                       : "cc", "memory");
 112         if (error)
 113                 printk(KERN_ERR "kvm: vmclear fail: %p/%llx\n",
 114                        vmcs, phys_addr);
 115 }
 116
 117 static void __vcpu_clear(void *arg)
 118 {
 119         struct kvm_vcpu *vcpu = arg;
 120         int cpu = raw_smp_processor_id();
 121
 122         if (vcpu->cpu == cpu)
 123                 vmcs_clear(vcpu->vmcs);
 124         if (per_cpu(current_vmcs, cpu) == vcpu->vmcs)
 125                 per_cpu(current_vmcs, cpu) = NULL;
 126 }
 127
 128 static void vcpu_clear(struct kvm_vcpu *vcpu)
 129 {
 130         if (vcpu->cpu != raw_smp_processor_id() && vcpu->cpu != -1)
 131                 smp_call_function_single(vcpu->cpu, __vcpu_clear, vcpu, 0, 1);
 132         else
 133                 __vcpu_clear(vcpu);
 134         vcpu->launched = 0;
 135 }
 136
 137 static unsigned long vmcs_readl(unsigned long field)
 138 {
 139         unsigned long value;
 140
 141         asm volatile (ASM_VMX_VMREAD_RDX_RAX
 142                       : "=a"(value) : "d"(field) : "cc");
 143         return value;
 144 }
 145
 146 static u16 vmcs_read16(unsigned long field)
 147 {
 148         return vmcs_readl(field);
 149 }
 150
 151 static u32 vmcs_read32(unsigned long field)
 152 {
 153         return vmcs_readl(field);
 154 }
 155
 156 static u64 vmcs_read64(unsigned long field)
 157 {
 158 #ifdef CONFIG_X86_64
 159         return vmcs_readl(field);
 160 #else
 161         return vmcs_readl(field) | ((u64)vmcs_readl(field+1) << 32);
 162 #endif
 163 }
 164
 165 static noinline void vmwrite_error(unsigned long field, unsigned long value)
 166 {
 167         printk(KERN_ERR "vmwrite error: reg %lx value %lx (err %d)\n",
 168                field, value, vmcs_read32(VM_INSTRUCTION_ERROR));
 169         dump_stack();
 170 }
 171
 172 static void vmcs_writel(unsigned long field, unsigned long value)
 173 {
 174         u8 error;
 175
 176         asm volatile (ASM_VMX_VMWRITE_RAX_RDX "; setna %0"
 177                        : "=q"(error) : "a"(value), "d"(field) : "cc" );
 178         if (unlikely(error))
 179                 vmwrite_error(field, value);
 180 }
 181
 182 static void vmcs_write16(unsigned long field, u16 value)
 183 {
 184         vmcs_writel(field, value);
 185 }
 186
 187 static void vmcs_write32(unsigned long field, u32 value)
 188 {
 189         vmcs_writel(field, value);
 190 }
 191
 192 static void vmcs_write64(unsigned long field, u64 value)
 193 {
 194 #ifdef CONFIG_X86_64
 195         vmcs_writel(field, value);
 196 #else
 197         vmcs_writel(field, value);
 198         asm volatile ("");
 199         vmcs_writel(field+1, value >> 32);
 200 #endif
 201 }
 202
 203 /*
 204  * Switches to specified vcpu, until a matching vcpu_put(), but assumes
 205  * vcpu mutex is already taken.
 206  */
 207 static struct kvm_vcpu *vmx_vcpu_load(struct kvm_vcpu *vcpu)
 208 {
 209         u64 phys_addr = __pa(vcpu->vmcs);
 210         int cpu;
 211
 212         cpu = get_cpu();
 213
 214         if (vcpu->cpu != cpu)
 215                 vcpu_clear(vcpu);
 216
 217         if (per_cpu(current_vmcs, cpu) != vcpu->vmcs) {
 218                 u8 error;
 219
 220                 per_cpu(current_vmcs, cpu) = vcpu->vmcs;
 221                 asm volatile (ASM_VMX_VMPTRLD_RAX "; setna %0"
 222                               : "=g"(error) : "a"(&phys_addr), "m"(phys_addr)
 223                               : "cc");
 224                 if (error)
 225                         printk(KERN_ERR "kvm: vmptrld %p/%llx fail\n",
 226                                vcpu->vmcs, phys_addr);
 227         }
 228
 229         if (vcpu->cpu != cpu) {
 230                 struct descriptor_table dt;
 231                 unsigned long sysenter_esp;
 232
 233                 vcpu->cpu = cpu;
 234                 /*
 235                  * Linux uses per-cpu TSS and GDT, so set these when switching
 236                  * processors.
 237                  */
 238                 vmcs_writel(HOST_TR_BASE, read_tr_base()); /* 22.2.4 */
 239                 get_gdt(&dt);
 240                 vmcs_writel(HOST_GDTR_BASE, dt.base);   /* 22.2.4 */
 241
 242                 rdmsrl(MSR_IA32_SYSENTER_ESP, sysenter_esp);
 243                 vmcs_writel(HOST_IA32_SYSENTER_ESP, sysenter_esp); /* 22.2.3 */
 244         }
 245         return vcpu;
 246 }
 247
 248 static void vmx_vcpu_put(struct kvm_vcpu *vcpu)
 249 {
 250         put_cpu();
 251 }
 252
 253 static void vmx_vcpu_decache(struct kvm_vcpu *vcpu)
 254 {
 255         vcpu_clear(vcpu);
 256 }
 257
 258 static unsigned long vmx_get_rflags(struct kvm_vcpu *vcpu)
 259 {
 260         return vmcs_readl(GUEST_RFLAGS);
 261 }
 262
 263 static void vmx_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
 264 {
 265         vmcs_writel(GUEST_RFLAGS, rflags);
 266 }
 267
 268 static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
 269 {
 270         unsigned long rip;
 271         u32 interruptibility;
 272
 273         rip = vmcs_readl(GUEST_RIP);
 274         rip += vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
 275         vmcs_writel(GUEST_RIP, rip);
 276
 277         /*
 278          * We emulated an instruction, so temporary interrupt blocking
 279          * should be removed, if set.
 280          */
 281         interruptibility = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
 282         if (interruptibility & 3)
 283                 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO,
 284                              interruptibility & ~3);
 285         vcpu->interrupt_window_open = 1;
 286 }
 287
 288 static void vmx_inject_gp(struct kvm_vcpu *vcpu, unsigned error_code)
 289 {
 290         printk(KERN_DEBUG "inject_general_protection: rip 0x%lx\n",
 291                vmcs_readl(GUEST_RIP));
 292         vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE, error_code);
 293         vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
 294                      GP_VECTOR |
 295                      INTR_TYPE_EXCEPTION |
 296                      INTR_INFO_DELIEVER_CODE_MASK |
 297                      INTR_INFO_VALID_MASK);
 298 }
 299
 300 /*
 301  * reads and returns guest's timestamp counter "register"
 302  * guest_tsc = host_tsc + tsc_offset    -- 21.3
 303  */
 304 static u64 guest_read_tsc(void)
 305 {
 306         u64 host_tsc, tsc_offset;
 307
 308         rdtscll(host_tsc);
 309         tsc_offset = vmcs_read64(TSC_OFFSET);
 310         return host_tsc + tsc_offset;
 311 }
 312
 313 /*
 314  * writes 'guest_tsc' into guest's timestamp counter "register"
 315  * guest_tsc = host_tsc + tsc_offset ==> tsc_offset = guest_tsc - host_tsc
 316  */
 317 static void guest_write_tsc(u64 guest_tsc)
 318 {
 319         u64 host_tsc;
 320
 321         rdtscll(host_tsc);
 322         vmcs_write64(TSC_OFFSET, guest_tsc - host_tsc);
 323 }
 324
 325 static void reload_tss(void)
 326 {
 327 #ifndef CONFIG_X86_64
 328
 329         /*
 330          * VT restores TR but not its size.  Useless.
 331          */
 332         struct descriptor_table gdt;
 333         struct segment_descriptor *descs;
 334
 335         get_gdt(&gdt);
 336         descs = (void *)gdt.base;
 337         descs[GDT_ENTRY_TSS].type = 9; /* available TSS */
 338         load_TR_desc();
 339 #endif
 340 }
 341
 342 /*
 343  * Reads an msr value (of 'msr_index') into 'pdata'.
 344  * Returns 0 on success, non-0 otherwise.
 345  * Assumes vcpu_load() was already called.
 346  */
 347 static int vmx_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
 348 {
 349         u64 data;
 350         struct vmx_msr_entry *msr;
 351
 352         if (!pdata) {
 353                 printk(KERN_ERR "BUG: get_msr called with NULL pdata\n");
 354                 return -EINVAL;
 355         }
 356
 357         switch (msr_index) {
 358 #ifdef CONFIG_X86_64
 359         case MSR_FS_BASE:
 360                 data = vmcs_readl(GUEST_FS_BASE);
 361                 break;
 362         case MSR_GS_BASE:
 363                 data = vmcs_readl(GUEST_GS_BASE);
 364                 break;
 365         case MSR_EFER:
 366                 return kvm_get_msr_common(vcpu, msr_index, pdata);
 367 #endif
 368         case MSR_IA32_TIME_STAMP_COUNTER:
 369                 data = guest_read_tsc();
 370                 break;
 371         case MSR_IA32_SYSENTER_CS:
 372                 data = vmcs_read32(GUEST_SYSENTER_CS);
 373                 break;
 374         case MSR_IA32_SYSENTER_EIP:
 375                 data = vmcs_read32(GUEST_SYSENTER_EIP);
 376                 break;
 377         case MSR_IA32_SYSENTER_ESP:
 378                 data = vmcs_read32(GUEST_SYSENTER_ESP);
 379                 break;
 380         default:
 381                 msr = find_msr_entry(vcpu, msr_index);
 382                 if (msr) {
 383                         data = msr->data;
 384                         break;
 385                 }
 386                 return kvm_get_msr_common(vcpu, msr_index, pdata);
 387         }
 388
 389         *pdata = data;
 390         return 0;
 391 }
 392
 393 /*
 394  * Writes msr value into into the appropriate "register".
 395  * Returns 0 on success, non-0 otherwise.
 396  * Assumes vcpu_load() was already called.
 397  */
 398 static int vmx_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
 399 {
 400         struct vmx_msr_entry *msr;
 401         switch (msr_index) {
 402 #ifdef CONFIG_X86_64
 403         case MSR_EFER:
 404                 return kvm_set_msr_common(vcpu, msr_index, data);
 405         case MSR_FS_BASE:
 406                 vmcs_writel(GUEST_FS_BASE, data);
 407                 break;
 408         case MSR_GS_BASE:
 409                 vmcs_writel(GUEST_GS_BASE, data);
 410                 break;
 411 #endif
 412         case MSR_IA32_SYSENTER_CS:
 413                 vmcs_write32(GUEST_SYSENTER_CS, data);
 414                 break;
 415         case MSR_IA32_SYSENTER_EIP:
 416                 vmcs_write32(GUEST_SYSENTER_EIP, data);
 417                 break;
 418         case MSR_IA32_SYSENTER_ESP:
 419                 vmcs_write32(GUEST_SYSENTER_ESP, data);
 420                 break;
 421         case MSR_IA32_TIME_STAMP_COUNTER: {
 422                 guest_write_tsc(data);
 423                 break;
 424         }
 425         default:
 426                 msr = find_msr_entry(vcpu, msr_index);
 427                 if (msr) {
 428                         msr->data = data;
 429                         break;
 430                 }
 431                 return kvm_set_msr_common(vcpu, msr_index, data);
 432                 msr->data = data;
 433                 break;
 434         }
 435
 436         return 0;
 437 }
 438
 439 /*
 440  * Sync the rsp and rip registers into the vcpu structure.  This allows
 441  * registers to be accessed by indexing vcpu->regs.
 442  */
 443 static void vcpu_load_rsp_rip(struct kvm_vcpu *vcpu)
 444 {
 445         vcpu->regs[VCPU_REGS_RSP] = vmcs_readl(GUEST_RSP);
 446         vcpu->rip = vmcs_readl(GUEST_RIP);
 447 }
 448
 449 /*
 450  * Syncs rsp and rip back into the vmcs.  Should be called after possible
 451  * modification.
 452  */
 453 static void vcpu_put_rsp_rip(struct kvm_vcpu *vcpu)
 454 {
 455         vmcs_writel(GUEST_RSP, vcpu->regs[VCPU_REGS_RSP]);
 456         vmcs_writel(GUEST_RIP, vcpu->rip);
 457 }
 458
 459 static int set_guest_debug(struct kvm_vcpu *vcpu, struct kvm_debug_guest *dbg)
 460 {
 461         unsigned long dr7 = 0x400;
 462         u32 exception_bitmap;
 463         int old_singlestep;
 464
 465         exception_bitmap = vmcs_read32(EXCEPTION_BITMAP);
 466         old_singlestep = vcpu->guest_debug.singlestep;
 467
 468         vcpu->guest_debug.enabled = dbg->enabled;
 469         if (vcpu->guest_debug.enabled) {
 470                 int i;
 471
 472                 dr7 |= 0x200;  /* exact */
 473                 for (i = 0; i < 4; ++i) {
 474                         if (!dbg->breakpoints[i].enabled)
 475                                 continue;
 476                         vcpu->guest_debug.bp[i] = dbg->breakpoints[i].address;
 477                         dr7 |= 2 << (i*2);    /* global enable */
 478                         dr7 |= 0 << (i*4+16); /* execution breakpoint */
 479                 }
 480
 481                 exception_bitmap |= (1u << 1);  /* Trap debug exceptions */
 482
 483                 vcpu->guest_debug.singlestep = dbg->singlestep;
 484         } else {
 485                 exception_bitmap &= ~(1u << 1); /* Ignore debug exceptions */
 486                 vcpu->guest_debug.singlestep = 0;
 487         }
 488
 489         if (old_singlestep && !vcpu->guest_debug.singlestep) {
 490                 unsigned long flags;
 491
 492                 flags = vmcs_readl(GUEST_RFLAGS);
 493                 flags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
 494                 vmcs_writel(GUEST_RFLAGS, flags);
 495         }
 496
 497         vmcs_write32(EXCEPTION_BITMAP, exception_bitmap);
 498         vmcs_writel(GUEST_DR7, dr7);
 499
 500         return 0;
 501 }
 502
 503 static __init int cpu_has_kvm_support(void)
 504 {
 505         unsigned long ecx = cpuid_ecx(1);
 506         return test_bit(5, &ecx); /* CPUID.1:ECX.VMX[bit 5] -> VT */
 507 }
 508
 509 static __init int vmx_disabled_by_bios(void)
 510 {
 511         u64 msr;
 512
 513         rdmsrl(MSR_IA32_FEATURE_CONTROL, msr);
 514         return (msr & 5) == 1; /* locked but not enabled */
 515 }
 516
 517 static void hardware_enable(void *garbage)
 518 {
 519         int cpu = raw_smp_processor_id();
 520         u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
 521         u64 old;
 522
 523         rdmsrl(MSR_IA32_FEATURE_CONTROL, old);
 524         if ((old & 5) != 5)
 525                 /* enable and lock */
 526                 wrmsrl(MSR_IA32_FEATURE_CONTROL, old | 5);
 527         write_cr4(read_cr4() | CR4_VMXE); /* FIXME: not cpu hotplug safe */
 528         asm volatile (ASM_VMX_VMXON_RAX : : "a"(&phys_addr), "m"(phys_addr)
 529                       : "memory", "cc");
 530 }
 531
 532 static void hardware_disable(void *garbage)
 533 {
 534         asm volatile (ASM_VMX_VMXOFF : : : "cc");
 535 }
 536
 537 static __init void setup_vmcs_descriptor(void)
 538 {
 539         u32 vmx_msr_low, vmx_msr_high;
 540
 541         rdmsr(MSR_IA32_VMX_BASIC, vmx_msr_low, vmx_msr_high);
 542         vmcs_descriptor.size = vmx_msr_high & 0x1fff;
 543         vmcs_descriptor.order = get_order(vmcs_descriptor.size);
 544         vmcs_descriptor.revision_id = vmx_msr_low;
 545 }
 546
 547 static struct vmcs *alloc_vmcs_cpu(int cpu)
 548 {
 549         int node = cpu_to_node(cpu);
 550         struct page *pages;
 551         struct vmcs *vmcs;
 552
 553         pages = alloc_pages_node(node, GFP_KERNEL, vmcs_descriptor.order);
 554         if (!pages)
 555                 return NULL;
 556         vmcs = page_address(pages);
 557         memset(vmcs, 0, vmcs_descriptor.size);
 558         vmcs->revision_id = vmcs_descriptor.revision_id; /* vmcs revision id */
 559         return vmcs;
 560 }
 561
 562 static struct vmcs *alloc_vmcs(void)
 563 {
 564         return alloc_vmcs_cpu(raw_smp_processor_id());
 565 }
 566
 567 static void free_vmcs(struct vmcs *vmcs)
 568 {
 569         free_pages((unsigned long)vmcs, vmcs_descriptor.order);
 570 }
 571
 572 static __exit void free_kvm_area(void)
 573 {
 574         int cpu;
 575
 576         for_each_online_cpu(cpu)
 577                 free_vmcs(per_cpu(vmxarea, cpu));
 578 }
 579
 580 extern struct vmcs *alloc_vmcs_cpu(int cpu);
 581
 582 static __init int alloc_kvm_area(void)
 583 {
 584         int cpu;
 585
 586         for_each_online_cpu(cpu) {
 587                 struct vmcs *vmcs;
 588
 589                 vmcs = alloc_vmcs_cpu(cpu);
 590                 if (!vmcs) {
 591                         free_kvm_area();
 592                         return -ENOMEM;
 593                 }
 594
 595                 per_cpu(vmxarea, cpu) = vmcs;
 596         }
 597         return 0;
 598 }
 599
 600 static __init int hardware_setup(void)
 601 {
 602         setup_vmcs_descriptor();
 603         return alloc_kvm_area();
 604 }
 605
 606 static __exit void hardware_unsetup(void)
 607 {
 608         free_kvm_area();
 609 }
 610
 611 static void update_exception_bitmap(struct kvm_vcpu *vcpu)
 612 {
 613         if (vcpu->rmode.active)
 614                 vmcs_write32(EXCEPTION_BITMAP, ~0);
 615         else
 616                 vmcs_write32(EXCEPTION_BITMAP, 1 << PF_VECTOR);
 617 }
 618
 619 static void fix_pmode_dataseg(int seg, struct kvm_save_segment *save)
 620 {
 621         struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
 622
 623         if (vmcs_readl(sf->base) == save->base) {
 624                 vmcs_write16(sf->selector, save->selector);
 625                 vmcs_writel(sf->base, save->base);
 626                 vmcs_write32(sf->limit, save->limit);
 627                 vmcs_write32(sf->ar_bytes, save->ar);
 628         } else {
 629                 u32 dpl = (vmcs_read16(sf->selector) & SELECTOR_RPL_MASK)
 630                         << AR_DPL_SHIFT;
 631                 vmcs_write32(sf->ar_bytes, 0x93 | dpl);
 632         }
 633 }
 634
 635 static void enter_pmode(struct kvm_vcpu *vcpu)
 636 {
 637         unsigned long flags;
 638
 639         vcpu->rmode.active = 0;
 640
 641         vmcs_writel(GUEST_TR_BASE, vcpu->rmode.tr.base);
 642         vmcs_write32(GUEST_TR_LIMIT, vcpu->rmode.tr.limit);
 643         vmcs_write32(GUEST_TR_AR_BYTES, vcpu->rmode.tr.ar);
 644
 645         flags = vmcs_readl(GUEST_RFLAGS);
 646         flags &= ~(IOPL_MASK | X86_EFLAGS_VM);
 647         flags |= (vcpu->rmode.save_iopl << IOPL_SHIFT);
 648         vmcs_writel(GUEST_RFLAGS, flags);
 649
 650         vmcs_writel(GUEST_CR4, (vmcs_readl(GUEST_CR4) & ~CR4_VME_MASK) |
 651                         (vmcs_readl(CR4_READ_SHADOW) & CR4_VME_MASK));
 652
 653         update_exception_bitmap(vcpu);
 654
 655         fix_pmode_dataseg(VCPU_SREG_ES, &vcpu->rmode.es);
 656         fix_pmode_dataseg(VCPU_SREG_DS, &vcpu->rmode.ds);
 657         fix_pmode_dataseg(VCPU_SREG_GS, &vcpu->rmode.gs);
 658         fix_pmode_dataseg(VCPU_SREG_FS, &vcpu->rmode.fs);
 659
 660         vmcs_write16(GUEST_SS_SELECTOR, 0);
 661         vmcs_write32(GUEST_SS_AR_BYTES, 0x93);
 662
 663         vmcs_write16(GUEST_CS_SELECTOR,
 664                      vmcs_read16(GUEST_CS_SELECTOR) & ~SELECTOR_RPL_MASK);
 665         vmcs_write32(GUEST_CS_AR_BYTES, 0x9b);
 666 }
 667
 668 static int rmode_tss_base(struct kvm* kvm)
 669 {
 670         gfn_t base_gfn = kvm->memslots[0].base_gfn + kvm->memslots[0].npages - 3;
 671         return base_gfn << PAGE_SHIFT;
 672 }
 673
 674 static void fix_rmode_seg(int seg, struct kvm_save_segment *save)
 675 {
 676         struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
 677
 678         save->selector = vmcs_read16(sf->selector);
 679         save->base = vmcs_readl(sf->base);
 680         save->limit = vmcs_read32(sf->limit);
 681         save->ar = vmcs_read32(sf->ar_bytes);
 682         vmcs_write16(sf->selector, vmcs_readl(sf->base) >> 4);
 683         vmcs_write32(sf->limit, 0xffff);
 684         vmcs_write32(sf->ar_bytes, 0xf3);
 685 }
 686
 687 static void enter_rmode(struct kvm_vcpu *vcpu)
 688 {
 689         unsigned long flags;
 690
 691         vcpu->rmode.active = 1;
 692
 693         vcpu->rmode.tr.base = vmcs_readl(GUEST_TR_BASE);
 694         vmcs_writel(GUEST_TR_BASE, rmode_tss_base(vcpu->kvm));
 695
 696         vcpu->rmode.tr.limit = vmcs_read32(GUEST_TR_LIMIT);
 697         vmcs_write32(GUEST_TR_LIMIT, RMODE_TSS_SIZE - 1);
 698
 699         vcpu->rmode.tr.ar = vmcs_read32(GUEST_TR_AR_BYTES);
 700         vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
 701
 702         flags = vmcs_readl(GUEST_RFLAGS);
 703         vcpu->rmode.save_iopl = (flags & IOPL_MASK) >> IOPL_SHIFT;
 704
 705         flags |= IOPL_MASK | X86_EFLAGS_VM;
 706
 707         vmcs_writel(GUEST_RFLAGS, flags);
 708         vmcs_writel(GUEST_CR4, vmcs_readl(GUEST_CR4) | CR4_VME_MASK);
 709         update_exception_bitmap(vcpu);
 710
 711         vmcs_write16(GUEST_SS_SELECTOR, vmcs_readl(GUEST_SS_BASE) >> 4);
 712         vmcs_write32(GUEST_SS_LIMIT, 0xffff);
 713         vmcs_write32(GUEST_SS_AR_BYTES, 0xf3);
 714
 715         vmcs_write32(GUEST_CS_AR_BYTES, 0xf3);
 716         vmcs_write32(GUEST_CS_LIMIT, 0xffff);
 717         vmcs_write16(GUEST_CS_SELECTOR, vmcs_readl(GUEST_CS_BASE) >> 4);
 718
 719         fix_rmode_seg(VCPU_SREG_ES, &vcpu->rmode.es);
 720         fix_rmode_seg(VCPU_SREG_DS, &vcpu->rmode.ds);
 721         fix_rmode_seg(VCPU_SREG_GS, &vcpu->rmode.gs);
 722         fix_rmode_seg(VCPU_SREG_FS, &vcpu->rmode.fs);
 723 }
 724
 725 #ifdef CONFIG_X86_64
 726
 727 static void enter_lmode(struct kvm_vcpu *vcpu)
 728 {
 729         u32 guest_tr_ar;
 730
 731         guest_tr_ar = vmcs_read32(GUEST_TR_AR_BYTES);
 732         if ((guest_tr_ar & AR_TYPE_MASK) != AR_TYPE_BUSY_64_TSS) {
 733                 printk(KERN_DEBUG "%s: tss fixup for long mode. \n",
 734                        __FUNCTION__);
 735                 vmcs_write32(GUEST_TR_AR_BYTES,
 736                              (guest_tr_ar & ~AR_TYPE_MASK)
 737                              | AR_TYPE_BUSY_64_TSS);
 738         }
 739
 740         vcpu->shadow_efer |= EFER_LMA;
 741
 742         find_msr_entry(vcpu, MSR_EFER)->data |= EFER_LMA | EFER_LME;
 743         vmcs_write32(VM_ENTRY_CONTROLS,
 744                      vmcs_read32(VM_ENTRY_CONTROLS)
 745                      | VM_ENTRY_CONTROLS_IA32E_MASK);
 746 }
 747
 748 static void exit_lmode(struct kvm_vcpu *vcpu)
 749 {
 750         vcpu->shadow_efer &= ~EFER_LMA;
 751
 752         vmcs_write32(VM_ENTRY_CONTROLS,
 753                      vmcs_read32(VM_ENTRY_CONTROLS)
 754                      & ~VM_ENTRY_CONTROLS_IA32E_MASK);
 755 }
 756
 757 #endif
 758
 759 static void vmx_decache_cr0_cr4_guest_bits(struct kvm_vcpu *vcpu)
 760 {
 761         vcpu->cr0 &= KVM_GUEST_CR0_MASK;
 762         vcpu->cr0 |= vmcs_readl(GUEST_CR0) & ~KVM_GUEST_CR0_MASK;
 763
 764         vcpu->cr4 &= KVM_GUEST_CR4_MASK;
 765         vcpu->cr4 |= vmcs_readl(GUEST_CR4) & ~KVM_GUEST_CR4_MASK;
 766 }
 767
 768 static void vmx_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
 769 {
 770         if (vcpu->rmode.active && (cr0 & CR0_PE_MASK))
 771                 enter_pmode(vcpu);
 772
 773         if (!vcpu->rmode.active && !(cr0 & CR0_PE_MASK))
 774                 enter_rmode(vcpu);
 775
 776 #ifdef CONFIG_X86_64
 777         if (vcpu->shadow_efer & EFER_LME) {
 778                 if (!is_paging(vcpu) && (cr0 & CR0_PG_MASK))
 779                         enter_lmode(vcpu);
 780                 if (is_paging(vcpu) && !(cr0 & CR0_PG_MASK))
 781                         exit_lmode(vcpu);
 782         }
 783 #endif
 784
 785         vmcs_writel(CR0_READ_SHADOW, cr0);
 786         vmcs_writel(GUEST_CR0,
 787                     (cr0 & ~KVM_GUEST_CR0_MASK) | KVM_VM_CR0_ALWAYS_ON);
 788         vcpu->cr0 = cr0;
 789 }
 790
 791 /*
 792  * Used when restoring the VM to avoid corrupting segment registers
 793  */
 794 static void vmx_set_cr0_no_modeswitch(struct kvm_vcpu *vcpu, unsigned long cr0)
 795 {
 796         vcpu->rmode.active = ((cr0 & CR0_PE_MASK) == 0);
 797         update_exception_bitmap(vcpu);
 798         vmcs_writel(CR0_READ_SHADOW, cr0);
 799         vmcs_writel(GUEST_CR0,
 800                     (cr0 & ~KVM_GUEST_CR0_MASK) | KVM_VM_CR0_ALWAYS_ON);
 801         vcpu->cr0 = cr0;
 802 }
 803
 804 static void vmx_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
 805 {
 806         vmcs_writel(GUEST_CR3, cr3);
 807 }
 808
 809 static void vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
 810 {
 811         vmcs_writel(CR4_READ_SHADOW, cr4);
 812         vmcs_writel(GUEST_CR4, cr4 | (vcpu->rmode.active ?
 813                     KVM_RMODE_VM_CR4_ALWAYS_ON : KVM_PMODE_VM_CR4_ALWAYS_ON));
 814         vcpu->cr4 = cr4;
 815 }
 816
 817 #ifdef CONFIG_X86_64
 818
 819 static void vmx_set_efer(struct kvm_vcpu *vcpu, u64 efer)
 820 {
 821         struct vmx_msr_entry *msr = find_msr_entry(vcpu, MSR_EFER);
 822
 823         vcpu->shadow_efer = efer;
 824         if (efer & EFER_LMA) {
 825                 vmcs_write32(VM_ENTRY_CONTROLS,
 826                                      vmcs_read32(VM_ENTRY_CONTROLS) |
 827                                      VM_ENTRY_CONTROLS_IA32E_MASK);
 828                 msr->data = efer;
 829
 830         } else {
 831                 vmcs_write32(VM_ENTRY_CONTROLS,
 832                                      vmcs_read32(VM_ENTRY_CONTROLS) &
 833                                      ~VM_ENTRY_CONTROLS_IA32E_MASK);
 834
 835                 msr->data = efer & ~EFER_LME;
 836         }
 837 }
 838
 839 #endif
 840
 841 static u64 vmx_get_segment_base(struct kvm_vcpu *vcpu, int seg)
 842 {
 843         struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
 844
 845         return vmcs_readl(sf->base);
 846 }
 847
 848 static void vmx_get_segment(struct kvm_vcpu *vcpu,
 849                             struct kvm_segment *var, int seg)
 850 {
 851         struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
 852         u32 ar;
 853
 854         var->base = vmcs_readl(sf->base);
 855         var->limit = vmcs_read32(sf->limit);
 856         var->selector = vmcs_read16(sf->selector);
 857         ar = vmcs_read32(sf->ar_bytes);
 858         if (ar & AR_UNUSABLE_MASK)
 859                 ar = 0;
 860         var->type = ar & 15;
 861         var->s = (ar >> 4) & 1;
 862         var->dpl = (ar >> 5) & 3;
 863         var->present = (ar >> 7) & 1;
 864         var->avl = (ar >> 12) & 1;
 865         var->l = (ar >> 13) & 1;
 866         var->db = (ar >> 14) & 1;
 867         var->g = (ar >> 15) & 1;
 868         var->unusable = (ar >> 16) & 1;
 869 }
 870
 871 static void vmx_set_segment(struct kvm_vcpu *vcpu,
 872                             struct kvm_segment *var, int seg)
 873 {
 874         struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
 875         u32 ar;
 876
 877         vmcs_writel(sf->base, var->base);
 878         vmcs_write32(sf->limit, var->limit);
 879         vmcs_write16(sf->selector, var->selector);
 880         if (var->unusable)
 881                 ar = 1 << 16;
 882         else {
 883                 ar = var->type & 15;
 884                 ar |= (var->s & 1) << 4;
 885                 ar |= (var->dpl & 3) << 5;
 886                 ar |= (var->present & 1) << 7;
 887                 ar |= (var->avl & 1) << 12;
 888                 ar |= (var->l & 1) << 13;
 889                 ar |= (var->db & 1) << 14;
 890                 ar |= (var->g & 1) << 15;
 891         }
 892         if (ar == 0) /* a 0 value means unusable */
 893                 ar = AR_UNUSABLE_MASK;
 894         vmcs_write32(sf->ar_bytes, ar);
 895 }
 896
 897 static void vmx_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
 898 {
 899         u32 ar = vmcs_read32(GUEST_CS_AR_BYTES);
 900
 901         *db = (ar >> 14) & 1;
 902         *l = (ar >> 13) & 1;
 903 }
 904
 905 static void vmx_get_idt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
 906 {
 907         dt->limit = vmcs_read32(GUEST_IDTR_LIMIT);
 908         dt->base = vmcs_readl(GUEST_IDTR_BASE);
 909 }
 910
 911 static void vmx_set_idt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
 912 {
 913         vmcs_write32(GUEST_IDTR_LIMIT, dt->limit);
 914         vmcs_writel(GUEST_IDTR_BASE, dt->base);
 915 }
 916
 917 static void vmx_get_gdt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
 918 {
 919         dt->limit = vmcs_read32(GUEST_GDTR_LIMIT);
 920         dt->base = vmcs_readl(GUEST_GDTR_BASE);
 921 }
 922
 923 static void vmx_set_gdt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
 924 {
 925         vmcs_write32(GUEST_GDTR_LIMIT, dt->limit);
 926         vmcs_writel(GUEST_GDTR_BASE, dt->base);
 927 }
 928
 929 static int init_rmode_tss(struct kvm* kvm)
 930 {
 931         struct page *p1, *p2, *p3;
 932         gfn_t fn = rmode_tss_base(kvm) >> PAGE_SHIFT;
 933         char *page;
 934
 935         p1 = _gfn_to_page(kvm, fn++);
 936         p2 = _gfn_to_page(kvm, fn++);
 937         p3 = _gfn_to_page(kvm, fn);
 938
 939         if (!p1 || !p2 || !p3) {
 940                 kvm_printf(kvm,"%s: gfn_to_page failed\n", __FUNCTION__);
 941                 return 0;
 942         }
 943
 944         page = kmap_atomic(p1, KM_USER0);
 945         memset(page, 0, PAGE_SIZE);
 946         *(u16*)(page + 0x66) = TSS_BASE_SIZE + TSS_REDIRECTION_SIZE;
 947         kunmap_atomic(page, KM_USER0);
 948
 949         page = kmap_atomic(p2, KM_USER0);
 950         memset(page, 0, PAGE_SIZE);
 951         kunmap_atomic(page, KM_USER0);
 952
 953         page = kmap_atomic(p3, KM_USER0);
 954         memset(page, 0, PAGE_SIZE);
 955         *(page + RMODE_TSS_SIZE - 2 * PAGE_SIZE - 1) = ~0;
 956         kunmap_atomic(page, KM_USER0);
 957
 958         return 1;
 959 }
 960
 961 static void vmcs_write32_fixedbits(u32 msr, u32 vmcs_field, u32 val)
 962 {
 963         u32 msr_high, msr_low;
 964
 965         rdmsr(msr, msr_low, msr_high);
 966
 967         val &= msr_high;
 968         val |= msr_low;
 969         vmcs_write32(vmcs_field, val);
 970 }
 971
 972 static void seg_setup(int seg)
 973 {
 974         struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
 975
 976         vmcs_write16(sf->selector, 0);
 977         vmcs_writel(sf->base, 0);
 978         vmcs_write32(sf->limit, 0xffff);
 979         vmcs_write32(sf->ar_bytes, 0x93);
 980 }
 981
 982 /*
 983  * Sets up the vmcs for emulated real mode.
 984  */
 985 static int vmx_vcpu_setup(struct kvm_vcpu *vcpu)
 986 {
 987         u32 host_sysenter_cs;
 988         u32 junk;
 989         unsigned long a;
 990         struct descriptor_table dt;
 991         int i;
 992         int ret = 0;
 993         int nr_good_msrs;
 994         extern asmlinkage void kvm_vmx_return(void);
 995
 996         if (!init_rmode_tss(vcpu->kvm)) {
 997                 ret = -ENOMEM;
 998                 goto out;
 999         }
1000
1001         memset(vcpu->regs, 0, sizeof(vcpu->regs));
1002         vcpu->regs[VCPU_REGS_RDX] = get_rdx_init_val();
1003         vcpu->cr8 = 0;
1004         vcpu->apic_base = 0xfee00000 |
1005                         /*for vcpu 0*/ MSR_IA32_APICBASE_BSP |
1006                         MSR_IA32_APICBASE_ENABLE;
1007
1008         fx_init(vcpu);
1009
1010         /*
1011          * GUEST_CS_BASE should really be 0xffff0000, but VT vm86 mode
1012          * insists on having GUEST_CS_BASE == GUEST_CS_SELECTOR << 4.  Sigh.
1013          */
1014         vmcs_write16(GUEST_CS_SELECTOR, 0xf000);
1015         vmcs_writel(GUEST_CS_BASE, 0x000f0000);
1016         vmcs_write32(GUEST_CS_LIMIT, 0xffff);
1017         vmcs_write32(GUEST_CS_AR_BYTES, 0x9b);
1018
1019         seg_setup(VCPU_SREG_DS);
1020         seg_setup(VCPU_SREG_ES);
1021         seg_setup(VCPU_SREG_FS);
1022         seg_setup(VCPU_SREG_GS);
1023         seg_setup(VCPU_SREG_SS);
1024
1025         vmcs_write16(GUEST_TR_SELECTOR, 0);
1026         vmcs_writel(GUEST_TR_BASE, 0);
1027         vmcs_write32(GUEST_TR_LIMIT, 0xffff);
1028         vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
1029
1030         vmcs_write16(GUEST_LDTR_SELECTOR, 0);
1031         vmcs_writel(GUEST_LDTR_BASE, 0);
1032         vmcs_write32(GUEST_LDTR_LIMIT, 0xffff);
1033         vmcs_write32(GUEST_LDTR_AR_BYTES, 0x00082);
1034
1035         vmcs_write32(GUEST_SYSENTER_CS, 0);
1036         vmcs_writel(GUEST_SYSENTER_ESP, 0);
1037         vmcs_writel(GUEST_SYSENTER_EIP, 0);
1038
1039         vmcs_writel(GUEST_RFLAGS, 0x02);
1040         vmcs_writel(GUEST_RIP, 0xfff0);
1041         vmcs_writel(GUEST_RSP, 0);
1042
1043         //todo: dr0 = dr1 = dr2 = dr3 = 0; dr6 = 0xffff0ff0
1044         vmcs_writel(GUEST_DR7, 0x400);
1045
1046         vmcs_writel(GUEST_GDTR_BASE, 0);
1047         vmcs_write32(GUEST_GDTR_LIMIT, 0xffff);
1048
1049         vmcs_writel(GUEST_IDTR_BASE, 0);
1050         vmcs_write32(GUEST_IDTR_LIMIT, 0xffff);
1051
1052         vmcs_write32(GUEST_ACTIVITY_STATE, 0);
1053         vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, 0);
1054         vmcs_write32(GUEST_PENDING_DBG_EXCEPTIONS, 0);
1055
1056         /* I/O */
1057         vmcs_write64(IO_BITMAP_A, 0);
1058         vmcs_write64(IO_BITMAP_B, 0);
1059
1060         guest_write_tsc(0);
1061
1062         vmcs_write64(VMCS_LINK_POINTER, -1ull); /* 22.3.1.5 */
1063
1064         /* Special registers */
1065         vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
1066
1067         /* Control */
1068         vmcs_write32_fixedbits(MSR_IA32_VMX_PINBASED_CTLS,
1069                                PIN_BASED_VM_EXEC_CONTROL,
1070                                PIN_BASED_EXT_INTR_MASK   /* 20.6.1 */
1071                                | PIN_BASED_NMI_EXITING   /* 20.6.1 */
1072                         );
1073         vmcs_write32_fixedbits(MSR_IA32_VMX_PROCBASED_CTLS,
1074                                CPU_BASED_VM_EXEC_CONTROL,
1075                                CPU_BASED_HLT_EXITING         /* 20.6.2 */
1076                                | CPU_BASED_CR8_LOAD_EXITING    /* 20.6.2 */
1077                                | CPU_BASED_CR8_STORE_EXITING   /* 20.6.2 */
1078                                | CPU_BASED_UNCOND_IO_EXITING   /* 20.6.2 */
1079                                | CPU_BASED_MOV_DR_EXITING
1080                                | CPU_BASED_USE_TSC_OFFSETING   /* 21.3 */
1081                         );
1082
1083         vmcs_write32(EXCEPTION_BITMAP, 1 << PF_VECTOR);
1084         vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, 0);
1085         vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, 0);
1086         vmcs_write32(CR3_TARGET_COUNT, 0);           /* 22.2.1 */
1087
1088         vmcs_writel(HOST_CR0, read_cr0());  /* 22.2.3 */
1089         vmcs_writel(HOST_CR4, read_cr4());  /* 22.2.3, 22.2.5 */
1090         vmcs_writel(HOST_CR3, read_cr3());  /* 22.2.3  FIXME: shadow tables */
1091
1092         vmcs_write16(HOST_CS_SELECTOR, __KERNEL_CS);  /* 22.2.4 */
1093         vmcs_write16(HOST_DS_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
1094         vmcs_write16(HOST_ES_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
1095         vmcs_write16(HOST_FS_SELECTOR, read_fs());    /* 22.2.4 */
1096         vmcs_write16(HOST_GS_SELECTOR, read_gs());    /* 22.2.4 */
1097         vmcs_write16(HOST_SS_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
1098 #ifdef CONFIG_X86_64
1099         rdmsrl(MSR_FS_BASE, a);
1100         vmcs_writel(HOST_FS_BASE, a); /* 22.2.4 */
1101         rdmsrl(MSR_GS_BASE, a);
1102         vmcs_writel(HOST_GS_BASE, a); /* 22.2.4 */
1103 #else
1104         vmcs_writel(HOST_FS_BASE, 0); /* 22.2.4 */
1105         vmcs_writel(HOST_GS_BASE, 0); /* 22.2.4 */
1106 #endif
1107
1108         vmcs_write16(HOST_TR_SELECTOR, GDT_ENTRY_TSS*8);  /* 22.2.4 */
1109
1110         get_idt(&dt);
1111         vmcs_writel(HOST_IDTR_BASE, dt.base);   /* 22.2.4 */
1112
1113
1114         vmcs_writel(HOST_RIP, (unsigned long)kvm_vmx_return); /* 22.2.5 */
1115
1116         rdmsr(MSR_IA32_SYSENTER_CS, host_sysenter_cs, junk);
1117         vmcs_write32(HOST_IA32_SYSENTER_CS, host_sysenter_cs);
1118         rdmsrl(MSR_IA32_SYSENTER_ESP, a);
1119         vmcs_writel(HOST_IA32_SYSENTER_ESP, a);   /* 22.2.3 */
1120         rdmsrl(MSR_IA32_SYSENTER_EIP, a);
1121         vmcs_writel(HOST_IA32_SYSENTER_EIP, a);   /* 22.2.3 */
1122
1123         for (i = 0; i < NR_VMX_MSR; ++i) {
1124                 u32 index = vmx_msr_index[i];
1125                 u32 data_low, data_high;
1126                 u64 data;
1127                 int j = vcpu->nmsrs;
1128
1129                 if (rdmsr_safe(index, &data_low, &data_high) < 0)
1130                         continue;
1131                 if (wrmsr_safe(index, data_low, data_high) < 0)
1132                         continue;
1133                 data = data_low | ((u64)data_high << 32);
1134                 vcpu->host_msrs[j].index = index;
1135                 vcpu->host_msrs[j].reserved = 0;
1136                 vcpu->host_msrs[j].data = data;
1137                 vcpu->guest_msrs[j] = vcpu->host_msrs[j];
1138                 ++vcpu->nmsrs;
1139         }
1140         printk(KERN_DEBUG "kvm: msrs: %d\n", vcpu->nmsrs);
1141
1142         nr_good_msrs = vcpu->nmsrs - NR_BAD_MSRS;
1143         vmcs_writel(VM_ENTRY_MSR_LOAD_ADDR,
1144                     virt_to_phys(vcpu->guest_msrs + NR_BAD_MSRS));
1145         vmcs_writel(VM_EXIT_MSR_STORE_ADDR,
1146                     virt_to_phys(vcpu->guest_msrs + NR_BAD_MSRS));
1147         vmcs_writel(VM_EXIT_MSR_LOAD_ADDR,
1148                     virt_to_phys(vcpu->host_msrs + NR_BAD_MSRS));
1149         vmcs_write32_fixedbits(MSR_IA32_VMX_EXIT_CTLS, VM_EXIT_CONTROLS,
1150                                (HOST_IS_64 << 9));  /* 22.2,1, 20.7.1 */
1151         vmcs_write32(VM_EXIT_MSR_STORE_COUNT, nr_good_msrs); /* 22.2.2 */
1152         vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, nr_good_msrs);  /* 22.2.2 */
1153         vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, nr_good_msrs); /* 22.2.2 */
1154
1155
1156         /* 22.2.1, 20.8.1 */
1157         vmcs_write32_fixedbits(MSR_IA32_VMX_ENTRY_CTLS,
1158                                VM_ENTRY_CONTROLS, 0);
1159         vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);  /* 22.2.1 */
1160
1161 #ifdef CONFIG_X86_64
1162         vmcs_writel(VIRTUAL_APIC_PAGE_ADDR, 0);
1163         vmcs_writel(TPR_THRESHOLD, 0);
1164 #endif
1165
1166         vmcs_writel(CR0_GUEST_HOST_MASK, KVM_GUEST_CR0_MASK);
1167         vmcs_writel(CR4_GUEST_HOST_MASK, KVM_GUEST_CR4_MASK);
1168
1169         vcpu->cr0 = 0x60000010;
1170         vmx_set_cr0(vcpu, vcpu->cr0); // enter rmode
1171         vmx_set_cr4(vcpu, 0);
1172 #ifdef CONFIG_X86_64
1173         vmx_set_efer(vcpu, 0);
1174 #endif
1175
1176         return 0;
1177
1178 out:
1179         return ret;
1180 }
1181
1182 static void inject_rmode_irq(struct kvm_vcpu *vcpu, int irq)
1183 {
1184         u16 ent[2];
1185         u16 cs;
1186         u16 ip;
1187         unsigned long flags;
1188         unsigned long ss_base = vmcs_readl(GUEST_SS_BASE);
1189         u16 sp =  vmcs_readl(GUEST_RSP);
1190         u32 ss_limit = vmcs_read32(GUEST_SS_LIMIT);
1191
1192         if (sp > ss_limit || sp - 6 > sp) {
1193                 vcpu_printf(vcpu, "%s: #SS, rsp 0x%lx ss 0x%lx limit 0x%x\n",
1194                             __FUNCTION__,
1195                             vmcs_readl(GUEST_RSP),
1196                             vmcs_readl(GUEST_SS_BASE),
1197                             vmcs_read32(GUEST_SS_LIMIT));
1198                 return;
1199         }
1200
1201         if (kvm_read_guest(vcpu, irq * sizeof(ent), sizeof(ent), &ent) !=
1202                                                                 sizeof(ent)) {
1203                 vcpu_printf(vcpu, "%s: read guest err\n", __FUNCTION__);
1204                 return;
1205         }
1206
1207         flags =  vmcs_readl(GUEST_RFLAGS);
1208         cs =  vmcs_readl(GUEST_CS_BASE) >> 4;
1209         ip =  vmcs_readl(GUEST_RIP);
1210
1211
1212         if (kvm_write_guest(vcpu, ss_base + sp - 2, 2, &flags) != 2 ||
1213             kvm_write_guest(vcpu, ss_base + sp - 4, 2, &cs) != 2 ||
1214             kvm_write_guest(vcpu, ss_base + sp - 6, 2, &ip) != 2) {
1215                 vcpu_printf(vcpu, "%s: write guest err\n", __FUNCTION__);
1216                 return;
1217         }
1218
1219         vmcs_writel(GUEST_RFLAGS, flags &
1220                     ~( X86_EFLAGS_IF | X86_EFLAGS_AC | X86_EFLAGS_TF));
1221         vmcs_write16(GUEST_CS_SELECTOR, ent[1]) ;
1222         vmcs_writel(GUEST_CS_BASE, ent[1] << 4);
1223         vmcs_writel(GUEST_RIP, ent[0]);
1224         vmcs_writel(GUEST_RSP, (vmcs_readl(GUEST_RSP) & ~0xffff) | (sp - 6));
1225 }
1226
1227 static void kvm_do_inject_irq(struct kvm_vcpu *vcpu)
1228 {
1229         int word_index = __ffs(vcpu->irq_summary);
1230         int bit_index = __ffs(vcpu->irq_pending[word_index]);
1231         int irq = word_index * BITS_PER_LONG + bit_index;
1232
1233         clear_bit(bit_index, &vcpu->irq_pending[word_index]);
1234         if (!vcpu->irq_pending[word_index])
1235                 clear_bit(word_index, &vcpu->irq_summary);
1236
1237         if (vcpu->rmode.active) {
1238                 inject_rmode_irq(vcpu, irq);
1239                 return;
1240         }
1241         vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
1242                         irq | INTR_TYPE_EXT_INTR | INTR_INFO_VALID_MASK);
1243 }
1244
1245
1246 static void do_interrupt_requests(struct kvm_vcpu *vcpu,
1247                                        struct kvm_run *kvm_run)
1248 {
1249         u32 cpu_based_vm_exec_control;
1250
1251         vcpu->interrupt_window_open =
1252                 ((vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_IF) &&
1253                  (vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & 3) == 0);
1254
1255         if (vcpu->interrupt_window_open &&
1256             vcpu->irq_summary &&
1257             !(vmcs_read32(VM_ENTRY_INTR_INFO_FIELD) & INTR_INFO_VALID_MASK))
1258                 /*
1259                  * If interrupts enabled, and not blocked by sti or mov ss. Good.
1260                  */
1261                 kvm_do_inject_irq(vcpu);
1262
1263         cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
1264         if (!vcpu->interrupt_window_open &&
1265             (vcpu->irq_summary || kvm_run->request_interrupt_window))
1266                 /*
1267                  * Interrupts blocked.  Wait for unblock.
1268                  */
1269                 cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_INTR_PENDING;
1270         else
1271                 cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_INTR_PENDING;
1272         vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
1273 }
1274
1275 static void kvm_guest_debug_pre(struct kvm_vcpu *vcpu)
1276 {
1277         struct kvm_guest_debug *dbg = &vcpu->guest_debug;
1278
1279         set_debugreg(dbg->bp[0], 0);
1280         set_debugreg(dbg->bp[1], 1);
1281         set_debugreg(dbg->bp[2], 2);
1282         set_debugreg(dbg->bp[3], 3);
1283
1284         if (dbg->singlestep) {
1285                 unsigned long flags;
1286
1287                 flags = vmcs_readl(GUEST_RFLAGS);
1288                 flags |= X86_EFLAGS_TF | X86_EFLAGS_RF;
1289                 vmcs_writel(GUEST_RFLAGS, flags);
1290         }
1291 }
1292
1293 static int handle_rmode_exception(struct kvm_vcpu *vcpu,
1294                                   int vec, u32 err_code)
1295 {
1296         if (!vcpu->rmode.active)
1297                 return 0;
1298
1299         if (vec == GP_VECTOR && err_code == 0)
1300                 if (emulate_instruction(vcpu, NULL, 0, 0) == EMULATE_DONE)
1301                         return 1;
1302         return 0;
1303 }
1304
1305 static int handle_exception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1306 {
1307         u32 intr_info, error_code;
1308         unsigned long cr2, rip;
1309         u32 vect_info;
1310         enum emulation_result er;
1311         int r;
1312
1313         vect_info = vmcs_read32(IDT_VECTORING_INFO_FIELD);
1314         intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
1315
1316         if ((vect_info & VECTORING_INFO_VALID_MASK) &&
1317                                                 !is_page_fault(intr_info)) {
1318                 printk(KERN_ERR "%s: unexpected, vectoring info 0x%x "
1319                        "intr info 0x%x\n", __FUNCTION__, vect_info, intr_info);
1320         }
1321
1322         if (is_external_interrupt(vect_info)) {
1323                 int irq = vect_info & VECTORING_INFO_VECTOR_MASK;
1324                 set_bit(irq, vcpu->irq_pending);
1325                 set_bit(irq / BITS_PER_LONG, &vcpu->irq_summary);
1326         }
1327
1328         if ((intr_info & INTR_INFO_INTR_TYPE_MASK) == 0x200) { /* nmi */
1329                 asm ("int $2");
1330                 return 1;
1331         }
1332         error_code = 0;
1333         rip = vmcs_readl(GUEST_RIP);
1334         if (intr_info & INTR_INFO_DELIEVER_CODE_MASK)
1335                 error_code = vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
1336         if (is_page_fault(intr_info)) {
1337                 cr2 = vmcs_readl(EXIT_QUALIFICATION);
1338
1339                 spin_lock(&vcpu->kvm->lock);
1340                 r = kvm_mmu_page_fault(vcpu, cr2, error_code);
1341                 if (r < 0) {
1342                         spin_unlock(&vcpu->kvm->lock);
1343                         return r;
1344                 }
1345                 if (!r) {
1346                         spin_unlock(&vcpu->kvm->lock);
1347                         return 1;
1348                 }
1349
1350                 er = emulate_instruction(vcpu, kvm_run, cr2, error_code);
1351                 spin_unlock(&vcpu->kvm->lock);
1352
1353                 switch (er) {
1354                 case EMULATE_DONE:
1355                         return 1;
1356                 case EMULATE_DO_MMIO:
1357                         ++kvm_stat.mmio_exits;
1358                         kvm_run->exit_reason = KVM_EXIT_MMIO;
1359                         return 0;
1360                  case EMULATE_FAIL:
1361                         vcpu_printf(vcpu, "%s: emulate fail\n", __FUNCTION__);
1362                         break;
1363                 default:
1364                         BUG();
1365                 }
1366         }
1367
1368         if (vcpu->rmode.active &&
1369             handle_rmode_exception(vcpu, intr_info & INTR_INFO_VECTOR_MASK,
1370                                                                 error_code))
1371                 return 1;
1372
1373         if ((intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK)) == (INTR_TYPE_EXCEPTION | 1)) {
1374                 kvm_run->exit_reason = KVM_EXIT_DEBUG;
1375                 return 0;
1376         }
1377         kvm_run->exit_reason = KVM_EXIT_EXCEPTION;
1378         kvm_run->ex.exception = intr_info & INTR_INFO_VECTOR_MASK;
1379         kvm_run->ex.error_code = error_code;
1380         return 0;
1381 }
1382
1383 static int handle_external_interrupt(struct kvm_vcpu *vcpu,
1384                                      struct kvm_run *kvm_run)
1385 {
1386         ++kvm_stat.irq_exits;
1387         return 1;
1388 }
1389
1390 static int handle_triple_fault(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1391 {
1392         kvm_run->exit_reason = KVM_EXIT_SHUTDOWN;
1393         return 0;
1394 }
1395
1396 static int get_io_count(struct kvm_vcpu *vcpu, u64 *count)
1397 {
1398         u64 inst;
1399         gva_t rip;
1400         int countr_size;
1401         int i, n;
1402
1403         if ((vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_VM)) {
1404                 countr_size = 2;
1405         } else {
1406                 u32 cs_ar = vmcs_read32(GUEST_CS_AR_BYTES);
1407
1408                 countr_size = (cs_ar & AR_L_MASK) ? 8:
1409                               (cs_ar & AR_DB_MASK) ? 4: 2;
1410         }
1411
1412         rip =  vmcs_readl(GUEST_RIP);
1413         if (countr_size != 8)
1414                 rip += vmcs_readl(GUEST_CS_BASE);
1415
1416         n = kvm_read_guest(vcpu, rip, sizeof(inst), &inst);
1417
1418         for (i = 0; i < n; i++) {
1419                 switch (((u8*)&inst)[i]) {
1420                 case 0xf0:
1421                 case 0xf2:
1422                 case 0xf3:
1423                 case 0x2e:
1424                 case 0x36:
1425                 case 0x3e:
1426                 case 0x26:
1427                 case 0x64:
1428                 case 0x65:
1429                 case 0x66:
1430                         break;
1431                 case 0x67:
1432                         countr_size = (countr_size == 2) ? 4: (countr_size >> 1);
1433                 default:
1434                         goto done;
1435                 }
1436         }
1437         return 0;
1438 done:
1439         countr_size *= 8;
1440         *count = vcpu->regs[VCPU_REGS_RCX] & (~0ULL >> (64 - countr_size));
1441         return 1;
1442 }
1443
1444 static int handle_io(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1445 {
1446         u64 exit_qualification;
1447
1448         ++kvm_stat.io_exits;
1449         exit_qualification = vmcs_read64(EXIT_QUALIFICATION);
1450         kvm_run->exit_reason = KVM_EXIT_IO;
1451         if (exit_qualification & 8)
1452                 kvm_run->io.direction = KVM_EXIT_IO_IN;
1453         else
1454                 kvm_run->io.direction = KVM_EXIT_IO_OUT;
1455         kvm_run->io.size = (exit_qualification & 7) + 1;
1456         kvm_run->io.string = (exit_qualification & 16) != 0;
1457         kvm_run->io.string_down
1458                 = (vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_DF) != 0;
1459         kvm_run->io.rep = (exit_qualification & 32) != 0;
1460         kvm_run->io.port = exit_qualification >> 16;
1461         if (kvm_run->io.string) {
1462                 if (!get_io_count(vcpu, &kvm_run->io.count))
1463                         return 1;
1464                 kvm_run->io.address = vmcs_readl(GUEST_LINEAR_ADDRESS);
1465         } else
1466                 kvm_run->io.value = vcpu->regs[VCPU_REGS_RAX]; /* rax */
1467         return 0;
1468 }
1469
1470 static int handle_cr(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1471 {
1472         u64 exit_qualification;
1473         int cr;
1474         int reg;
1475
1476         exit_qualification = vmcs_read64(EXIT_QUALIFICATION);
1477         cr = exit_qualification & 15;
1478         reg = (exit_qualification >> 8) & 15;
1479         switch ((exit_qualification >> 4) & 3) {
1480         case 0: /* mov to cr */
1481                 switch (cr) {
1482                 case 0:
1483                         vcpu_load_rsp_rip(vcpu);
1484                         set_cr0(vcpu, vcpu->regs[reg]);
1485                         skip_emulated_instruction(vcpu);
1486                         return 1;
1487                 case 3:
1488                         vcpu_load_rsp_rip(vcpu);
1489                         set_cr3(vcpu, vcpu->regs[reg]);
1490                         skip_emulated_instruction(vcpu);
1491                         return 1;
1492                 case 4:
1493                         vcpu_load_rsp_rip(vcpu);
1494                         set_cr4(vcpu, vcpu->regs[reg]);
1495                         skip_emulated_instruction(vcpu);
1496                         return 1;
1497                 case 8:
1498                         vcpu_load_rsp_rip(vcpu);
1499                         set_cr8(vcpu, vcpu->regs[reg]);
1500                         skip_emulated_instruction(vcpu);
1501                         return 1;
1502                 };
1503                 break;
1504         case 1: /*mov from cr*/
1505                 switch (cr) {
1506                 case 3:
1507                         vcpu_load_rsp_rip(vcpu);
1508                         vcpu->regs[reg] = vcpu->cr3;
1509                         vcpu_put_rsp_rip(vcpu);
1510                         skip_emulated_instruction(vcpu);
1511                         return 1;
1512                 case 8:
1513                         printk(KERN_DEBUG "handle_cr: read CR8 "
1514                                "cpu erratum AA15\n");
1515                         vcpu_load_rsp_rip(vcpu);
1516                         vcpu->regs[reg] = vcpu->cr8;
1517                         vcpu_put_rsp_rip(vcpu);
1518                         skip_emulated_instruction(vcpu);
1519                         return 1;
1520                 }
1521                 break;
1522         case 3: /* lmsw */
1523                 lmsw(vcpu, (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f);
1524
1525                 skip_emulated_instruction(vcpu);
1526                 return 1;
1527         default:
1528                 break;
1529         }
1530         kvm_run->exit_reason = 0;
1531         printk(KERN_ERR "kvm: unhandled control register: op %d cr %d\n",
1532                (int)(exit_qualification >> 4) & 3, cr);
1533         return 0;
1534 }
1535
1536 static int handle_dr(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1537 {
1538         u64 exit_qualification;
1539         unsigned long val;
1540         int dr, reg;
1541
1542         /*
1543          * FIXME: this code assumes the host is debugging the guest.
1544          *        need to deal with guest debugging itself too.
1545          */
1546         exit_qualification = vmcs_read64(EXIT_QUALIFICATION);
1547         dr = exit_qualification & 7;
1548         reg = (exit_qualification >> 8) & 15;
1549         vcpu_load_rsp_rip(vcpu);
1550         if (exit_qualification & 16) {
1551                 /* mov from dr */
1552                 switch (dr) {
1553                 case 6:
1554                         val = 0xffff0ff0;
1555                         break;
1556                 case 7:
1557                         val = 0x400;
1558                         break;
1559                 default:
1560                         val = 0;
1561                 }
1562                 vcpu->regs[reg] = val;
1563         } else {
1564                 /* mov to dr */
1565         }
1566         vcpu_put_rsp_rip(vcpu);
1567         skip_emulated_instruction(vcpu);
1568         return 1;
1569 }
1570
1571 static int handle_cpuid(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1572 {
1573         kvm_run->exit_reason = KVM_EXIT_CPUID;
1574         return 0;
1575 }
1576
1577 static int handle_rdmsr(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1578 {
1579         u32 ecx = vcpu->regs[VCPU_REGS_RCX];
1580         u64 data;
1581
1582         if (vmx_get_msr(vcpu, ecx, &data)) {
1583                 vmx_inject_gp(vcpu, 0);
1584                 return 1;
1585         }
1586
1587         /* FIXME: handling of bits 32:63 of rax, rdx */
1588         vcpu->regs[VCPU_REGS_RAX] = data & -1u;
1589         vcpu->regs[VCPU_REGS_RDX] = (data >> 32) & -1u;
1590         skip_emulated_instruction(vcpu);
1591         return 1;
1592 }
1593
1594 static int handle_wrmsr(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1595 {
1596         u32 ecx = vcpu->regs[VCPU_REGS_RCX];
1597         u64 data = (vcpu->regs[VCPU_REGS_RAX] & -1u)
1598                 | ((u64)(vcpu->regs[VCPU_REGS_RDX] & -1u) << 32);
1599
1600         if (vmx_set_msr(vcpu, ecx, data) != 0) {
1601                 vmx_inject_gp(vcpu, 0);
1602                 return 1;
1603         }
1604
1605         skip_emulated_instruction(vcpu);
1606         return 1;
1607 }
1608
1609 static void post_kvm_run_save(struct kvm_vcpu *vcpu,
1610                               struct kvm_run *kvm_run)
1611 {
1612         kvm_run->if_flag = (vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_IF) != 0;
1613         kvm_run->cr8 = vcpu->cr8;
1614         kvm_run->apic_base = vcpu->apic_base;
1615         kvm_run->ready_for_interrupt_injection = (vcpu->interrupt_window_open &&
1616                                                   vcpu->irq_summary == 0);
1617 }
1618
1619 static int handle_interrupt_window(struct kvm_vcpu *vcpu,
1620                                    struct kvm_run *kvm_run)
1621 {
1622         /*
1623          * If the user space waits to inject interrupts, exit as soon as
1624          * possible
1625          */
1626         if (kvm_run->request_interrupt_window &&
1627             !vcpu->irq_summary) {
1628                 kvm_run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
1629                 ++kvm_stat.irq_window_exits;
1630                 return 0;
1631         }
1632         return 1;
1633 }
1634
1635 static int handle_halt(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1636 {
1637         skip_emulated_instruction(vcpu);
1638         if (vcpu->irq_summary)
1639                 return 1;
1640
1641         kvm_run->exit_reason = KVM_EXIT_HLT;
1642         ++kvm_stat.halt_exits;
1643         return 0;
1644 }
1645
1646 /*
1647  * The exit handlers return 1 if the exit was handled fully and guest execution
1648  * may resume.  Otherwise they set the kvm_run parameter to indicate what needs
1649  * to be done to userspace and return 0.
1650  */
1651 static int (*kvm_vmx_exit_handlers[])(struct kvm_vcpu *vcpu,
1652                                       struct kvm_run *kvm_run) = {
1653         [EXIT_REASON_EXCEPTION_NMI]           = handle_exception,
1654         [EXIT_REASON_EXTERNAL_INTERRUPT]      = handle_external_interrupt,
1655         [EXIT_REASON_TRIPLE_FAULT]            = handle_triple_fault,
1656         [EXIT_REASON_IO_INSTRUCTION]          = handle_io,
1657         [EXIT_REASON_CR_ACCESS]               = handle_cr,
1658         [EXIT_REASON_DR_ACCESS]               = handle_dr,
1659         [EXIT_REASON_CPUID]                   = handle_cpuid,
1660         [EXIT_REASON_MSR_READ]                = handle_rdmsr,
1661         [EXIT_REASON_MSR_WRITE]               = handle_wrmsr,
1662         [EXIT_REASON_PENDING_INTERRUPT]       = handle_interrupt_window,
1663         [EXIT_REASON_HLT]                     = handle_halt,
1664 };
1665
1666 static const int kvm_vmx_max_exit_handlers =
1667         sizeof(kvm_vmx_exit_handlers) / sizeof(*kvm_vmx_exit_handlers);
1668
1669 /*
1670  * The guest has exited.  See if we can fix it or if we need userspace
1671  * assistance.
1672  */
1673 static int kvm_handle_exit(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
1674 {
1675         u32 vectoring_info = vmcs_read32(IDT_VECTORING_INFO_FIELD);
1676         u32 exit_reason = vmcs_read32(VM_EXIT_REASON);
1677
1678         if ( (vectoring_info & VECTORING_INFO_VALID_MASK) &&
1679                                 exit_reason != EXIT_REASON_EXCEPTION_NMI )
1680                 printk(KERN_WARNING "%s: unexpected, valid vectoring info and "
1681                        "exit reason is 0x%x\n", __FUNCTION__, exit_reason);
1682         kvm_run->instruction_length = vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
1683         if (exit_reason < kvm_vmx_max_exit_handlers
1684             && kvm_vmx_exit_handlers[exit_reason])
1685                 return kvm_vmx_exit_handlers[exit_reason](vcpu, kvm_run);
1686         else {
1687                 kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
1688                 kvm_run->hw.hardware_exit_reason = exit_reason;
1689         }
1690         return 0;
1691 }
1692
1693 /*
1694  * Check if userspace requested an interrupt window, and that the
1695  * interrupt window is open.
1696  *
1697  * No need to exit to userspace if we already have an interrupt queued.
1698  */
1699 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu,
1700                                           struct kvm_run *kvm_run)
1701 {
1702         return (!vcpu->irq_summary &&
1703                 kvm_run->request_interrupt_window &&
1704                 vcpu->interrupt_window_open &&
1705                 (vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_IF));
1706 }
1707
1708 static int vmx_vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1709 {
1710         u8 fail;
1711         u16 fs_sel, gs_sel, ldt_sel;
1712         int fs_gs_ldt_reload_needed;
1713         int r;
1714
1715 again:
1716         /*
1717          * Set host fs and gs selectors.  Unfortunately, 22.2.3 does not
1718          * allow segment selectors with cpl > 0 or ti == 1.
1719          */
1720         fs_sel = read_fs();
1721         gs_sel = read_gs();
1722         ldt_sel = read_ldt();
1723         fs_gs_ldt_reload_needed = (fs_sel & 7) | (gs_sel & 7) | ldt_sel;
1724         if (!fs_gs_ldt_reload_needed) {
1725                 vmcs_write16(HOST_FS_SELECTOR, fs_sel);
1726                 vmcs_write16(HOST_GS_SELECTOR, gs_sel);
1727         } else {
1728                 vmcs_write16(HOST_FS_SELECTOR, 0);
1729                 vmcs_write16(HOST_GS_SELECTOR, 0);
1730         }
1731
1732 #ifdef CONFIG_X86_64
1733         vmcs_writel(HOST_FS_BASE, read_msr(MSR_FS_BASE));
1734         vmcs_writel(HOST_GS_BASE, read_msr(MSR_GS_BASE));
1735 #else
1736         vmcs_writel(HOST_FS_BASE, segment_base(fs_sel));
1737         vmcs_writel(HOST_GS_BASE, segment_base(gs_sel));
1738 #endif
1739
1740         if (!vcpu->mmio_read_completed)
1741                 do_interrupt_requests(vcpu, kvm_run);
1742
1743         if (vcpu->guest_debug.enabled)
1744                 kvm_guest_debug_pre(vcpu);
1745
1746         fx_save(vcpu->host_fx_image);
1747         fx_restore(vcpu->guest_fx_image);
1748
1749         save_msrs(vcpu->host_msrs, vcpu->nmsrs);
1750         load_msrs(vcpu->guest_msrs, NR_BAD_MSRS);
1751
1752         asm (
1753                 /* Store host registers */
1754                 "pushf \n\t"
1755 #ifdef CONFIG_X86_64
1756                 "push %%rax; push %%rbx; push %%rdx;"
1757                 "push %%rsi; push %%rdi; push %%rbp;"
1758                 "push %%r8;  push %%r9;  push %%r10; push %%r11;"
1759                 "push %%r12; push %%r13; push %%r14; push %%r15;"
1760                 "push %%rcx \n\t"
1761                 ASM_VMX_VMWRITE_RSP_RDX "\n\t"
1762 #else
1763                 "pusha; push %%ecx \n\t"
1764                 ASM_VMX_VMWRITE_RSP_RDX "\n\t"
1765 #endif
1766                 /* Check if vmlaunch of vmresume is needed */
1767                 "cmp $0, %1 \n\t"
1768                 /* Load guest registers.  Don't clobber flags. */
1769 #ifdef CONFIG_X86_64
1770                 "mov %c[cr2](%3), %%rax \n\t"
1771                 "mov %%rax, %%cr2 \n\t"
1772                 "mov %c[rax](%3), %%rax \n\t"
1773                 "mov %c[rbx](%3), %%rbx \n\t"
1774                 "mov %c[rdx](%3), %%rdx \n\t"
1775                 "mov %c[rsi](%3), %%rsi \n\t"
1776                 "mov %c[rdi](%3), %%rdi \n\t"
1777                 "mov %c[rbp](%3), %%rbp \n\t"
1778                 "mov %c[r8](%3),  %%r8  \n\t"
1779                 "mov %c[r9](%3),  %%r9  \n\t"
1780                 "mov %c[r10](%3), %%r10 \n\t"
1781                 "mov %c[r11](%3), %%r11 \n\t"
1782                 "mov %c[r12](%3), %%r12 \n\t"
1783                 "mov %c[r13](%3), %%r13 \n\t"
1784                 "mov %c[r14](%3), %%r14 \n\t"
1785                 "mov %c[r15](%3), %%r15 \n\t"
1786                 "mov %c[rcx](%3), %%rcx \n\t" /* kills %3 (rcx) */
1787 #else
1788                 "mov %c[cr2](%3), %%eax \n\t"
1789                 "mov %%eax,   %%cr2 \n\t"
1790                 "mov %c[rax](%3), %%eax \n\t"
1791                 "mov %c[rbx](%3), %%ebx \n\t"
1792                 "mov %c[rdx](%3), %%edx \n\t"
1793                 "mov %c[rsi](%3), %%esi \n\t"
1794                 "mov %c[rdi](%3), %%edi \n\t"
1795                 "mov %c[rbp](%3), %%ebp \n\t"
1796                 "mov %c[rcx](%3), %%ecx \n\t" /* kills %3 (ecx) */
1797 #endif
1798                 /* Enter guest mode */
1799                 "jne launched \n\t"
1800                 ASM_VMX_VMLAUNCH "\n\t"
1801                 "jmp kvm_vmx_return \n\t"
1802                 "launched: " ASM_VMX_VMRESUME "\n\t"
1803                 ".globl kvm_vmx_return \n\t"
1804                 "kvm_vmx_return: "
1805                 /* Save guest registers, load host registers, keep flags */
1806 #ifdef CONFIG_X86_64
1807                 "xchg %3,     (%%rsp) \n\t"
1808                 "mov %%rax, %c[rax](%3) \n\t"
1809                 "mov %%rbx, %c[rbx](%3) \n\t"
1810                 "pushq (%%rsp); popq %c[rcx](%3) \n\t"
1811                 "mov %%rdx, %c[rdx](%3) \n\t"
1812                 "mov %%rsi, %c[rsi](%3) \n\t"
1813                 "mov %%rdi, %c[rdi](%3) \n\t"
1814                 "mov %%rbp, %c[rbp](%3) \n\t"
1815                 "mov %%r8,  %c[r8](%3) \n\t"
1816                 "mov %%r9,  %c[r9](%3) \n\t"
1817                 "mov %%r10, %c[r10](%3) \n\t"
1818                 "mov %%r11, %c[r11](%3) \n\t"
1819                 "mov %%r12, %c[r12](%3) \n\t"
1820                 "mov %%r13, %c[r13](%3) \n\t"
1821                 "mov %%r14, %c[r14](%3) \n\t"
1822                 "mov %%r15, %c[r15](%3) \n\t"
1823                 "mov %%cr2, %%rax   \n\t"
1824                 "mov %%rax, %c[cr2](%3) \n\t"
1825                 "mov (%%rsp), %3 \n\t"
1826
1827                 "pop  %%rcx; pop  %%r15; pop  %%r14; pop  %%r13; pop  %%r12;"
1828                 "pop  %%r11; pop  %%r10; pop  %%r9;  pop  %%r8;"
1829                 "pop  %%rbp; pop  %%rdi; pop  %%rsi;"
1830                 "pop  %%rdx; pop  %%rbx; pop  %%rax \n\t"
1831 #else
1832                 "xchg %3, (%%esp) \n\t"
1833                 "mov %%eax, %c[rax](%3) \n\t"
1834                 "mov %%ebx, %c[rbx](%3) \n\t"
1835                 "pushl (%%esp); popl %c[rcx](%3) \n\t"
1836                 "mov %%edx, %c[rdx](%3) \n\t"
1837                 "mov %%esi, %c[rsi](%3) \n\t"
1838                 "mov %%edi, %c[rdi](%3) \n\t"
1839                 "mov %%ebp, %c[rbp](%3) \n\t"
1840                 "mov %%cr2, %%eax  \n\t"
1841                 "mov %%eax, %c[cr2](%3) \n\t"
1842                 "mov (%%esp), %3 \n\t"
1843
1844                 "pop %%ecx; popa \n\t"
1845 #endif
1846                 "setbe %0 \n\t"
1847                 "popf \n\t"
1848               : "=q" (fail)
1849               : "r"(vcpu->launched), "d"((unsigned long)HOST_RSP),
1850                 "c"(vcpu),
1851                 [rax]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_RAX])),
1852                 [rbx]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_RBX])),
1853                 [rcx]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_RCX])),
1854                 [rdx]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_RDX])),
1855                 [rsi]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_RSI])),
1856                 [rdi]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_RDI])),
1857                 [rbp]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_RBP])),
1858 #ifdef CONFIG_X86_64
1859                 [r8 ]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_R8 ])),
1860                 [r9 ]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_R9 ])),
1861                 [r10]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_R10])),
1862                 [r11]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_R11])),
1863                 [r12]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_R12])),
1864                 [r13]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_R13])),
1865                 [r14]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_R14])),
1866                 [r15]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_R15])),
1867 #endif
1868                 [cr2]"i"(offsetof(struct kvm_vcpu, cr2))
1869               : "cc", "memory" );
1870
1871         ++kvm_stat.exits;
1872
1873         save_msrs(vcpu->guest_msrs, NR_BAD_MSRS);
1874         load_msrs(vcpu->host_msrs, NR_BAD_MSRS);
1875
1876         fx_save(vcpu->guest_fx_image);
1877         fx_restore(vcpu->host_fx_image);
1878         vcpu->interrupt_window_open = (vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & 3) == 0;
1879
1880         asm ("mov %0, %%ds; mov %0, %%es" : : "r"(__USER_DS));
1881
1882         kvm_run->exit_type = 0;
1883         if (fail) {
1884                 kvm_run->exit_type = KVM_EXIT_TYPE_FAIL_ENTRY;
1885                 kvm_run->exit_reason = vmcs_read32(VM_INSTRUCTION_ERROR);
1886                 r = 0;
1887         } else {
1888                 if (fs_gs_ldt_reload_needed) {
1889                         load_ldt(ldt_sel);
1890                         load_fs(fs_sel);
1891                         /*
1892                          * If we have to reload gs, we must take care to
1893                          * preserve our gs base.
1894                          */
1895                         local_irq_disable();
1896                         load_gs(gs_sel);
1897 #ifdef CONFIG_X86_64
1898                         wrmsrl(MSR_GS_BASE, vmcs_readl(HOST_GS_BASE));
1899 #endif
1900                         local_irq_enable();
1901
1902                         reload_tss();
1903                 }
1904                 /*
1905                  * Profile KVM exit RIPs:
1906                  */
1907                 if (unlikely(prof_on == KVM_PROFILING))
1908                         profile_hit(KVM_PROFILING, (void *)vmcs_readl(GUEST_RIP));
1909
1910                 vcpu->launched = 1;
1911                 kvm_run->exit_type = KVM_EXIT_TYPE_VM_EXIT;
1912                 r = kvm_handle_exit(kvm_run, vcpu);
1913                 if (r > 0) {
1914                         /* Give scheduler a change to reschedule. */
1915                         if (signal_pending(current)) {
1916                                 ++kvm_stat.signal_exits;
1917                                 post_kvm_run_save(vcpu, kvm_run);
1918                                 return -EINTR;
1919                         }
1920
1921                         if (dm_request_for_irq_injection(vcpu, kvm_run)) {
1922                                 ++kvm_stat.request_irq_exits;
1923                                 post_kvm_run_save(vcpu, kvm_run);
1924                                 return -EINTR;
1925                         }
1926
1927                         kvm_resched(vcpu);
1928                         goto again;
1929                 }
1930         }
1931
1932         post_kvm_run_save(vcpu, kvm_run);
1933         return r;
1934 }
1935
1936 static void vmx_flush_tlb(struct kvm_vcpu *vcpu)
1937 {
1938         vmcs_writel(GUEST_CR3, vmcs_readl(GUEST_CR3));
1939 }
1940
1941 static void vmx_inject_page_fault(struct kvm_vcpu *vcpu,
1942                                   unsigned long addr,
1943                                   u32 err_code)
1944 {
1945         u32 vect_info = vmcs_read32(IDT_VECTORING_INFO_FIELD);
1946
1947         ++kvm_stat.pf_guest;
1948
1949         if (is_page_fault(vect_info)) {
1950                 printk(KERN_DEBUG "inject_page_fault: "
1951                        "double fault 0x%lx @ 0x%lx\n",
1952                        addr, vmcs_readl(GUEST_RIP));
1953                 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE, 0);
1954                 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
1955                              DF_VECTOR |
1956                              INTR_TYPE_EXCEPTION |
1957                              INTR_INFO_DELIEVER_CODE_MASK |
1958                              INTR_INFO_VALID_MASK);
1959                 return;
1960         }
1961         vcpu->cr2 = addr;
1962         vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE, err_code);
1963         vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
1964                      PF_VECTOR |
1965                      INTR_TYPE_EXCEPTION |
1966                      INTR_INFO_DELIEVER_CODE_MASK |
1967                      INTR_INFO_VALID_MASK);
1968
1969 }
1970
1971 static void vmx_free_vmcs(struct kvm_vcpu *vcpu)
1972 {
1973         if (vcpu->vmcs) {
1974                 on_each_cpu(__vcpu_clear, vcpu, 0, 1);
1975                 free_vmcs(vcpu->vmcs);
1976                 vcpu->vmcs = NULL;
1977         }
1978 }
1979
1980 static void vmx_free_vcpu(struct kvm_vcpu *vcpu)
1981 {
1982         vmx_free_vmcs(vcpu);
1983 }
1984
1985 static int vmx_create_vcpu(struct kvm_vcpu *vcpu)
1986 {
1987         struct vmcs *vmcs;
1988
1989         vcpu->guest_msrs = kmalloc(PAGE_SIZE, GFP_KERNEL);
1990         if (!vcpu->guest_msrs)
1991                 return -ENOMEM;
1992
1993         vcpu->host_msrs = kmalloc(PAGE_SIZE, GFP_KERNEL);
1994         if (!vcpu->host_msrs)
1995                 goto out_free_guest_msrs;
1996
1997         vmcs = alloc_vmcs();
1998         if (!vmcs)
1999                 goto out_free_msrs;
2000
2001         vmcs_clear(vmcs);
2002         vcpu->vmcs = vmcs;
2003         vcpu->launched = 0;
2004
2005         return 0;
2006
2007 out_free_msrs:
2008         kfree(vcpu->host_msrs);
2009         vcpu->host_msrs = NULL;
2010
2011 out_free_guest_msrs:
2012         kfree(vcpu->guest_msrs);
2013         vcpu->guest_msrs = NULL;
2014
2015         return -ENOMEM;
2016 }
2017
2018 static struct kvm_arch_ops vmx_arch_ops = {
2019         .cpu_has_kvm_support = cpu_has_kvm_support,
2020         .disabled_by_bios = vmx_disabled_by_bios,
2021         .hardware_setup = hardware_setup,
2022         .hardware_unsetup = hardware_unsetup,
2023         .hardware_enable = hardware_enable,
2024         .hardware_disable = hardware_disable,
2025
2026         .vcpu_create = vmx_create_vcpu,
2027         .vcpu_free = vmx_free_vcpu,
2028
2029         .vcpu_load = vmx_vcpu_load,
2030         .vcpu_put = vmx_vcpu_put,
2031         .vcpu_decache = vmx_vcpu_decache,
2032
2033         .set_guest_debug = set_guest_debug,
2034         .get_msr = vmx_get_msr,
2035         .set_msr = vmx_set_msr,
2036         .get_segment_base = vmx_get_segment_base,
2037         .get_segment = vmx_get_segment,
2038         .set_segment = vmx_set_segment,
2039         .get_cs_db_l_bits = vmx_get_cs_db_l_bits,
2040         .decache_cr0_cr4_guest_bits = vmx_decache_cr0_cr4_guest_bits,
2041         .set_cr0 = vmx_set_cr0,
2042         .set_cr0_no_modeswitch = vmx_set_cr0_no_modeswitch,
2043         .set_cr3 = vmx_set_cr3,
2044         .set_cr4 = vmx_set_cr4,
2045 #ifdef CONFIG_X86_64
2046         .set_efer = vmx_set_efer,
2047 #endif
2048         .get_idt = vmx_get_idt,
2049         .set_idt = vmx_set_idt,
2050         .get_gdt = vmx_get_gdt,
2051         .set_gdt = vmx_set_gdt,
2052         .cache_regs = vcpu_load_rsp_rip,
2053         .decache_regs = vcpu_put_rsp_rip,
2054         .get_rflags = vmx_get_rflags,
2055         .set_rflags = vmx_set_rflags,
2056
2057         .tlb_flush = vmx_flush_tlb,
2058         .inject_page_fault = vmx_inject_page_fault,
2059
2060         .inject_gp = vmx_inject_gp,
2061
2062         .run = vmx_vcpu_run,
2063         .skip_emulated_instruction = skip_emulated_instruction,
2064         .vcpu_setup = vmx_vcpu_setup,
2065 };
2066
2067 static int __init vmx_init(void)
2068 {
2069         return kvm_init_arch(&vmx_arch_ops, THIS_MODULE);
2070 }
2071
2072 static void __exit vmx_exit(void)
2073 {
2074         kvm_exit_arch();
2075 }
2076
2077 module_init(vmx_init)
2078 module_exit(vmx_exit)