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