arch/x86/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 "irq.h"
  19 #include "vmx.h"
  20 #include "mmu.h"
  21
  22 #include <linux/kvm_host.h>
  23 #include <linux/module.h>
  24 #include <linux/kernel.h>
  25 #include <linux/mm.h>
  26 #include <linux/highmem.h>
  27 #include <linux/sched.h>
  28 #include <linux/moduleparam.h>
  29 #include "kvm_cache_regs.h"
  30 #include "x86.h"
  31
  32 #include <asm/io.h>
  33 #include <asm/desc.h>
  34
  35 #define __ex(x) __kvm_handle_fault_on_reboot(x)
  36
  37 MODULE_AUTHOR("Qumranet");
  38 MODULE_LICENSE("GPL");
  39
  40 static int bypass_guest_pf = 1;
  41 module_param(bypass_guest_pf, bool, 0);
  42
  43 static int enable_vpid = 1;
  44 module_param(enable_vpid, bool, 0);
  45
  46 static int flexpriority_enabled = 1;
  47 module_param(flexpriority_enabled, bool, 0);
  48
  49 static int enable_ept = 1;
  50 module_param(enable_ept, bool, 0);
  51
  52 static int emulate_invalid_guest_state = 0;
  53 module_param(emulate_invalid_guest_state, bool, 0);
  54
  55 struct vmcs {
  56         u32 revision_id;
  57         u32 abort;
  58         char data[0];
  59 };
  60
  61 struct vcpu_vmx {
  62         struct kvm_vcpu       vcpu;
  63         struct list_head      local_vcpus_link;
  64         unsigned long         host_rsp;
  65         int                   launched;
  66         u8                    fail;
  67         u32                   idt_vectoring_info;
  68         struct kvm_msr_entry *guest_msrs;
  69         struct kvm_msr_entry *host_msrs;
  70         int                   nmsrs;
  71         int                   save_nmsrs;
  72         int                   msr_offset_efer;
  73 #ifdef CONFIG_X86_64
  74         int                   msr_offset_kernel_gs_base;
  75 #endif
  76         struct vmcs          *vmcs;
  77         struct {
  78                 int           loaded;
  79                 u16           fs_sel, gs_sel, ldt_sel;
  80                 int           gs_ldt_reload_needed;
  81                 int           fs_reload_needed;
  82                 int           guest_efer_loaded;
  83         } host_state;
  84         struct {
  85                 struct {
  86                         bool pending;
  87                         u8 vector;
  88                         unsigned rip;
  89                 } irq;
  90         } rmode;
  91         int vpid;
  92         bool emulation_required;
  93 };
  94
  95 static inline struct vcpu_vmx *to_vmx(struct kvm_vcpu *vcpu)
  96 {
  97         return container_of(vcpu, struct vcpu_vmx, vcpu);
  98 }
  99
 100 static int init_rmode(struct kvm *kvm);
 101 static u64 construct_eptp(unsigned long root_hpa);
 102
 103 static DEFINE_PER_CPU(struct vmcs *, vmxarea);
 104 static DEFINE_PER_CPU(struct vmcs *, current_vmcs);
 105 static DEFINE_PER_CPU(struct list_head, vcpus_on_cpu);
 106
 107 static struct page *vmx_io_bitmap_a;
 108 static struct page *vmx_io_bitmap_b;
 109 static struct page *vmx_msr_bitmap;
 110
 111 static DECLARE_BITMAP(vmx_vpid_bitmap, VMX_NR_VPIDS);
 112 static DEFINE_SPINLOCK(vmx_vpid_lock);
 113
 114 static struct vmcs_config {
 115         int size;
 116         int order;
 117         u32 revision_id;
 118         u32 pin_based_exec_ctrl;
 119         u32 cpu_based_exec_ctrl;
 120         u32 cpu_based_2nd_exec_ctrl;
 121         u32 vmexit_ctrl;
 122         u32 vmentry_ctrl;
 123 } vmcs_config;
 124
 125 struct vmx_capability {
 126         u32 ept;
 127         u32 vpid;
 128 } vmx_capability;
 129
 130 #define VMX_SEGMENT_FIELD(seg)                                  \
 131         [VCPU_SREG_##seg] = {                                   \
 132                 .selector = GUEST_##seg##_SELECTOR,             \
 133                 .base = GUEST_##seg##_BASE,                     \
 134                 .limit = GUEST_##seg##_LIMIT,                   \
 135                 .ar_bytes = GUEST_##seg##_AR_BYTES,             \
 136         }
 137
 138 static struct kvm_vmx_segment_field {
 139         unsigned selector;
 140         unsigned base;
 141         unsigned limit;
 142         unsigned ar_bytes;
 143 } kvm_vmx_segment_fields[] = {
 144         VMX_SEGMENT_FIELD(CS),
 145         VMX_SEGMENT_FIELD(DS),
 146         VMX_SEGMENT_FIELD(ES),
 147         VMX_SEGMENT_FIELD(FS),
 148         VMX_SEGMENT_FIELD(GS),
 149         VMX_SEGMENT_FIELD(SS),
 150         VMX_SEGMENT_FIELD(TR),
 151         VMX_SEGMENT_FIELD(LDTR),
 152 };
 153
 154 /*
 155  * Keep MSR_K6_STAR at the end, as setup_msrs() will try to optimize it
 156  * away by decrementing the array size.
 157  */
 158 static const u32 vmx_msr_index[] = {
 159 #ifdef CONFIG_X86_64
 160         MSR_SYSCALL_MASK, MSR_LSTAR, MSR_CSTAR, MSR_KERNEL_GS_BASE,
 161 #endif
 162         MSR_EFER, MSR_K6_STAR,
 163 };
 164 #define NR_VMX_MSR ARRAY_SIZE(vmx_msr_index)
 165
 166 static void load_msrs(struct kvm_msr_entry *e, int n)
 167 {
 168         int i;
 169
 170         for (i = 0; i < n; ++i)
 171                 wrmsrl(e[i].index, e[i].data);
 172 }
 173
 174 static void save_msrs(struct kvm_msr_entry *e, int n)
 175 {
 176         int i;
 177
 178         for (i = 0; i < n; ++i)
 179                 rdmsrl(e[i].index, e[i].data);
 180 }
 181
 182 static inline int is_page_fault(u32 intr_info)
 183 {
 184         return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
 185                              INTR_INFO_VALID_MASK)) ==
 186                 (INTR_TYPE_EXCEPTION | PF_VECTOR | INTR_INFO_VALID_MASK);
 187 }
 188
 189 static inline int is_no_device(u32 intr_info)
 190 {
 191         return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
 192                              INTR_INFO_VALID_MASK)) ==
 193                 (INTR_TYPE_EXCEPTION | NM_VECTOR | INTR_INFO_VALID_MASK);
 194 }
 195
 196 static inline int is_invalid_opcode(u32 intr_info)
 197 {
 198         return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
 199                              INTR_INFO_VALID_MASK)) ==
 200                 (INTR_TYPE_EXCEPTION | UD_VECTOR | INTR_INFO_VALID_MASK);
 201 }
 202
 203 static inline int is_external_interrupt(u32 intr_info)
 204 {
 205         return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VALID_MASK))
 206                 == (INTR_TYPE_EXT_INTR | INTR_INFO_VALID_MASK);
 207 }
 208
 209 static inline int cpu_has_vmx_msr_bitmap(void)
 210 {
 211         return (vmcs_config.cpu_based_exec_ctrl & CPU_BASED_USE_MSR_BITMAPS);
 212 }
 213
 214 static inline int cpu_has_vmx_tpr_shadow(void)
 215 {
 216         return (vmcs_config.cpu_based_exec_ctrl & CPU_BASED_TPR_SHADOW);
 217 }
 218
 219 static inline int vm_need_tpr_shadow(struct kvm *kvm)
 220 {
 221         return ((cpu_has_vmx_tpr_shadow()) && (irqchip_in_kernel(kvm)));
 222 }
 223
 224 static inline int cpu_has_secondary_exec_ctrls(void)
 225 {
 226         return (vmcs_config.cpu_based_exec_ctrl &
 227                 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS);
 228 }
 229
 230 static inline bool cpu_has_vmx_virtualize_apic_accesses(void)
 231 {
 232         return flexpriority_enabled
 233                 && (vmcs_config.cpu_based_2nd_exec_ctrl &
 234                     SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES);
 235 }
 236
 237 static inline int cpu_has_vmx_invept_individual_addr(void)
 238 {
 239         return (!!(vmx_capability.ept & VMX_EPT_EXTENT_INDIVIDUAL_BIT));
 240 }
 241
 242 static inline int cpu_has_vmx_invept_context(void)
 243 {
 244         return (!!(vmx_capability.ept & VMX_EPT_EXTENT_CONTEXT_BIT));
 245 }
 246
 247 static inline int cpu_has_vmx_invept_global(void)
 248 {
 249         return (!!(vmx_capability.ept & VMX_EPT_EXTENT_GLOBAL_BIT));
 250 }
 251
 252 static inline int cpu_has_vmx_ept(void)
 253 {
 254         return (vmcs_config.cpu_based_2nd_exec_ctrl &
 255                 SECONDARY_EXEC_ENABLE_EPT);
 256 }
 257
 258 static inline int vm_need_ept(void)
 259 {
 260         return (cpu_has_vmx_ept() && enable_ept);
 261 }
 262
 263 static inline int vm_need_virtualize_apic_accesses(struct kvm *kvm)
 264 {
 265         return ((cpu_has_vmx_virtualize_apic_accesses()) &&
 266                 (irqchip_in_kernel(kvm)));
 267 }
 268
 269 static inline int cpu_has_vmx_vpid(void)
 270 {
 271         return (vmcs_config.cpu_based_2nd_exec_ctrl &
 272                 SECONDARY_EXEC_ENABLE_VPID);
 273 }
 274
 275 static inline int cpu_has_virtual_nmis(void)
 276 {
 277         return vmcs_config.pin_based_exec_ctrl & PIN_BASED_VIRTUAL_NMIS;
 278 }
 279
 280 static int __find_msr_index(struct vcpu_vmx *vmx, u32 msr)
 281 {
 282         int i;
 283
 284         for (i = 0; i < vmx->nmsrs; ++i)
 285                 if (vmx->guest_msrs[i].index == msr)
 286                         return i;
 287         return -1;
 288 }
 289
 290 static inline void __invvpid(int ext, u16 vpid, gva_t gva)
 291 {
 292     struct {
 293         u64 vpid : 16;
 294         u64 rsvd : 48;
 295         u64 gva;
 296     } operand = { vpid, 0, gva };
 297
 298     asm volatile (__ex(ASM_VMX_INVVPID)
 299                   /* CF==1 or ZF==1 --> rc = -1 */
 300                   "; ja 1f ; ud2 ; 1:"
 301                   : : "a"(&operand), "c"(ext) : "cc", "memory");
 302 }
 303
 304 static inline void __invept(int ext, u64 eptp, gpa_t gpa)
 305 {
 306         struct {
 307                 u64 eptp, gpa;
 308         } operand = {eptp, gpa};
 309
 310         asm volatile (__ex(ASM_VMX_INVEPT)
 311                         /* CF==1 or ZF==1 --> rc = -1 */
 312                         "; ja 1f ; ud2 ; 1:\n"
 313                         : : "a" (&operand), "c" (ext) : "cc", "memory");
 314 }
 315
 316 static struct kvm_msr_entry *find_msr_entry(struct vcpu_vmx *vmx, u32 msr)
 317 {
 318         int i;
 319
 320         i = __find_msr_index(vmx, msr);
 321         if (i >= 0)
 322                 return &vmx->guest_msrs[i];
 323         return NULL;
 324 }
 325
 326 static void vmcs_clear(struct vmcs *vmcs)
 327 {
 328         u64 phys_addr = __pa(vmcs);
 329         u8 error;
 330
 331         asm volatile (__ex(ASM_VMX_VMCLEAR_RAX) "; setna %0"
 332                       : "=g"(error) : "a"(&phys_addr), "m"(phys_addr)
 333                       : "cc", "memory");
 334         if (error)
 335                 printk(KERN_ERR "kvm: vmclear fail: %p/%llx\n",
 336                        vmcs, phys_addr);
 337 }
 338
 339 static void __vcpu_clear(void *arg)
 340 {
 341         struct vcpu_vmx *vmx = arg;
 342         int cpu = raw_smp_processor_id();
 343
 344         if (vmx->vcpu.cpu == cpu)
 345                 vmcs_clear(vmx->vmcs);
 346         if (per_cpu(current_vmcs, cpu) == vmx->vmcs)
 347                 per_cpu(current_vmcs, cpu) = NULL;
 348         rdtscll(vmx->vcpu.arch.host_tsc);
 349         list_del(&vmx->local_vcpus_link);
 350         vmx->vcpu.cpu = -1;
 351         vmx->launched = 0;
 352 }
 353
 354 static void vcpu_clear(struct vcpu_vmx *vmx)
 355 {
 356         if (vmx->vcpu.cpu == -1)
 357                 return;
 358         smp_call_function_single(vmx->vcpu.cpu, __vcpu_clear, vmx, 1);
 359 }
 360
 361 static inline void vpid_sync_vcpu_all(struct vcpu_vmx *vmx)
 362 {
 363         if (vmx->vpid == 0)
 364                 return;
 365
 366         __invvpid(VMX_VPID_EXTENT_SINGLE_CONTEXT, vmx->vpid, 0);
 367 }
 368
 369 static inline void ept_sync_global(void)
 370 {
 371         if (cpu_has_vmx_invept_global())
 372                 __invept(VMX_EPT_EXTENT_GLOBAL, 0, 0);
 373 }
 374
 375 static inline void ept_sync_context(u64 eptp)
 376 {
 377         if (vm_need_ept()) {
 378                 if (cpu_has_vmx_invept_context())
 379                         __invept(VMX_EPT_EXTENT_CONTEXT, eptp, 0);
 380                 else
 381                         ept_sync_global();
 382         }
 383 }
 384
 385 static inline void ept_sync_individual_addr(u64 eptp, gpa_t gpa)
 386 {
 387         if (vm_need_ept()) {
 388                 if (cpu_has_vmx_invept_individual_addr())
 389                         __invept(VMX_EPT_EXTENT_INDIVIDUAL_ADDR,
 390                                         eptp, gpa);
 391                 else
 392                         ept_sync_context(eptp);
 393         }
 394 }
 395
 396 static unsigned long vmcs_readl(unsigned long field)
 397 {
 398         unsigned long value;
 399
 400         asm volatile (__ex(ASM_VMX_VMREAD_RDX_RAX)
 401                       : "=a"(value) : "d"(field) : "cc");
 402         return value;
 403 }
 404
 405 static u16 vmcs_read16(unsigned long field)
 406 {
 407         return vmcs_readl(field);
 408 }
 409
 410 static u32 vmcs_read32(unsigned long field)
 411 {
 412         return vmcs_readl(field);
 413 }
 414
 415 static u64 vmcs_read64(unsigned long field)
 416 {
 417 #ifdef CONFIG_X86_64
 418         return vmcs_readl(field);
 419 #else
 420         return vmcs_readl(field) | ((u64)vmcs_readl(field+1) << 32);
 421 #endif
 422 }
 423
 424 static noinline void vmwrite_error(unsigned long field, unsigned long value)
 425 {
 426         printk(KERN_ERR "vmwrite error: reg %lx value %lx (err %d)\n",
 427                field, value, vmcs_read32(VM_INSTRUCTION_ERROR));
 428         dump_stack();
 429 }
 430
 431 static void vmcs_writel(unsigned long field, unsigned long value)
 432 {
 433         u8 error;
 434
 435         asm volatile (__ex(ASM_VMX_VMWRITE_RAX_RDX) "; setna %0"
 436                        : "=q"(error) : "a"(value), "d"(field) : "cc");
 437         if (unlikely(error))
 438                 vmwrite_error(field, value);
 439 }
 440
 441 static void vmcs_write16(unsigned long field, u16 value)
 442 {
 443         vmcs_writel(field, value);
 444 }
 445
 446 static void vmcs_write32(unsigned long field, u32 value)
 447 {
 448         vmcs_writel(field, value);
 449 }
 450
 451 static void vmcs_write64(unsigned long field, u64 value)
 452 {
 453         vmcs_writel(field, value);
 454 #ifndef CONFIG_X86_64
 455         asm volatile ("");
 456         vmcs_writel(field+1, value >> 32);
 457 #endif
 458 }
 459
 460 static void vmcs_clear_bits(unsigned long field, u32 mask)
 461 {
 462         vmcs_writel(field, vmcs_readl(field) & ~mask);
 463 }
 464
 465 static void vmcs_set_bits(unsigned long field, u32 mask)
 466 {
 467         vmcs_writel(field, vmcs_readl(field) | mask);
 468 }
 469
 470 static void update_exception_bitmap(struct kvm_vcpu *vcpu)
 471 {
 472         u32 eb;
 473
 474         eb = (1u << PF_VECTOR) | (1u << UD_VECTOR);
 475         if (!vcpu->fpu_active)
 476                 eb |= 1u << NM_VECTOR;
 477         if (vcpu->guest_debug.enabled)
 478                 eb |= 1u << DB_VECTOR;
 479         if (vcpu->arch.rmode.active)
 480                 eb = ~0;
 481         if (vm_need_ept())
 482                 eb &= ~(1u << PF_VECTOR); /* bypass_guest_pf = 0 */
 483         vmcs_write32(EXCEPTION_BITMAP, eb);
 484 }
 485
 486 static void reload_tss(void)
 487 {
 488         /*
 489          * VT restores TR but not its size.  Useless.
 490          */
 491         struct descriptor_table gdt;
 492         struct desc_struct *descs;
 493
 494         kvm_get_gdt(&gdt);
 495         descs = (void *)gdt.base;
 496         descs[GDT_ENTRY_TSS].type = 9; /* available TSS */
 497         load_TR_desc();
 498 }
 499
 500 static void load_transition_efer(struct vcpu_vmx *vmx)
 501 {
 502         int efer_offset = vmx->msr_offset_efer;
 503         u64 host_efer = vmx->host_msrs[efer_offset].data;
 504         u64 guest_efer = vmx->guest_msrs[efer_offset].data;
 505         u64 ignore_bits;
 506
 507         if (efer_offset < 0)
 508                 return;
 509         /*
 510          * NX is emulated; LMA and LME handled by hardware; SCE meaninless
 511          * outside long mode
 512          */
 513         ignore_bits = EFER_NX | EFER_SCE;
 514 #ifdef CONFIG_X86_64
 515         ignore_bits |= EFER_LMA | EFER_LME;
 516         /* SCE is meaningful only in long mode on Intel */
 517         if (guest_efer & EFER_LMA)
 518                 ignore_bits &= ~(u64)EFER_SCE;
 519 #endif
 520         if ((guest_efer & ~ignore_bits) == (host_efer & ~ignore_bits))
 521                 return;
 522
 523         vmx->host_state.guest_efer_loaded = 1;
 524         guest_efer &= ~ignore_bits;
 525         guest_efer |= host_efer & ignore_bits;
 526         wrmsrl(MSR_EFER, guest_efer);
 527         vmx->vcpu.stat.efer_reload++;
 528 }
 529
 530 static void reload_host_efer(struct vcpu_vmx *vmx)
 531 {
 532         if (vmx->host_state.guest_efer_loaded) {
 533                 vmx->host_state.guest_efer_loaded = 0;
 534                 load_msrs(vmx->host_msrs + vmx->msr_offset_efer, 1);
 535         }
 536 }
 537
 538 static void vmx_save_host_state(struct kvm_vcpu *vcpu)
 539 {
 540         struct vcpu_vmx *vmx = to_vmx(vcpu);
 541
 542         if (vmx->host_state.loaded)
 543                 return;
 544
 545         vmx->host_state.loaded = 1;
 546         /*
 547          * Set host fs and gs selectors.  Unfortunately, 22.2.3 does not
 548          * allow segment selectors with cpl > 0 or ti == 1.
 549          */
 550         vmx->host_state.ldt_sel = kvm_read_ldt();
 551         vmx->host_state.gs_ldt_reload_needed = vmx->host_state.ldt_sel;
 552         vmx->host_state.fs_sel = kvm_read_fs();
 553         if (!(vmx->host_state.fs_sel & 7)) {
 554                 vmcs_write16(HOST_FS_SELECTOR, vmx->host_state.fs_sel);
 555                 vmx->host_state.fs_reload_needed = 0;
 556         } else {
 557                 vmcs_write16(HOST_FS_SELECTOR, 0);
 558                 vmx->host_state.fs_reload_needed = 1;
 559         }
 560         vmx->host_state.gs_sel = kvm_read_gs();
 561         if (!(vmx->host_state.gs_sel & 7))
 562                 vmcs_write16(HOST_GS_SELECTOR, vmx->host_state.gs_sel);
 563         else {
 564                 vmcs_write16(HOST_GS_SELECTOR, 0);
 565                 vmx->host_state.gs_ldt_reload_needed = 1;
 566         }
 567
 568 #ifdef CONFIG_X86_64
 569         vmcs_writel(HOST_FS_BASE, read_msr(MSR_FS_BASE));
 570         vmcs_writel(HOST_GS_BASE, read_msr(MSR_GS_BASE));
 571 #else
 572         vmcs_writel(HOST_FS_BASE, segment_base(vmx->host_state.fs_sel));
 573         vmcs_writel(HOST_GS_BASE, segment_base(vmx->host_state.gs_sel));
 574 #endif
 575
 576 #ifdef CONFIG_X86_64
 577         if (is_long_mode(&vmx->vcpu))
 578                 save_msrs(vmx->host_msrs +
 579                           vmx->msr_offset_kernel_gs_base, 1);
 580
 581 #endif
 582         load_msrs(vmx->guest_msrs, vmx->save_nmsrs);
 583         load_transition_efer(vmx);
 584 }
 585
 586 static void __vmx_load_host_state(struct vcpu_vmx *vmx)
 587 {
 588         unsigned long flags;
 589
 590         if (!vmx->host_state.loaded)
 591                 return;
 592
 593         ++vmx->vcpu.stat.host_state_reload;
 594         vmx->host_state.loaded = 0;
 595         if (vmx->host_state.fs_reload_needed)
 596                 kvm_load_fs(vmx->host_state.fs_sel);
 597         if (vmx->host_state.gs_ldt_reload_needed) {
 598                 kvm_load_ldt(vmx->host_state.ldt_sel);
 599                 /*
 600                  * If we have to reload gs, we must take care to
 601                  * preserve our gs base.
 602                  */
 603                 local_irq_save(flags);
 604                 kvm_load_gs(vmx->host_state.gs_sel);
 605 #ifdef CONFIG_X86_64
 606                 wrmsrl(MSR_GS_BASE, vmcs_readl(HOST_GS_BASE));
 607 #endif
 608                 local_irq_restore(flags);
 609         }
 610         reload_tss();
 611         save_msrs(vmx->guest_msrs, vmx->save_nmsrs);
 612         load_msrs(vmx->host_msrs, vmx->save_nmsrs);
 613         reload_host_efer(vmx);
 614 }
 615
 616 static void vmx_load_host_state(struct vcpu_vmx *vmx)
 617 {
 618         preempt_disable();
 619         __vmx_load_host_state(vmx);
 620         preempt_enable();
 621 }
 622
 623 /*
 624  * Switches to specified vcpu, until a matching vcpu_put(), but assumes
 625  * vcpu mutex is already taken.
 626  */
 627 static void vmx_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
 628 {
 629         struct vcpu_vmx *vmx = to_vmx(vcpu);
 630         u64 phys_addr = __pa(vmx->vmcs);
 631         u64 tsc_this, delta, new_offset;
 632
 633         if (vcpu->cpu != cpu) {
 634                 vcpu_clear(vmx);
 635                 kvm_migrate_timers(vcpu);
 636                 vpid_sync_vcpu_all(vmx);
 637                 local_irq_disable();
 638                 list_add(&vmx->local_vcpus_link,
 639                          &per_cpu(vcpus_on_cpu, cpu));
 640                 local_irq_enable();
 641         }
 642
 643         if (per_cpu(current_vmcs, cpu) != vmx->vmcs) {
 644                 u8 error;
 645
 646                 per_cpu(current_vmcs, cpu) = vmx->vmcs;
 647                 asm volatile (__ex(ASM_VMX_VMPTRLD_RAX) "; setna %0"
 648                               : "=g"(error) : "a"(&phys_addr), "m"(phys_addr)
 649                               : "cc");
 650                 if (error)
 651                         printk(KERN_ERR "kvm: vmptrld %p/%llx fail\n",
 652                                vmx->vmcs, phys_addr);
 653         }
 654
 655         if (vcpu->cpu != cpu) {
 656                 struct descriptor_table dt;
 657                 unsigned long sysenter_esp;
 658
 659                 vcpu->cpu = cpu;
 660                 /*
 661                  * Linux uses per-cpu TSS and GDT, so set these when switching
 662                  * processors.
 663                  */
 664                 vmcs_writel(HOST_TR_BASE, kvm_read_tr_base()); /* 22.2.4 */
 665                 kvm_get_gdt(&dt);
 666                 vmcs_writel(HOST_GDTR_BASE, dt.base);   /* 22.2.4 */
 667
 668                 rdmsrl(MSR_IA32_SYSENTER_ESP, sysenter_esp);
 669                 vmcs_writel(HOST_IA32_SYSENTER_ESP, sysenter_esp); /* 22.2.3 */
 670
 671                 /*
 672                  * Make sure the time stamp counter is monotonous.
 673                  */
 674                 rdtscll(tsc_this);
 675                 if (tsc_this < vcpu->arch.host_tsc) {
 676                         delta = vcpu->arch.host_tsc - tsc_this;
 677                         new_offset = vmcs_read64(TSC_OFFSET) + delta;
 678                         vmcs_write64(TSC_OFFSET, new_offset);
 679                 }
 680         }
 681 }
 682
 683 static void vmx_vcpu_put(struct kvm_vcpu *vcpu)
 684 {
 685         __vmx_load_host_state(to_vmx(vcpu));
 686 }
 687
 688 static void vmx_fpu_activate(struct kvm_vcpu *vcpu)
 689 {
 690         if (vcpu->fpu_active)
 691                 return;
 692         vcpu->fpu_active = 1;
 693         vmcs_clear_bits(GUEST_CR0, X86_CR0_TS);
 694         if (vcpu->arch.cr0 & X86_CR0_TS)
 695                 vmcs_set_bits(GUEST_CR0, X86_CR0_TS);
 696         update_exception_bitmap(vcpu);
 697 }
 698
 699 static void vmx_fpu_deactivate(struct kvm_vcpu *vcpu)
 700 {
 701         if (!vcpu->fpu_active)
 702                 return;
 703         vcpu->fpu_active = 0;
 704         vmcs_set_bits(GUEST_CR0, X86_CR0_TS);
 705         update_exception_bitmap(vcpu);
 706 }
 707
 708 static unsigned long vmx_get_rflags(struct kvm_vcpu *vcpu)
 709 {
 710         return vmcs_readl(GUEST_RFLAGS);
 711 }
 712
 713 static void vmx_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
 714 {
 715         if (vcpu->arch.rmode.active)
 716                 rflags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
 717         vmcs_writel(GUEST_RFLAGS, rflags);
 718 }
 719
 720 static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
 721 {
 722         unsigned long rip;
 723         u32 interruptibility;
 724
 725         rip = kvm_rip_read(vcpu);
 726         rip += vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
 727         kvm_rip_write(vcpu, rip);
 728
 729         /*
 730          * We emulated an instruction, so temporary interrupt blocking
 731          * should be removed, if set.
 732          */
 733         interruptibility = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
 734         if (interruptibility & 3)
 735                 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO,
 736                              interruptibility & ~3);
 737         vcpu->arch.interrupt_window_open = 1;
 738 }
 739
 740 static void vmx_queue_exception(struct kvm_vcpu *vcpu, unsigned nr,
 741                                 bool has_error_code, u32 error_code)
 742 {
 743         struct vcpu_vmx *vmx = to_vmx(vcpu);
 744
 745         if (has_error_code)
 746                 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE, error_code);
 747
 748         if (vcpu->arch.rmode.active) {
 749                 vmx->rmode.irq.pending = true;
 750                 vmx->rmode.irq.vector = nr;
 751                 vmx->rmode.irq.rip = kvm_rip_read(vcpu);
 752                 if (nr == BP_VECTOR)
 753                         vmx->rmode.irq.rip++;
 754                 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
 755                              nr | INTR_TYPE_SOFT_INTR
 756                              | (has_error_code ? INTR_INFO_DELIVER_CODE_MASK : 0)
 757                              | INTR_INFO_VALID_MASK);
 758                 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN, 1);
 759                 kvm_rip_write(vcpu, vmx->rmode.irq.rip - 1);
 760                 return;
 761         }
 762
 763         vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
 764                      nr | INTR_TYPE_EXCEPTION
 765                      | (has_error_code ? INTR_INFO_DELIVER_CODE_MASK : 0)
 766                      | INTR_INFO_VALID_MASK);
 767 }
 768
 769 static bool vmx_exception_injected(struct kvm_vcpu *vcpu)
 770 {
 771         return false;
 772 }
 773
 774 /*
 775  * Swap MSR entry in host/guest MSR entry array.
 776  */
 777 #ifdef CONFIG_X86_64
 778 static void move_msr_up(struct vcpu_vmx *vmx, int from, int to)
 779 {
 780         struct kvm_msr_entry tmp;
 781
 782         tmp = vmx->guest_msrs[to];
 783         vmx->guest_msrs[to] = vmx->guest_msrs[from];
 784         vmx->guest_msrs[from] = tmp;
 785         tmp = vmx->host_msrs[to];
 786         vmx->host_msrs[to] = vmx->host_msrs[from];
 787         vmx->host_msrs[from] = tmp;
 788 }
 789 #endif
 790
 791 /*
 792  * Set up the vmcs to automatically save and restore system
 793  * msrs.  Don't touch the 64-bit msrs if the guest is in legacy
 794  * mode, as fiddling with msrs is very expensive.
 795  */
 796 static void setup_msrs(struct vcpu_vmx *vmx)
 797 {
 798         int save_nmsrs;
 799
 800         vmx_load_host_state(vmx);
 801         save_nmsrs = 0;
 802 #ifdef CONFIG_X86_64
 803         if (is_long_mode(&vmx->vcpu)) {
 804                 int index;
 805
 806                 index = __find_msr_index(vmx, MSR_SYSCALL_MASK);
 807                 if (index >= 0)
 808                         move_msr_up(vmx, index, save_nmsrs++);
 809                 index = __find_msr_index(vmx, MSR_LSTAR);
 810                 if (index >= 0)
 811                         move_msr_up(vmx, index, save_nmsrs++);
 812                 index = __find_msr_index(vmx, MSR_CSTAR);
 813                 if (index >= 0)
 814                         move_msr_up(vmx, index, save_nmsrs++);
 815                 index = __find_msr_index(vmx, MSR_KERNEL_GS_BASE);
 816                 if (index >= 0)
 817                         move_msr_up(vmx, index, save_nmsrs++);
 818                 /*
 819                  * MSR_K6_STAR is only needed on long mode guests, and only
 820                  * if efer.sce is enabled.
 821                  */
 822                 index = __find_msr_index(vmx, MSR_K6_STAR);
 823                 if ((index >= 0) && (vmx->vcpu.arch.shadow_efer & EFER_SCE))
 824                         move_msr_up(vmx, index, save_nmsrs++);
 825         }
 826 #endif
 827         vmx->save_nmsrs = save_nmsrs;
 828
 829 #ifdef CONFIG_X86_64
 830         vmx->msr_offset_kernel_gs_base =
 831                 __find_msr_index(vmx, MSR_KERNEL_GS_BASE);
 832 #endif
 833         vmx->msr_offset_efer = __find_msr_index(vmx, MSR_EFER);
 834 }
 835
 836 /*
 837  * reads and returns guest's timestamp counter "register"
 838  * guest_tsc = host_tsc + tsc_offset    -- 21.3
 839  */
 840 static u64 guest_read_tsc(void)
 841 {
 842         u64 host_tsc, tsc_offset;
 843
 844         rdtscll(host_tsc);
 845         tsc_offset = vmcs_read64(TSC_OFFSET);
 846         return host_tsc + tsc_offset;
 847 }
 848
 849 /*
 850  * writes 'guest_tsc' into guest's timestamp counter "register"
 851  * guest_tsc = host_tsc + tsc_offset ==> tsc_offset = guest_tsc - host_tsc
 852  */
 853 static void guest_write_tsc(u64 guest_tsc)
 854 {
 855         u64 host_tsc;
 856
 857         rdtscll(host_tsc);
 858         vmcs_write64(TSC_OFFSET, guest_tsc - host_tsc);
 859 }
 860
 861 /*
 862  * Reads an msr value (of 'msr_index') into 'pdata'.
 863  * Returns 0 on success, non-0 otherwise.
 864  * Assumes vcpu_load() was already called.
 865  */
 866 static int vmx_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
 867 {
 868         u64 data;
 869         struct kvm_msr_entry *msr;
 870
 871         if (!pdata) {
 872                 printk(KERN_ERR "BUG: get_msr called with NULL pdata\n");
 873                 return -EINVAL;
 874         }
 875
 876         switch (msr_index) {
 877 #ifdef CONFIG_X86_64
 878         case MSR_FS_BASE:
 879                 data = vmcs_readl(GUEST_FS_BASE);
 880                 break;
 881         case MSR_GS_BASE:
 882                 data = vmcs_readl(GUEST_GS_BASE);
 883                 break;
 884         case MSR_EFER:
 885                 return kvm_get_msr_common(vcpu, msr_index, pdata);
 886 #endif
 887         case MSR_IA32_TIME_STAMP_COUNTER:
 888                 data = guest_read_tsc();
 889                 break;
 890         case MSR_IA32_SYSENTER_CS:
 891                 data = vmcs_read32(GUEST_SYSENTER_CS);
 892                 break;
 893         case MSR_IA32_SYSENTER_EIP:
 894                 data = vmcs_readl(GUEST_SYSENTER_EIP);
 895                 break;
 896         case MSR_IA32_SYSENTER_ESP:
 897                 data = vmcs_readl(GUEST_SYSENTER_ESP);
 898                 break;
 899         default:
 900                 msr = find_msr_entry(to_vmx(vcpu), msr_index);
 901                 if (msr) {
 902                         data = msr->data;
 903                         break;
 904                 }
 905                 return kvm_get_msr_common(vcpu, msr_index, pdata);
 906         }
 907
 908         *pdata = data;
 909         return 0;
 910 }
 911
 912 /*
 913  * Writes msr value into into the appropriate "register".
 914  * Returns 0 on success, non-0 otherwise.
 915  * Assumes vcpu_load() was already called.
 916  */
 917 static int vmx_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
 918 {
 919         struct vcpu_vmx *vmx = to_vmx(vcpu);
 920         struct kvm_msr_entry *msr;
 921         int ret = 0;
 922
 923         switch (msr_index) {
 924 #ifdef CONFIG_X86_64
 925         case MSR_EFER:
 926                 vmx_load_host_state(vmx);
 927                 ret = kvm_set_msr_common(vcpu, msr_index, data);
 928                 break;
 929         case MSR_FS_BASE:
 930                 vmcs_writel(GUEST_FS_BASE, data);
 931                 break;
 932         case MSR_GS_BASE:
 933                 vmcs_writel(GUEST_GS_BASE, data);
 934                 break;
 935 #endif
 936         case MSR_IA32_SYSENTER_CS:
 937                 vmcs_write32(GUEST_SYSENTER_CS, data);
 938                 break;
 939         case MSR_IA32_SYSENTER_EIP:
 940                 vmcs_writel(GUEST_SYSENTER_EIP, data);
 941                 break;
 942         case MSR_IA32_SYSENTER_ESP:
 943                 vmcs_writel(GUEST_SYSENTER_ESP, data);
 944                 break;
 945         case MSR_IA32_TIME_STAMP_COUNTER:
 946                 guest_write_tsc(data);
 947                 break;
 948         case MSR_P6_PERFCTR0:
 949         case MSR_P6_PERFCTR1:
 950         case MSR_P6_EVNTSEL0:
 951         case MSR_P6_EVNTSEL1:
 952                 /*
 953                  * Just discard all writes to the performance counters; this
 954                  * should keep both older linux and windows 64-bit guests
 955                  * happy
 956                  */
 957                 pr_unimpl(vcpu, "unimplemented perfctr wrmsr: 0x%x data 0x%llx\n", msr_index, data);
 958
 959                 break;
 960         default:
 961                 vmx_load_host_state(vmx);
 962                 msr = find_msr_entry(vmx, msr_index);
 963                 if (msr) {
 964                         msr->data = data;
 965                         break;
 966                 }
 967                 ret = kvm_set_msr_common(vcpu, msr_index, data);
 968         }
 969
 970         return ret;
 971 }
 972
 973 static void vmx_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
 974 {
 975         __set_bit(reg, (unsigned long *)&vcpu->arch.regs_avail);
 976         switch (reg) {
 977         case VCPU_REGS_RSP:
 978                 vcpu->arch.regs[VCPU_REGS_RSP] = vmcs_readl(GUEST_RSP);
 979                 break;
 980         case VCPU_REGS_RIP:
 981                 vcpu->arch.regs[VCPU_REGS_RIP] = vmcs_readl(GUEST_RIP);
 982                 break;
 983         default:
 984                 break;
 985         }
 986 }
 987
 988 static int set_guest_debug(struct kvm_vcpu *vcpu, struct kvm_debug_guest *dbg)
 989 {
 990         unsigned long dr7 = 0x400;
 991         int old_singlestep;
 992
 993         old_singlestep = vcpu->guest_debug.singlestep;
 994
 995         vcpu->guest_debug.enabled = dbg->enabled;
 996         if (vcpu->guest_debug.enabled) {
 997                 int i;
 998
 999                 dr7 |= 0x200;  /* exact */
1000                 for (i = 0; i < 4; ++i) {
1001                         if (!dbg->breakpoints[i].enabled)
1002                                 continue;
1003                         vcpu->guest_debug.bp[i] = dbg->breakpoints[i].address;
1004                         dr7 |= 2 << (i*2);    /* global enable */
1005                         dr7 |= 0 << (i*4+16); /* execution breakpoint */
1006                 }
1007
1008                 vcpu->guest_debug.singlestep = dbg->singlestep;
1009         } else
1010                 vcpu->guest_debug.singlestep = 0;
1011
1012         if (old_singlestep && !vcpu->guest_debug.singlestep) {
1013                 unsigned long flags;
1014
1015                 flags = vmcs_readl(GUEST_RFLAGS);
1016                 flags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
1017                 vmcs_writel(GUEST_RFLAGS, flags);
1018         }
1019
1020         update_exception_bitmap(vcpu);
1021         vmcs_writel(GUEST_DR7, dr7);
1022
1023         return 0;
1024 }
1025
1026 static int vmx_get_irq(struct kvm_vcpu *vcpu)
1027 {
1028         if (!vcpu->arch.interrupt.pending)
1029                 return -1;
1030         return vcpu->arch.interrupt.nr;
1031 }
1032
1033 static __init int cpu_has_kvm_support(void)
1034 {
1035         unsigned long ecx = cpuid_ecx(1);
1036         return test_bit(5, &ecx); /* CPUID.1:ECX.VMX[bit 5] -> VT */
1037 }
1038
1039 static __init int vmx_disabled_by_bios(void)
1040 {
1041         u64 msr;
1042
1043         rdmsrl(MSR_IA32_FEATURE_CONTROL, msr);
1044         return (msr & (IA32_FEATURE_CONTROL_LOCKED_BIT |
1045                        IA32_FEATURE_CONTROL_VMXON_ENABLED_BIT))
1046             == IA32_FEATURE_CONTROL_LOCKED_BIT;
1047         /* locked but not enabled */
1048 }
1049
1050 static void hardware_enable(void *garbage)
1051 {
1052         int cpu = raw_smp_processor_id();
1053         u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
1054         u64 old;
1055
1056         INIT_LIST_HEAD(&per_cpu(vcpus_on_cpu, cpu));
1057         rdmsrl(MSR_IA32_FEATURE_CONTROL, old);
1058         if ((old & (IA32_FEATURE_CONTROL_LOCKED_BIT |
1059                     IA32_FEATURE_CONTROL_VMXON_ENABLED_BIT))
1060             != (IA32_FEATURE_CONTROL_LOCKED_BIT |
1061                 IA32_FEATURE_CONTROL_VMXON_ENABLED_BIT))
1062                 /* enable and lock */
1063                 wrmsrl(MSR_IA32_FEATURE_CONTROL, old |
1064                        IA32_FEATURE_CONTROL_LOCKED_BIT |
1065                        IA32_FEATURE_CONTROL_VMXON_ENABLED_BIT);
1066         write_cr4(read_cr4() | X86_CR4_VMXE); /* FIXME: not cpu hotplug safe */
1067         asm volatile (ASM_VMX_VMXON_RAX
1068                       : : "a"(&phys_addr), "m"(phys_addr)
1069                       : "memory", "cc");
1070 }
1071
1072 static void vmclear_local_vcpus(void)
1073 {
1074         int cpu = raw_smp_processor_id();
1075         struct vcpu_vmx *vmx, *n;
1076
1077         list_for_each_entry_safe(vmx, n, &per_cpu(vcpus_on_cpu, cpu),
1078                                  local_vcpus_link)
1079                 __vcpu_clear(vmx);
1080 }
1081
1082 static void hardware_disable(void *garbage)
1083 {
1084         vmclear_local_vcpus();
1085         asm volatile (__ex(ASM_VMX_VMXOFF) : : : "cc");
1086         write_cr4(read_cr4() & ~X86_CR4_VMXE);
1087 }
1088
1089 static __init int adjust_vmx_controls(u32 ctl_min, u32 ctl_opt,
1090                                       u32 msr, u32 *result)
1091 {
1092         u32 vmx_msr_low, vmx_msr_high;
1093         u32 ctl = ctl_min | ctl_opt;
1094
1095         rdmsr(msr, vmx_msr_low, vmx_msr_high);
1096
1097         ctl &= vmx_msr_high; /* bit == 0 in high word ==> must be zero */
1098         ctl |= vmx_msr_low;  /* bit == 1 in low word  ==> must be one  */
1099
1100         /* Ensure minimum (required) set of control bits are supported. */
1101         if (ctl_min & ~ctl)
1102                 return -EIO;
1103
1104         *result = ctl;
1105         return 0;
1106 }
1107
1108 static __init int setup_vmcs_config(struct vmcs_config *vmcs_conf)
1109 {
1110         u32 vmx_msr_low, vmx_msr_high;
1111         u32 min, opt, min2, opt2;
1112         u32 _pin_based_exec_control = 0;
1113         u32 _cpu_based_exec_control = 0;
1114         u32 _cpu_based_2nd_exec_control = 0;
1115         u32 _vmexit_control = 0;
1116         u32 _vmentry_control = 0;
1117
1118         min = PIN_BASED_EXT_INTR_MASK | PIN_BASED_NMI_EXITING;
1119         opt = PIN_BASED_VIRTUAL_NMIS;
1120         if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PINBASED_CTLS,
1121                                 &_pin_based_exec_control) < 0)
1122                 return -EIO;
1123
1124         min = CPU_BASED_HLT_EXITING |
1125 #ifdef CONFIG_X86_64
1126               CPU_BASED_CR8_LOAD_EXITING |
1127               CPU_BASED_CR8_STORE_EXITING |
1128 #endif
1129               CPU_BASED_CR3_LOAD_EXITING |
1130               CPU_BASED_CR3_STORE_EXITING |
1131               CPU_BASED_USE_IO_BITMAPS |
1132               CPU_BASED_MOV_DR_EXITING |
1133               CPU_BASED_USE_TSC_OFFSETING;
1134         opt = CPU_BASED_TPR_SHADOW |
1135               CPU_BASED_USE_MSR_BITMAPS |
1136               CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
1137         if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PROCBASED_CTLS,
1138                                 &_cpu_based_exec_control) < 0)
1139                 return -EIO;
1140 #ifdef CONFIG_X86_64
1141         if ((_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
1142                 _cpu_based_exec_control &= ~CPU_BASED_CR8_LOAD_EXITING &
1143                                            ~CPU_BASED_CR8_STORE_EXITING;
1144 #endif
1145         if (_cpu_based_exec_control & CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) {
1146                 min2 = 0;
1147                 opt2 = SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
1148                         SECONDARY_EXEC_WBINVD_EXITING |
1149                         SECONDARY_EXEC_ENABLE_VPID |
1150                         SECONDARY_EXEC_ENABLE_EPT;
1151                 if (adjust_vmx_controls(min2, opt2,
1152                                         MSR_IA32_VMX_PROCBASED_CTLS2,
1153                                         &_cpu_based_2nd_exec_control) < 0)
1154                         return -EIO;
1155         }
1156 #ifndef CONFIG_X86_64
1157         if (!(_cpu_based_2nd_exec_control &
1158                                 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
1159                 _cpu_based_exec_control &= ~CPU_BASED_TPR_SHADOW;
1160 #endif
1161         if (_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_EPT) {
1162                 /* CR3 accesses don't need to cause VM Exits when EPT enabled */
1163                 min &= ~(CPU_BASED_CR3_LOAD_EXITING |
1164                          CPU_BASED_CR3_STORE_EXITING);
1165                 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PROCBASED_CTLS,
1166                                         &_cpu_based_exec_control) < 0)
1167                         return -EIO;
1168                 rdmsr(MSR_IA32_VMX_EPT_VPID_CAP,
1169                       vmx_capability.ept, vmx_capability.vpid);
1170         }
1171
1172         min = 0;
1173 #ifdef CONFIG_X86_64
1174         min |= VM_EXIT_HOST_ADDR_SPACE_SIZE;
1175 #endif
1176         opt = 0;
1177         if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_EXIT_CTLS,
1178                                 &_vmexit_control) < 0)
1179                 return -EIO;
1180
1181         min = opt = 0;
1182         if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_ENTRY_CTLS,
1183                                 &_vmentry_control) < 0)
1184                 return -EIO;
1185
1186         rdmsr(MSR_IA32_VMX_BASIC, vmx_msr_low, vmx_msr_high);
1187
1188         /* IA-32 SDM Vol 3B: VMCS size is never greater than 4kB. */
1189         if ((vmx_msr_high & 0x1fff) > PAGE_SIZE)
1190                 return -EIO;
1191
1192 #ifdef CONFIG_X86_64
1193         /* IA-32 SDM Vol 3B: 64-bit CPUs always have VMX_BASIC_MSR[48]==0. */
1194         if (vmx_msr_high & (1u<<16))
1195                 return -EIO;
1196 #endif
1197
1198         /* Require Write-Back (WB) memory type for VMCS accesses. */
1199         if (((vmx_msr_high >> 18) & 15) != 6)
1200                 return -EIO;
1201
1202         vmcs_conf->size = vmx_msr_high & 0x1fff;
1203         vmcs_conf->order = get_order(vmcs_config.size);
1204         vmcs_conf->revision_id = vmx_msr_low;
1205
1206         vmcs_conf->pin_based_exec_ctrl = _pin_based_exec_control;
1207         vmcs_conf->cpu_based_exec_ctrl = _cpu_based_exec_control;
1208         vmcs_conf->cpu_based_2nd_exec_ctrl = _cpu_based_2nd_exec_control;
1209         vmcs_conf->vmexit_ctrl         = _vmexit_control;
1210         vmcs_conf->vmentry_ctrl        = _vmentry_control;
1211
1212         return 0;
1213 }
1214
1215 static struct vmcs *alloc_vmcs_cpu(int cpu)
1216 {
1217         int node = cpu_to_node(cpu);
1218         struct page *pages;
1219         struct vmcs *vmcs;
1220
1221         pages = alloc_pages_node(node, GFP_KERNEL, vmcs_config.order);
1222         if (!pages)
1223                 return NULL;
1224         vmcs = page_address(pages);
1225         memset(vmcs, 0, vmcs_config.size);
1226         vmcs->revision_id = vmcs_config.revision_id; /* vmcs revision id */
1227         return vmcs;
1228 }
1229
1230 static struct vmcs *alloc_vmcs(void)
1231 {
1232         return alloc_vmcs_cpu(raw_smp_processor_id());
1233 }
1234
1235 static void free_vmcs(struct vmcs *vmcs)
1236 {
1237         free_pages((unsigned long)vmcs, vmcs_config.order);
1238 }
1239
1240 static void free_kvm_area(void)
1241 {
1242         int cpu;
1243
1244         for_each_online_cpu(cpu)
1245                 free_vmcs(per_cpu(vmxarea, cpu));
1246 }
1247
1248 static __init int alloc_kvm_area(void)
1249 {
1250         int cpu;
1251
1252         for_each_online_cpu(cpu) {
1253                 struct vmcs *vmcs;
1254
1255                 vmcs = alloc_vmcs_cpu(cpu);
1256                 if (!vmcs) {
1257                         free_kvm_area();
1258                         return -ENOMEM;
1259                 }
1260
1261                 per_cpu(vmxarea, cpu) = vmcs;
1262         }
1263         return 0;
1264 }
1265
1266 static __init int hardware_setup(void)
1267 {
1268         if (setup_vmcs_config(&vmcs_config) < 0)
1269                 return -EIO;
1270
1271         if (boot_cpu_has(X86_FEATURE_NX))
1272                 kvm_enable_efer_bits(EFER_NX);
1273
1274         return alloc_kvm_area();
1275 }
1276
1277 static __exit void hardware_unsetup(void)
1278 {
1279         free_kvm_area();
1280 }
1281
1282 static void fix_pmode_dataseg(int seg, struct kvm_save_segment *save)
1283 {
1284         struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1285
1286         if (vmcs_readl(sf->base) == save->base && (save->base & AR_S_MASK)) {
1287                 vmcs_write16(sf->selector, save->selector);
1288                 vmcs_writel(sf->base, save->base);
1289                 vmcs_write32(sf->limit, save->limit);
1290                 vmcs_write32(sf->ar_bytes, save->ar);
1291         } else {
1292                 u32 dpl = (vmcs_read16(sf->selector) & SELECTOR_RPL_MASK)
1293                         << AR_DPL_SHIFT;
1294                 vmcs_write32(sf->ar_bytes, 0x93 | dpl);
1295         }
1296 }
1297
1298 static void enter_pmode(struct kvm_vcpu *vcpu)
1299 {
1300         unsigned long flags;
1301         struct vcpu_vmx *vmx = to_vmx(vcpu);
1302
1303         vmx->emulation_required = 1;
1304         vcpu->arch.rmode.active = 0;
1305
1306         vmcs_writel(GUEST_TR_BASE, vcpu->arch.rmode.tr.base);
1307         vmcs_write32(GUEST_TR_LIMIT, vcpu->arch.rmode.tr.limit);
1308         vmcs_write32(GUEST_TR_AR_BYTES, vcpu->arch.rmode.tr.ar);
1309
1310         flags = vmcs_readl(GUEST_RFLAGS);
1311         flags &= ~(X86_EFLAGS_IOPL | X86_EFLAGS_VM);
1312         flags |= (vcpu->arch.rmode.save_iopl << IOPL_SHIFT);
1313         vmcs_writel(GUEST_RFLAGS, flags);
1314
1315         vmcs_writel(GUEST_CR4, (vmcs_readl(GUEST_CR4) & ~X86_CR4_VME) |
1316                         (vmcs_readl(CR4_READ_SHADOW) & X86_CR4_VME));
1317
1318         update_exception_bitmap(vcpu);
1319
1320         if (emulate_invalid_guest_state)
1321                 return;
1322
1323         fix_pmode_dataseg(VCPU_SREG_ES, &vcpu->arch.rmode.es);
1324         fix_pmode_dataseg(VCPU_SREG_DS, &vcpu->arch.rmode.ds);
1325         fix_pmode_dataseg(VCPU_SREG_GS, &vcpu->arch.rmode.gs);
1326         fix_pmode_dataseg(VCPU_SREG_FS, &vcpu->arch.rmode.fs);
1327
1328         vmcs_write16(GUEST_SS_SELECTOR, 0);
1329         vmcs_write32(GUEST_SS_AR_BYTES, 0x93);
1330
1331         vmcs_write16(GUEST_CS_SELECTOR,
1332                      vmcs_read16(GUEST_CS_SELECTOR) & ~SELECTOR_RPL_MASK);
1333         vmcs_write32(GUEST_CS_AR_BYTES, 0x9b);
1334 }
1335
1336 static gva_t rmode_tss_base(struct kvm *kvm)
1337 {
1338         if (!kvm->arch.tss_addr) {
1339                 gfn_t base_gfn = kvm->memslots[0].base_gfn +
1340                                  kvm->memslots[0].npages - 3;
1341                 return base_gfn << PAGE_SHIFT;
1342         }
1343         return kvm->arch.tss_addr;
1344 }
1345
1346 static void fix_rmode_seg(int seg, struct kvm_save_segment *save)
1347 {
1348         struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1349
1350         save->selector = vmcs_read16(sf->selector);
1351         save->base = vmcs_readl(sf->base);
1352         save->limit = vmcs_read32(sf->limit);
1353         save->ar = vmcs_read32(sf->ar_bytes);
1354         vmcs_write16(sf->selector, save->base >> 4);
1355         vmcs_write32(sf->base, save->base & 0xfffff);
1356         vmcs_write32(sf->limit, 0xffff);
1357         vmcs_write32(sf->ar_bytes, 0xf3);
1358 }
1359
1360 static void enter_rmode(struct kvm_vcpu *vcpu)
1361 {
1362         unsigned long flags;
1363         struct vcpu_vmx *vmx = to_vmx(vcpu);
1364
1365         vmx->emulation_required = 1;
1366         vcpu->arch.rmode.active = 1;
1367
1368         vcpu->arch.rmode.tr.base = vmcs_readl(GUEST_TR_BASE);
1369         vmcs_writel(GUEST_TR_BASE, rmode_tss_base(vcpu->kvm));
1370
1371         vcpu->arch.rmode.tr.limit = vmcs_read32(GUEST_TR_LIMIT);
1372         vmcs_write32(GUEST_TR_LIMIT, RMODE_TSS_SIZE - 1);
1373
1374         vcpu->arch.rmode.tr.ar = vmcs_read32(GUEST_TR_AR_BYTES);
1375         vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
1376
1377         flags = vmcs_readl(GUEST_RFLAGS);
1378         vcpu->arch.rmode.save_iopl
1379                 = (flags & X86_EFLAGS_IOPL) >> IOPL_SHIFT;
1380
1381         flags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
1382
1383         vmcs_writel(GUEST_RFLAGS, flags);
1384         vmcs_writel(GUEST_CR4, vmcs_readl(GUEST_CR4) | X86_CR4_VME);
1385         update_exception_bitmap(vcpu);
1386
1387         if (emulate_invalid_guest_state)
1388                 goto continue_rmode;
1389
1390         vmcs_write16(GUEST_SS_SELECTOR, vmcs_readl(GUEST_SS_BASE) >> 4);
1391         vmcs_write32(GUEST_SS_LIMIT, 0xffff);
1392         vmcs_write32(GUEST_SS_AR_BYTES, 0xf3);
1393
1394         vmcs_write32(GUEST_CS_AR_BYTES, 0xf3);
1395         vmcs_write32(GUEST_CS_LIMIT, 0xffff);
1396         if (vmcs_readl(GUEST_CS_BASE) == 0xffff0000)
1397                 vmcs_writel(GUEST_CS_BASE, 0xf0000);
1398         vmcs_write16(GUEST_CS_SELECTOR, vmcs_readl(GUEST_CS_BASE) >> 4);
1399
1400         fix_rmode_seg(VCPU_SREG_ES, &vcpu->arch.rmode.es);
1401         fix_rmode_seg(VCPU_SREG_DS, &vcpu->arch.rmode.ds);
1402         fix_rmode_seg(VCPU_SREG_GS, &vcpu->arch.rmode.gs);
1403         fix_rmode_seg(VCPU_SREG_FS, &vcpu->arch.rmode.fs);
1404
1405 continue_rmode:
1406         kvm_mmu_reset_context(vcpu);
1407         init_rmode(vcpu->kvm);
1408 }
1409
1410 #ifdef CONFIG_X86_64
1411
1412 static void enter_lmode(struct kvm_vcpu *vcpu)
1413 {
1414         u32 guest_tr_ar;
1415
1416         guest_tr_ar = vmcs_read32(GUEST_TR_AR_BYTES);
1417         if ((guest_tr_ar & AR_TYPE_MASK) != AR_TYPE_BUSY_64_TSS) {
1418                 printk(KERN_DEBUG "%s: tss fixup for long mode. \n",
1419                        __func__);
1420                 vmcs_write32(GUEST_TR_AR_BYTES,
1421                              (guest_tr_ar & ~AR_TYPE_MASK)
1422                              | AR_TYPE_BUSY_64_TSS);
1423         }
1424
1425         vcpu->arch.shadow_efer |= EFER_LMA;
1426
1427         find_msr_entry(to_vmx(vcpu), MSR_EFER)->data |= EFER_LMA | EFER_LME;
1428         vmcs_write32(VM_ENTRY_CONTROLS,
1429                      vmcs_read32(VM_ENTRY_CONTROLS)
1430                      | VM_ENTRY_IA32E_MODE);
1431 }
1432
1433 static void exit_lmode(struct kvm_vcpu *vcpu)
1434 {
1435         vcpu->arch.shadow_efer &= ~EFER_LMA;
1436
1437         vmcs_write32(VM_ENTRY_CONTROLS,
1438                      vmcs_read32(VM_ENTRY_CONTROLS)
1439                      & ~VM_ENTRY_IA32E_MODE);
1440 }
1441
1442 #endif
1443
1444 static void vmx_flush_tlb(struct kvm_vcpu *vcpu)
1445 {
1446         vpid_sync_vcpu_all(to_vmx(vcpu));
1447         if (vm_need_ept())
1448                 ept_sync_context(construct_eptp(vcpu->arch.mmu.root_hpa));
1449 }
1450
1451 static void vmx_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
1452 {
1453         vcpu->arch.cr4 &= KVM_GUEST_CR4_MASK;
1454         vcpu->arch.cr4 |= vmcs_readl(GUEST_CR4) & ~KVM_GUEST_CR4_MASK;
1455 }
1456
1457 static void ept_load_pdptrs(struct kvm_vcpu *vcpu)
1458 {
1459         if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
1460                 if (!load_pdptrs(vcpu, vcpu->arch.cr3)) {
1461                         printk(KERN_ERR "EPT: Fail to load pdptrs!\n");
1462                         return;
1463                 }
1464                 vmcs_write64(GUEST_PDPTR0, vcpu->arch.pdptrs[0]);
1465                 vmcs_write64(GUEST_PDPTR1, vcpu->arch.pdptrs[1]);
1466                 vmcs_write64(GUEST_PDPTR2, vcpu->arch.pdptrs[2]);
1467                 vmcs_write64(GUEST_PDPTR3, vcpu->arch.pdptrs[3]);
1468         }
1469 }
1470
1471 static void vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4);
1472
1473 static void ept_update_paging_mode_cr0(unsigned long *hw_cr0,
1474                                         unsigned long cr0,
1475                                         struct kvm_vcpu *vcpu)
1476 {
1477         if (!(cr0 & X86_CR0_PG)) {
1478                 /* From paging/starting to nonpaging */
1479                 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
1480                              vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) |
1481                              (CPU_BASED_CR3_LOAD_EXITING |
1482                               CPU_BASED_CR3_STORE_EXITING));
1483                 vcpu->arch.cr0 = cr0;
1484                 vmx_set_cr4(vcpu, vcpu->arch.cr4);
1485                 *hw_cr0 |= X86_CR0_PE | X86_CR0_PG;
1486                 *hw_cr0 &= ~X86_CR0_WP;
1487         } else if (!is_paging(vcpu)) {
1488                 /* From nonpaging to paging */
1489                 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
1490                              vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) &
1491                              ~(CPU_BASED_CR3_LOAD_EXITING |
1492                                CPU_BASED_CR3_STORE_EXITING));
1493                 vcpu->arch.cr0 = cr0;
1494                 vmx_set_cr4(vcpu, vcpu->arch.cr4);
1495                 if (!(vcpu->arch.cr0 & X86_CR0_WP))
1496                         *hw_cr0 &= ~X86_CR0_WP;
1497         }
1498 }
1499
1500 static void ept_update_paging_mode_cr4(unsigned long *hw_cr4,
1501                                         struct kvm_vcpu *vcpu)
1502 {
1503         if (!is_paging(vcpu)) {
1504                 *hw_cr4 &= ~X86_CR4_PAE;
1505                 *hw_cr4 |= X86_CR4_PSE;
1506         } else if (!(vcpu->arch.cr4 & X86_CR4_PAE))
1507                 *hw_cr4 &= ~X86_CR4_PAE;
1508 }
1509
1510 static void vmx_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
1511 {
1512         unsigned long hw_cr0 = (cr0 & ~KVM_GUEST_CR0_MASK) |
1513                                 KVM_VM_CR0_ALWAYS_ON;
1514
1515         vmx_fpu_deactivate(vcpu);
1516
1517         if (vcpu->arch.rmode.active && (cr0 & X86_CR0_PE))
1518                 enter_pmode(vcpu);
1519
1520         if (!vcpu->arch.rmode.active && !(cr0 & X86_CR0_PE))
1521                 enter_rmode(vcpu);
1522
1523 #ifdef CONFIG_X86_64
1524         if (vcpu->arch.shadow_efer & EFER_LME) {
1525                 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG))
1526                         enter_lmode(vcpu);
1527                 if (is_paging(vcpu) && !(cr0 & X86_CR0_PG))
1528                         exit_lmode(vcpu);
1529         }
1530 #endif
1531
1532         if (vm_need_ept())
1533                 ept_update_paging_mode_cr0(&hw_cr0, cr0, vcpu);
1534
1535         vmcs_writel(CR0_READ_SHADOW, cr0);
1536         vmcs_writel(GUEST_CR0, hw_cr0);
1537         vcpu->arch.cr0 = cr0;
1538
1539         if (!(cr0 & X86_CR0_TS) || !(cr0 & X86_CR0_PE))
1540                 vmx_fpu_activate(vcpu);
1541 }
1542
1543 static u64 construct_eptp(unsigned long root_hpa)
1544 {
1545         u64 eptp;
1546
1547         /* TODO write the value reading from MSR */
1548         eptp = VMX_EPT_DEFAULT_MT |
1549                 VMX_EPT_DEFAULT_GAW << VMX_EPT_GAW_EPTP_SHIFT;
1550         eptp |= (root_hpa & PAGE_MASK);
1551
1552         return eptp;
1553 }
1554
1555 static void vmx_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
1556 {
1557         unsigned long guest_cr3;
1558         u64 eptp;
1559
1560         guest_cr3 = cr3;
1561         if (vm_need_ept()) {
1562                 eptp = construct_eptp(cr3);
1563                 vmcs_write64(EPT_POINTER, eptp);
1564                 ept_sync_context(eptp);
1565                 ept_load_pdptrs(vcpu);
1566                 guest_cr3 = is_paging(vcpu) ? vcpu->arch.cr3 :
1567                         VMX_EPT_IDENTITY_PAGETABLE_ADDR;
1568         }
1569
1570         vmx_flush_tlb(vcpu);
1571         vmcs_writel(GUEST_CR3, guest_cr3);
1572         if (vcpu->arch.cr0 & X86_CR0_PE)
1573                 vmx_fpu_deactivate(vcpu);
1574 }
1575
1576 static void vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
1577 {
1578         unsigned long hw_cr4 = cr4 | (vcpu->arch.rmode.active ?
1579                     KVM_RMODE_VM_CR4_ALWAYS_ON : KVM_PMODE_VM_CR4_ALWAYS_ON);
1580
1581         vcpu->arch.cr4 = cr4;
1582         if (vm_need_ept())
1583                 ept_update_paging_mode_cr4(&hw_cr4, vcpu);
1584
1585         vmcs_writel(CR4_READ_SHADOW, cr4);
1586         vmcs_writel(GUEST_CR4, hw_cr4);
1587 }
1588
1589 static void vmx_set_efer(struct kvm_vcpu *vcpu, u64 efer)
1590 {
1591         struct vcpu_vmx *vmx = to_vmx(vcpu);
1592         struct kvm_msr_entry *msr = find_msr_entry(vmx, MSR_EFER);
1593
1594         vcpu->arch.shadow_efer = efer;
1595         if (!msr)
1596                 return;
1597         if (efer & EFER_LMA) {
1598                 vmcs_write32(VM_ENTRY_CONTROLS,
1599                                      vmcs_read32(VM_ENTRY_CONTROLS) |
1600                                      VM_ENTRY_IA32E_MODE);
1601                 msr->data = efer;
1602
1603         } else {
1604                 vmcs_write32(VM_ENTRY_CONTROLS,
1605                                      vmcs_read32(VM_ENTRY_CONTROLS) &
1606                                      ~VM_ENTRY_IA32E_MODE);
1607
1608                 msr->data = efer & ~EFER_LME;
1609         }
1610         setup_msrs(vmx);
1611 }
1612
1613 static u64 vmx_get_segment_base(struct kvm_vcpu *vcpu, int seg)
1614 {
1615         struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1616
1617         return vmcs_readl(sf->base);
1618 }
1619
1620 static void vmx_get_segment(struct kvm_vcpu *vcpu,
1621                             struct kvm_segment *var, int seg)
1622 {
1623         struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1624         u32 ar;
1625
1626         var->base = vmcs_readl(sf->base);
1627         var->limit = vmcs_read32(sf->limit);
1628         var->selector = vmcs_read16(sf->selector);
1629         ar = vmcs_read32(sf->ar_bytes);
1630         if (ar & AR_UNUSABLE_MASK)
1631                 ar = 0;
1632         var->type = ar & 15;
1633         var->s = (ar >> 4) & 1;
1634         var->dpl = (ar >> 5) & 3;
1635         var->present = (ar >> 7) & 1;
1636         var->avl = (ar >> 12) & 1;
1637         var->l = (ar >> 13) & 1;
1638         var->db = (ar >> 14) & 1;
1639         var->g = (ar >> 15) & 1;
1640         var->unusable = (ar >> 16) & 1;
1641 }
1642
1643 static int vmx_get_cpl(struct kvm_vcpu *vcpu)
1644 {
1645         struct kvm_segment kvm_seg;
1646
1647         if (!(vcpu->arch.cr0 & X86_CR0_PE)) /* if real mode */
1648                 return 0;
1649
1650         if (vmx_get_rflags(vcpu) & X86_EFLAGS_VM) /* if virtual 8086 */
1651                 return 3;
1652
1653         vmx_get_segment(vcpu, &kvm_seg, VCPU_SREG_CS);
1654         return kvm_seg.selector & 3;
1655 }
1656
1657 static u32 vmx_segment_access_rights(struct kvm_segment *var)
1658 {
1659         u32 ar;
1660
1661         if (var->unusable)
1662                 ar = 1 << 16;
1663         else {
1664                 ar = var->type & 15;
1665                 ar |= (var->s & 1) << 4;
1666                 ar |= (var->dpl & 3) << 5;
1667                 ar |= (var->present & 1) << 7;
1668                 ar |= (var->avl & 1) << 12;
1669                 ar |= (var->l & 1) << 13;
1670                 ar |= (var->db & 1) << 14;
1671                 ar |= (var->g & 1) << 15;
1672         }
1673         if (ar == 0) /* a 0 value means unusable */
1674                 ar = AR_UNUSABLE_MASK;
1675
1676         return ar;
1677 }
1678
1679 static void vmx_set_segment(struct kvm_vcpu *vcpu,
1680                             struct kvm_segment *var, int seg)
1681 {
1682         struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1683         u32 ar;
1684
1685         if (vcpu->arch.rmode.active && seg == VCPU_SREG_TR) {
1686                 vcpu->arch.rmode.tr.selector = var->selector;
1687                 vcpu->arch.rmode.tr.base = var->base;
1688                 vcpu->arch.rmode.tr.limit = var->limit;
1689                 vcpu->arch.rmode.tr.ar = vmx_segment_access_rights(var);
1690                 return;
1691         }
1692         vmcs_writel(sf->base, var->base);
1693         vmcs_write32(sf->limit, var->limit);
1694         vmcs_write16(sf->selector, var->selector);
1695         if (vcpu->arch.rmode.active && var->s) {
1696                 /*
1697                  * Hack real-mode segments into vm86 compatibility.
1698                  */
1699                 if (var->base == 0xffff0000 && var->selector == 0xf000)
1700                         vmcs_writel(sf->base, 0xf0000);
1701                 ar = 0xf3;
1702         } else
1703                 ar = vmx_segment_access_rights(var);
1704         vmcs_write32(sf->ar_bytes, ar);
1705 }
1706
1707 static void vmx_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
1708 {
1709         u32 ar = vmcs_read32(GUEST_CS_AR_BYTES);
1710
1711         *db = (ar >> 14) & 1;
1712         *l = (ar >> 13) & 1;
1713 }
1714
1715 static void vmx_get_idt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
1716 {
1717         dt->limit = vmcs_read32(GUEST_IDTR_LIMIT);
1718         dt->base = vmcs_readl(GUEST_IDTR_BASE);
1719 }
1720
1721 static void vmx_set_idt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
1722 {
1723         vmcs_write32(GUEST_IDTR_LIMIT, dt->limit);
1724         vmcs_writel(GUEST_IDTR_BASE, dt->base);
1725 }
1726
1727 static void vmx_get_gdt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
1728 {
1729         dt->limit = vmcs_read32(GUEST_GDTR_LIMIT);
1730         dt->base = vmcs_readl(GUEST_GDTR_BASE);
1731 }
1732
1733 static void vmx_set_gdt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
1734 {
1735         vmcs_write32(GUEST_GDTR_LIMIT, dt->limit);
1736         vmcs_writel(GUEST_GDTR_BASE, dt->base);
1737 }
1738
1739 static bool rmode_segment_valid(struct kvm_vcpu *vcpu, int seg)
1740 {
1741         struct kvm_segment var;
1742         u32 ar;
1743
1744         vmx_get_segment(vcpu, &var, seg);
1745         ar = vmx_segment_access_rights(&var);
1746
1747         if (var.base != (var.selector << 4))
1748                 return false;
1749         if (var.limit != 0xffff)
1750                 return false;
1751         if (ar != 0xf3)
1752                 return false;
1753
1754         return true;
1755 }
1756
1757 static bool code_segment_valid(struct kvm_vcpu *vcpu)
1758 {
1759         struct kvm_segment cs;
1760         unsigned int cs_rpl;
1761
1762         vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
1763         cs_rpl = cs.selector & SELECTOR_RPL_MASK;
1764
1765         if (~cs.type & (AR_TYPE_CODE_MASK|AR_TYPE_ACCESSES_MASK))
1766                 return false;
1767         if (!cs.s)
1768                 return false;
1769         if (!(~cs.type & (AR_TYPE_CODE_MASK|AR_TYPE_WRITEABLE_MASK))) {
1770                 if (cs.dpl > cs_rpl)
1771                         return false;
1772         } else if (cs.type & AR_TYPE_CODE_MASK) {
1773                 if (cs.dpl != cs_rpl)
1774                         return false;
1775         }
1776         if (!cs.present)
1777                 return false;
1778
1779         /* TODO: Add Reserved field check, this'll require a new member in the kvm_segment_field structure */
1780         return true;
1781 }
1782
1783 static bool stack_segment_valid(struct kvm_vcpu *vcpu)
1784 {
1785         struct kvm_segment ss;
1786         unsigned int ss_rpl;
1787
1788         vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
1789         ss_rpl = ss.selector & SELECTOR_RPL_MASK;
1790
1791         if ((ss.type != 3) || (ss.type != 7))
1792                 return false;
1793         if (!ss.s)
1794                 return false;
1795         if (ss.dpl != ss_rpl) /* DPL != RPL */
1796                 return false;
1797         if (!ss.present)
1798                 return false;
1799
1800         return true;
1801 }
1802
1803 static bool data_segment_valid(struct kvm_vcpu *vcpu, int seg)
1804 {
1805         struct kvm_segment var;
1806         unsigned int rpl;
1807
1808         vmx_get_segment(vcpu, &var, seg);
1809         rpl = var.selector & SELECTOR_RPL_MASK;
1810
1811         if (!var.s)
1812                 return false;
1813         if (!var.present)
1814                 return false;
1815         if (~var.type & (AR_TYPE_CODE_MASK|AR_TYPE_WRITEABLE_MASK)) {
1816                 if (var.dpl < rpl) /* DPL < RPL */
1817                         return false;
1818         }
1819
1820         /* TODO: Add other members to kvm_segment_field to allow checking for other access
1821          * rights flags
1822          */
1823         return true;
1824 }
1825
1826 static bool tr_valid(struct kvm_vcpu *vcpu)
1827 {
1828         struct kvm_segment tr;
1829
1830         vmx_get_segment(vcpu, &tr, VCPU_SREG_TR);
1831
1832         if (tr.selector & SELECTOR_TI_MASK)     /* TI = 1 */
1833                 return false;
1834         if ((tr.type != 3) || (tr.type != 11)) /* TODO: Check if guest is in IA32e mode */
1835                 return false;
1836         if (!tr.present)
1837                 return false;
1838
1839         return true;
1840 }
1841
1842 static bool ldtr_valid(struct kvm_vcpu *vcpu)
1843 {
1844         struct kvm_segment ldtr;
1845
1846         vmx_get_segment(vcpu, &ldtr, VCPU_SREG_LDTR);
1847
1848         if (ldtr.selector & SELECTOR_TI_MASK)   /* TI = 1 */
1849                 return false;
1850         if (ldtr.type != 2)
1851                 return false;
1852         if (!ldtr.present)
1853                 return false;
1854
1855         return true;
1856 }
1857
1858 static bool cs_ss_rpl_check(struct kvm_vcpu *vcpu)
1859 {
1860         struct kvm_segment cs, ss;
1861
1862         vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
1863         vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
1864
1865         return ((cs.selector & SELECTOR_RPL_MASK) ==
1866                  (ss.selector & SELECTOR_RPL_MASK));
1867 }
1868
1869 /*
1870  * Check if guest state is valid. Returns true if valid, false if
1871  * not.
1872  * We assume that registers are always usable
1873  */
1874 static bool guest_state_valid(struct kvm_vcpu *vcpu)
1875 {
1876         /* real mode guest state checks */
1877         if (!(vcpu->arch.cr0 & X86_CR0_PE)) {
1878                 if (!rmode_segment_valid(vcpu, VCPU_SREG_CS))
1879                         return false;
1880                 if (!rmode_segment_valid(vcpu, VCPU_SREG_SS))
1881                         return false;
1882                 if (!rmode_segment_valid(vcpu, VCPU_SREG_DS))
1883                         return false;
1884                 if (!rmode_segment_valid(vcpu, VCPU_SREG_ES))
1885                         return false;
1886                 if (!rmode_segment_valid(vcpu, VCPU_SREG_FS))
1887                         return false;
1888                 if (!rmode_segment_valid(vcpu, VCPU_SREG_GS))
1889                         return false;
1890         } else {
1891         /* protected mode guest state checks */
1892                 if (!cs_ss_rpl_check(vcpu))
1893                         return false;
1894                 if (!code_segment_valid(vcpu))
1895                         return false;
1896                 if (!stack_segment_valid(vcpu))
1897                         return false;
1898                 if (!data_segment_valid(vcpu, VCPU_SREG_DS))
1899                         return false;
1900                 if (!data_segment_valid(vcpu, VCPU_SREG_ES))
1901                         return false;
1902                 if (!data_segment_valid(vcpu, VCPU_SREG_FS))
1903                         return false;
1904                 if (!data_segment_valid(vcpu, VCPU_SREG_GS))
1905                         return false;
1906                 if (!tr_valid(vcpu))
1907                         return false;
1908                 if (!ldtr_valid(vcpu))
1909                         return false;
1910         }
1911         /* TODO:
1912          * - Add checks on RIP
1913          * - Add checks on RFLAGS
1914          */
1915
1916         return true;
1917 }
1918
1919 static int init_rmode_tss(struct kvm *kvm)
1920 {
1921         gfn_t fn = rmode_tss_base(kvm) >> PAGE_SHIFT;
1922         u16 data = 0;
1923         int ret = 0;
1924         int r;
1925
1926         r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
1927         if (r < 0)
1928                 goto out;
1929         data = TSS_BASE_SIZE + TSS_REDIRECTION_SIZE;
1930         r = kvm_write_guest_page(kvm, fn++, &data,
1931                         TSS_IOPB_BASE_OFFSET, sizeof(u16));
1932         if (r < 0)
1933                 goto out;
1934         r = kvm_clear_guest_page(kvm, fn++, 0, PAGE_SIZE);
1935         if (r < 0)
1936                 goto out;
1937         r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
1938         if (r < 0)
1939                 goto out;
1940         data = ~0;
1941         r = kvm_write_guest_page(kvm, fn, &data,
1942                                  RMODE_TSS_SIZE - 2 * PAGE_SIZE - 1,
1943                                  sizeof(u8));
1944         if (r < 0)
1945                 goto out;
1946
1947         ret = 1;
1948 out:
1949         return ret;
1950 }
1951
1952 static int init_rmode_identity_map(struct kvm *kvm)
1953 {
1954         int i, r, ret;
1955         pfn_t identity_map_pfn;
1956         u32 tmp;
1957
1958         if (!vm_need_ept())
1959                 return 1;
1960         if (unlikely(!kvm->arch.ept_identity_pagetable)) {
1961                 printk(KERN_ERR "EPT: identity-mapping pagetable "
1962                         "haven't been allocated!\n");
1963                 return 0;
1964         }
1965         if (likely(kvm->arch.ept_identity_pagetable_done))
1966                 return 1;
1967         ret = 0;
1968         identity_map_pfn = VMX_EPT_IDENTITY_PAGETABLE_ADDR >> PAGE_SHIFT;
1969         r = kvm_clear_guest_page(kvm, identity_map_pfn, 0, PAGE_SIZE);
1970         if (r < 0)
1971                 goto out;
1972         /* Set up identity-mapping pagetable for EPT in real mode */
1973         for (i = 0; i < PT32_ENT_PER_PAGE; i++) {
1974                 tmp = (i << 22) + (_PAGE_PRESENT | _PAGE_RW | _PAGE_USER |
1975                         _PAGE_ACCESSED | _PAGE_DIRTY | _PAGE_PSE);
1976                 r = kvm_write_guest_page(kvm, identity_map_pfn,
1977                                 &tmp, i * sizeof(tmp), sizeof(tmp));
1978                 if (r < 0)
1979                         goto out;
1980         }
1981         kvm->arch.ept_identity_pagetable_done = true;
1982         ret = 1;
1983 out:
1984         return ret;
1985 }
1986
1987 static void seg_setup(int seg)
1988 {
1989         struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1990
1991         vmcs_write16(sf->selector, 0);
1992         vmcs_writel(sf->base, 0);
1993         vmcs_write32(sf->limit, 0xffff);
1994         vmcs_write32(sf->ar_bytes, 0xf3);
1995 }
1996
1997 static int alloc_apic_access_page(struct kvm *kvm)
1998 {
1999         struct kvm_userspace_memory_region kvm_userspace_mem;
2000         int r = 0;
2001
2002         down_write(&kvm->slots_lock);
2003         if (kvm->arch.apic_access_page)
2004                 goto out;
2005         kvm_userspace_mem.slot = APIC_ACCESS_PAGE_PRIVATE_MEMSLOT;
2006         kvm_userspace_mem.flags = 0;
2007         kvm_userspace_mem.guest_phys_addr = 0xfee00000ULL;
2008         kvm_userspace_mem.memory_size = PAGE_SIZE;
2009         r = __kvm_set_memory_region(kvm, &kvm_userspace_mem, 0);
2010         if (r)
2011                 goto out;
2012
2013         down_read(&current->mm->mmap_sem);
2014         kvm->arch.apic_access_page = gfn_to_page(kvm, 0xfee00);
2015         up_read(&current->mm->mmap_sem);
2016 out:
2017         up_write(&kvm->slots_lock);
2018         return r;
2019 }
2020
2021 static int alloc_identity_pagetable(struct kvm *kvm)
2022 {
2023         struct kvm_userspace_memory_region kvm_userspace_mem;
2024         int r = 0;
2025
2026         down_write(&kvm->slots_lock);
2027         if (kvm->arch.ept_identity_pagetable)
2028                 goto out;
2029         kvm_userspace_mem.slot = IDENTITY_PAGETABLE_PRIVATE_MEMSLOT;
2030         kvm_userspace_mem.flags = 0;
2031         kvm_userspace_mem.guest_phys_addr = VMX_EPT_IDENTITY_PAGETABLE_ADDR;
2032         kvm_userspace_mem.memory_size = PAGE_SIZE;
2033         r = __kvm_set_memory_region(kvm, &kvm_userspace_mem, 0);
2034         if (r)
2035                 goto out;
2036
2037         down_read(&current->mm->mmap_sem);
2038         kvm->arch.ept_identity_pagetable = gfn_to_page(kvm,
2039                         VMX_EPT_IDENTITY_PAGETABLE_ADDR >> PAGE_SHIFT);
2040         up_read(&current->mm->mmap_sem);
2041 out:
2042         up_write(&kvm->slots_lock);
2043         return r;
2044 }
2045
2046 static void allocate_vpid(struct vcpu_vmx *vmx)
2047 {
2048         int vpid;
2049
2050         vmx->vpid = 0;
2051         if (!enable_vpid || !cpu_has_vmx_vpid())
2052                 return;
2053         spin_lock(&vmx_vpid_lock);
2054         vpid = find_first_zero_bit(vmx_vpid_bitmap, VMX_NR_VPIDS);
2055         if (vpid < VMX_NR_VPIDS) {
2056                 vmx->vpid = vpid;
2057                 __set_bit(vpid, vmx_vpid_bitmap);
2058         }
2059         spin_unlock(&vmx_vpid_lock);
2060 }
2061
2062 static void vmx_disable_intercept_for_msr(struct page *msr_bitmap, u32 msr)
2063 {
2064         void *va;
2065
2066         if (!cpu_has_vmx_msr_bitmap())
2067                 return;
2068
2069         /*
2070          * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
2071          * have the write-low and read-high bitmap offsets the wrong way round.
2072          * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
2073          */
2074         va = kmap(msr_bitmap);
2075         if (msr <= 0x1fff) {
2076                 __clear_bit(msr, va + 0x000); /* read-low */
2077                 __clear_bit(msr, va + 0x800); /* write-low */
2078         } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
2079                 msr &= 0x1fff;
2080                 __clear_bit(msr, va + 0x400); /* read-high */
2081                 __clear_bit(msr, va + 0xc00); /* write-high */
2082         }
2083         kunmap(msr_bitmap);
2084 }
2085
2086 /*
2087  * Sets up the vmcs for emulated real mode.
2088  */
2089 static int vmx_vcpu_setup(struct vcpu_vmx *vmx)
2090 {
2091         u32 host_sysenter_cs;
2092         u32 junk;
2093         unsigned long a;
2094         struct descriptor_table dt;
2095         int i;
2096         unsigned long kvm_vmx_return;
2097         u32 exec_control;
2098
2099         /* I/O */
2100         vmcs_write64(IO_BITMAP_A, page_to_phys(vmx_io_bitmap_a));
2101         vmcs_write64(IO_BITMAP_B, page_to_phys(vmx_io_bitmap_b));
2102
2103         if (cpu_has_vmx_msr_bitmap())
2104                 vmcs_write64(MSR_BITMAP, page_to_phys(vmx_msr_bitmap));
2105
2106         vmcs_write64(VMCS_LINK_POINTER, -1ull); /* 22.3.1.5 */
2107
2108         /* Control */
2109         vmcs_write32(PIN_BASED_VM_EXEC_CONTROL,
2110                 vmcs_config.pin_based_exec_ctrl);
2111
2112         exec_control = vmcs_config.cpu_based_exec_ctrl;
2113         if (!vm_need_tpr_shadow(vmx->vcpu.kvm)) {
2114                 exec_control &= ~CPU_BASED_TPR_SHADOW;
2115 #ifdef CONFIG_X86_64
2116                 exec_control |= CPU_BASED_CR8_STORE_EXITING |
2117                                 CPU_BASED_CR8_LOAD_EXITING;
2118 #endif
2119         }
2120         if (!vm_need_ept())
2121                 exec_control |= CPU_BASED_CR3_STORE_EXITING |
2122                                 CPU_BASED_CR3_LOAD_EXITING;
2123         vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, exec_control);
2124
2125         if (cpu_has_secondary_exec_ctrls()) {
2126                 exec_control = vmcs_config.cpu_based_2nd_exec_ctrl;
2127                 if (!vm_need_virtualize_apic_accesses(vmx->vcpu.kvm))
2128                         exec_control &=
2129                                 ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
2130                 if (vmx->vpid == 0)
2131                         exec_control &= ~SECONDARY_EXEC_ENABLE_VPID;
2132                 if (!vm_need_ept())
2133                         exec_control &= ~SECONDARY_EXEC_ENABLE_EPT;
2134                 vmcs_write32(SECONDARY_VM_EXEC_CONTROL, exec_control);
2135         }
2136
2137         vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, !!bypass_guest_pf);
2138         vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, !!bypass_guest_pf);
2139         vmcs_write32(CR3_TARGET_COUNT, 0);           /* 22.2.1 */
2140
2141         vmcs_writel(HOST_CR0, read_cr0());  /* 22.2.3 */
2142         vmcs_writel(HOST_CR4, read_cr4());  /* 22.2.3, 22.2.5 */
2143         vmcs_writel(HOST_CR3, read_cr3());  /* 22.2.3  FIXME: shadow tables */
2144
2145         vmcs_write16(HOST_CS_SELECTOR, __KERNEL_CS);  /* 22.2.4 */
2146         vmcs_write16(HOST_DS_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
2147         vmcs_write16(HOST_ES_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
2148         vmcs_write16(HOST_FS_SELECTOR, kvm_read_fs());    /* 22.2.4 */
2149         vmcs_write16(HOST_GS_SELECTOR, kvm_read_gs());    /* 22.2.4 */
2150         vmcs_write16(HOST_SS_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
2151 #ifdef CONFIG_X86_64
2152         rdmsrl(MSR_FS_BASE, a);
2153         vmcs_writel(HOST_FS_BASE, a); /* 22.2.4 */
2154         rdmsrl(MSR_GS_BASE, a);
2155         vmcs_writel(HOST_GS_BASE, a); /* 22.2.4 */
2156 #else
2157         vmcs_writel(HOST_FS_BASE, 0); /* 22.2.4 */
2158         vmcs_writel(HOST_GS_BASE, 0); /* 22.2.4 */
2159 #endif
2160
2161         vmcs_write16(HOST_TR_SELECTOR, GDT_ENTRY_TSS*8);  /* 22.2.4 */
2162
2163         kvm_get_idt(&dt);
2164         vmcs_writel(HOST_IDTR_BASE, dt.base);   /* 22.2.4 */
2165
2166         asm("mov $.Lkvm_vmx_return, %0" : "=r"(kvm_vmx_return));
2167         vmcs_writel(HOST_RIP, kvm_vmx_return); /* 22.2.5 */
2168         vmcs_write32(VM_EXIT_MSR_STORE_COUNT, 0);
2169         vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, 0);
2170         vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, 0);
2171
2172         rdmsr(MSR_IA32_SYSENTER_CS, host_sysenter_cs, junk);
2173         vmcs_write32(HOST_IA32_SYSENTER_CS, host_sysenter_cs);
2174         rdmsrl(MSR_IA32_SYSENTER_ESP, a);
2175         vmcs_writel(HOST_IA32_SYSENTER_ESP, a);   /* 22.2.3 */
2176         rdmsrl(MSR_IA32_SYSENTER_EIP, a);
2177         vmcs_writel(HOST_IA32_SYSENTER_EIP, a);   /* 22.2.3 */
2178
2179         for (i = 0; i < NR_VMX_MSR; ++i) {
2180                 u32 index = vmx_msr_index[i];
2181                 u32 data_low, data_high;
2182                 u64 data;
2183                 int j = vmx->nmsrs;
2184
2185                 if (rdmsr_safe(index, &data_low, &data_high) < 0)
2186                         continue;
2187                 if (wrmsr_safe(index, data_low, data_high) < 0)
2188                         continue;
2189                 data = data_low | ((u64)data_high << 32);
2190                 vmx->host_msrs[j].index = index;
2191                 vmx->host_msrs[j].reserved = 0;
2192                 vmx->host_msrs[j].data = data;
2193                 vmx->guest_msrs[j] = vmx->host_msrs[j];
2194                 ++vmx->nmsrs;
2195         }
2196
2197         vmcs_write32(VM_EXIT_CONTROLS, vmcs_config.vmexit_ctrl);
2198
2199         /* 22.2.1, 20.8.1 */
2200         vmcs_write32(VM_ENTRY_CONTROLS, vmcs_config.vmentry_ctrl);
2201
2202         vmcs_writel(CR0_GUEST_HOST_MASK, ~0UL);
2203         vmcs_writel(CR4_GUEST_HOST_MASK, KVM_GUEST_CR4_MASK);
2204
2205
2206         return 0;
2207 }
2208
2209 static int init_rmode(struct kvm *kvm)
2210 {
2211         if (!init_rmode_tss(kvm))
2212                 return 0;
2213         if (!init_rmode_identity_map(kvm))
2214                 return 0;
2215         return 1;
2216 }
2217
2218 static int vmx_vcpu_reset(struct kvm_vcpu *vcpu)
2219 {
2220         struct vcpu_vmx *vmx = to_vmx(vcpu);
2221         u64 msr;
2222         int ret;
2223
2224         vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP));
2225         down_read(&vcpu->kvm->slots_lock);
2226         if (!init_rmode(vmx->vcpu.kvm)) {
2227                 ret = -ENOMEM;
2228                 goto out;
2229         }
2230
2231         vmx->vcpu.arch.rmode.active = 0;
2232
2233         vmx->vcpu.arch.regs[VCPU_REGS_RDX] = get_rdx_init_val();
2234         kvm_set_cr8(&vmx->vcpu, 0);
2235         msr = 0xfee00000 | MSR_IA32_APICBASE_ENABLE;
2236         if (vmx->vcpu.vcpu_id == 0)
2237                 msr |= MSR_IA32_APICBASE_BSP;
2238         kvm_set_apic_base(&vmx->vcpu, msr);
2239
2240         fx_init(&vmx->vcpu);
2241
2242         seg_setup(VCPU_SREG_CS);
2243         /*
2244          * GUEST_CS_BASE should really be 0xffff0000, but VT vm86 mode
2245          * insists on having GUEST_CS_BASE == GUEST_CS_SELECTOR << 4.  Sigh.
2246          */
2247         if (vmx->vcpu.vcpu_id == 0) {
2248                 vmcs_write16(GUEST_CS_SELECTOR, 0xf000);
2249                 vmcs_writel(GUEST_CS_BASE, 0x000f0000);
2250         } else {
2251                 vmcs_write16(GUEST_CS_SELECTOR, vmx->vcpu.arch.sipi_vector << 8);
2252                 vmcs_writel(GUEST_CS_BASE, vmx->vcpu.arch.sipi_vector << 12);
2253         }
2254
2255         seg_setup(VCPU_SREG_DS);
2256         seg_setup(VCPU_SREG_ES);
2257         seg_setup(VCPU_SREG_FS);
2258         seg_setup(VCPU_SREG_GS);
2259         seg_setup(VCPU_SREG_SS);
2260
2261         vmcs_write16(GUEST_TR_SELECTOR, 0);
2262         vmcs_writel(GUEST_TR_BASE, 0);
2263         vmcs_write32(GUEST_TR_LIMIT, 0xffff);
2264         vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
2265
2266         vmcs_write16(GUEST_LDTR_SELECTOR, 0);
2267         vmcs_writel(GUEST_LDTR_BASE, 0);
2268         vmcs_write32(GUEST_LDTR_LIMIT, 0xffff);
2269         vmcs_write32(GUEST_LDTR_AR_BYTES, 0x00082);
2270
2271         vmcs_write32(GUEST_SYSENTER_CS, 0);
2272         vmcs_writel(GUEST_SYSENTER_ESP, 0);
2273         vmcs_writel(GUEST_SYSENTER_EIP, 0);
2274
2275         vmcs_writel(GUEST_RFLAGS, 0x02);
2276         if (vmx->vcpu.vcpu_id == 0)
2277                 kvm_rip_write(vcpu, 0xfff0);
2278         else
2279                 kvm_rip_write(vcpu, 0);
2280         kvm_register_write(vcpu, VCPU_REGS_RSP, 0);
2281
2282         /* todo: dr0 = dr1 = dr2 = dr3 = 0; dr6 = 0xffff0ff0 */
2283         vmcs_writel(GUEST_DR7, 0x400);
2284
2285         vmcs_writel(GUEST_GDTR_BASE, 0);
2286         vmcs_write32(GUEST_GDTR_LIMIT, 0xffff);
2287
2288         vmcs_writel(GUEST_IDTR_BASE, 0);
2289         vmcs_write32(GUEST_IDTR_LIMIT, 0xffff);
2290
2291         vmcs_write32(GUEST_ACTIVITY_STATE, 0);
2292         vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, 0);
2293         vmcs_write32(GUEST_PENDING_DBG_EXCEPTIONS, 0);
2294
2295         guest_write_tsc(0);
2296
2297         /* Special registers */
2298         vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
2299
2300         setup_msrs(vmx);
2301
2302         vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);  /* 22.2.1 */
2303
2304         if (cpu_has_vmx_tpr_shadow()) {
2305                 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, 0);
2306                 if (vm_need_tpr_shadow(vmx->vcpu.kvm))
2307                         vmcs_write64(VIRTUAL_APIC_PAGE_ADDR,
2308                                 page_to_phys(vmx->vcpu.arch.apic->regs_page));
2309                 vmcs_write32(TPR_THRESHOLD, 0);
2310         }
2311
2312         if (vm_need_virtualize_apic_accesses(vmx->vcpu.kvm))
2313                 vmcs_write64(APIC_ACCESS_ADDR,
2314                              page_to_phys(vmx->vcpu.kvm->arch.apic_access_page));
2315
2316         if (vmx->vpid != 0)
2317                 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
2318
2319         vmx->vcpu.arch.cr0 = 0x60000010;
2320         vmx_set_cr0(&vmx->vcpu, vmx->vcpu.arch.cr0); /* enter rmode */
2321         vmx_set_cr4(&vmx->vcpu, 0);
2322         vmx_set_efer(&vmx->vcpu, 0);
2323         vmx_fpu_activate(&vmx->vcpu);
2324         update_exception_bitmap(&vmx->vcpu);
2325
2326         vpid_sync_vcpu_all(vmx);
2327
2328         ret = 0;
2329
2330         /* HACK: Don't enable emulation on guest boot/reset */
2331         vmx->emulation_required = 0;
2332
2333 out:
2334         up_read(&vcpu->kvm->slots_lock);
2335         return ret;
2336 }
2337
2338 static void vmx_inject_irq(struct kvm_vcpu *vcpu, int irq)
2339 {
2340         struct vcpu_vmx *vmx = to_vmx(vcpu);
2341
2342         KVMTRACE_1D(INJ_VIRQ, vcpu, (u32)irq, handler);
2343
2344         if (vcpu->arch.rmode.active) {
2345                 vmx->rmode.irq.pending = true;
2346                 vmx->rmode.irq.vector = irq;
2347                 vmx->rmode.irq.rip = kvm_rip_read(vcpu);
2348                 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
2349                              irq | INTR_TYPE_SOFT_INTR | INTR_INFO_VALID_MASK);
2350                 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN, 1);
2351                 kvm_rip_write(vcpu, vmx->rmode.irq.rip - 1);
2352                 return;
2353         }
2354         vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
2355                         irq | INTR_TYPE_EXT_INTR | INTR_INFO_VALID_MASK);
2356 }
2357
2358 static void vmx_inject_nmi(struct kvm_vcpu *vcpu)
2359 {
2360         vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
2361                         INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR);
2362 }
2363
2364 static void kvm_do_inject_irq(struct kvm_vcpu *vcpu)
2365 {
2366         int word_index = __ffs(vcpu->arch.irq_summary);
2367         int bit_index = __ffs(vcpu->arch.irq_pending[word_index]);
2368         int irq = word_index * BITS_PER_LONG + bit_index;
2369
2370         clear_bit(bit_index, &vcpu->arch.irq_pending[word_index]);
2371         if (!vcpu->arch.irq_pending[word_index])
2372                 clear_bit(word_index, &vcpu->arch.irq_summary);
2373         kvm_queue_interrupt(vcpu, irq);
2374 }
2375
2376
2377 static void do_interrupt_requests(struct kvm_vcpu *vcpu,
2378                                        struct kvm_run *kvm_run)
2379 {
2380         u32 cpu_based_vm_exec_control;
2381
2382         vcpu->arch.interrupt_window_open =
2383                 ((vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_IF) &&
2384                  (vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & 3) == 0);
2385
2386         if (vcpu->arch.interrupt_window_open &&
2387             vcpu->arch.irq_summary && !vcpu->arch.interrupt.pending)
2388                 kvm_do_inject_irq(vcpu);
2389
2390         if (vcpu->arch.interrupt_window_open && vcpu->arch.interrupt.pending)
2391                 vmx_inject_irq(vcpu, vcpu->arch.interrupt.nr);
2392
2393         cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
2394         if (!vcpu->arch.interrupt_window_open &&
2395             (vcpu->arch.irq_summary || kvm_run->request_interrupt_window))
2396                 /*
2397                  * Interrupts blocked.  Wait for unblock.
2398                  */
2399                 cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_INTR_PENDING;
2400         else
2401                 cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_INTR_PENDING;
2402         vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
2403 }
2404
2405 static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr)
2406 {
2407         int ret;
2408         struct kvm_userspace_memory_region tss_mem = {
2409                 .slot = 8,
2410                 .guest_phys_addr = addr,
2411                 .memory_size = PAGE_SIZE * 3,
2412                 .flags = 0,
2413         };
2414
2415         ret = kvm_set_memory_region(kvm, &tss_mem, 0);
2416         if (ret)
2417                 return ret;
2418         kvm->arch.tss_addr = addr;
2419         return 0;
2420 }
2421
2422 static void kvm_guest_debug_pre(struct kvm_vcpu *vcpu)
2423 {
2424         struct kvm_guest_debug *dbg = &vcpu->guest_debug;
2425
2426         set_debugreg(dbg->bp[0], 0);
2427         set_debugreg(dbg->bp[1], 1);
2428         set_debugreg(dbg->bp[2], 2);
2429         set_debugreg(dbg->bp[3], 3);
2430
2431         if (dbg->singlestep) {
2432                 unsigned long flags;
2433
2434                 flags = vmcs_readl(GUEST_RFLAGS);
2435                 flags |= X86_EFLAGS_TF | X86_EFLAGS_RF;
2436                 vmcs_writel(GUEST_RFLAGS, flags);
2437         }
2438 }
2439
2440 static int handle_rmode_exception(struct kvm_vcpu *vcpu,
2441                                   int vec, u32 err_code)
2442 {
2443         /*
2444          * Instruction with address size override prefix opcode 0x67
2445          * Cause the #SS fault with 0 error code in VM86 mode.
2446          */
2447         if (((vec == GP_VECTOR) || (vec == SS_VECTOR)) && err_code == 0)
2448                 if (emulate_instruction(vcpu, NULL, 0, 0, 0) == EMULATE_DONE)
2449                         return 1;
2450         /*
2451          * Forward all other exceptions that are valid in real mode.
2452          * FIXME: Breaks guest debugging in real mode, needs to be fixed with
2453          *        the required debugging infrastructure rework.
2454          */
2455         switch (vec) {
2456         case DE_VECTOR:
2457         case DB_VECTOR:
2458         case BP_VECTOR:
2459         case OF_VECTOR:
2460         case BR_VECTOR:
2461         case UD_VECTOR:
2462         case DF_VECTOR:
2463         case SS_VECTOR:
2464         case GP_VECTOR:
2465         case MF_VECTOR:
2466                 kvm_queue_exception(vcpu, vec);
2467                 return 1;
2468         }
2469         return 0;
2470 }
2471
2472 static int handle_exception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2473 {
2474         struct vcpu_vmx *vmx = to_vmx(vcpu);
2475         u32 intr_info, error_code;
2476         unsigned long cr2, rip;
2477         u32 vect_info;
2478         enum emulation_result er;
2479
2480         vect_info = vmx->idt_vectoring_info;
2481         intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
2482
2483         if ((vect_info & VECTORING_INFO_VALID_MASK) &&
2484                                                 !is_page_fault(intr_info))
2485                 printk(KERN_ERR "%s: unexpected, vectoring info 0x%x "
2486                        "intr info 0x%x\n", __func__, vect_info, intr_info);
2487
2488         if (!irqchip_in_kernel(vcpu->kvm) && is_external_interrupt(vect_info)) {
2489                 int irq = vect_info & VECTORING_INFO_VECTOR_MASK;
2490                 set_bit(irq, vcpu->arch.irq_pending);
2491                 set_bit(irq / BITS_PER_LONG, &vcpu->arch.irq_summary);
2492         }
2493
2494         if ((intr_info & INTR_INFO_INTR_TYPE_MASK) == 0x200) /* nmi */
2495                 return 1;  /* already handled by vmx_vcpu_run() */
2496
2497         if (is_no_device(intr_info)) {
2498                 vmx_fpu_activate(vcpu);
2499                 return 1;
2500         }
2501
2502         if (is_invalid_opcode(intr_info)) {
2503                 er = emulate_instruction(vcpu, kvm_run, 0, 0, EMULTYPE_TRAP_UD);
2504                 if (er != EMULATE_DONE)
2505                         kvm_queue_exception(vcpu, UD_VECTOR);
2506                 return 1;
2507         }
2508
2509         error_code = 0;
2510         rip = kvm_rip_read(vcpu);
2511         if (intr_info & INTR_INFO_DELIVER_CODE_MASK)
2512                 error_code = vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
2513         if (is_page_fault(intr_info)) {
2514                 /* EPT won't cause page fault directly */
2515                 if (vm_need_ept())
2516                         BUG();
2517                 cr2 = vmcs_readl(EXIT_QUALIFICATION);
2518                 KVMTRACE_3D(PAGE_FAULT, vcpu, error_code, (u32)cr2,
2519                             (u32)((u64)cr2 >> 32), handler);
2520                 if (vcpu->arch.interrupt.pending || vcpu->arch.exception.pending)
2521                         kvm_mmu_unprotect_page_virt(vcpu, cr2);
2522                 return kvm_mmu_page_fault(vcpu, cr2, error_code);
2523         }
2524
2525         if (vcpu->arch.rmode.active &&
2526             handle_rmode_exception(vcpu, intr_info & INTR_INFO_VECTOR_MASK,
2527                                                                 error_code)) {
2528                 if (vcpu->arch.halt_request) {
2529                         vcpu->arch.halt_request = 0;
2530                         return kvm_emulate_halt(vcpu);
2531                 }
2532                 return 1;
2533         }
2534
2535         if ((intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK)) ==
2536             (INTR_TYPE_EXCEPTION | 1)) {
2537                 kvm_run->exit_reason = KVM_EXIT_DEBUG;
2538                 return 0;
2539         }
2540         kvm_run->exit_reason = KVM_EXIT_EXCEPTION;
2541         kvm_run->ex.exception = intr_info & INTR_INFO_VECTOR_MASK;
2542         kvm_run->ex.error_code = error_code;
2543         return 0;
2544 }
2545
2546 static int handle_external_interrupt(struct kvm_vcpu *vcpu,
2547                                      struct kvm_run *kvm_run)
2548 {
2549         ++vcpu->stat.irq_exits;
2550         KVMTRACE_1D(INTR, vcpu, vmcs_read32(VM_EXIT_INTR_INFO), handler);
2551         return 1;
2552 }
2553
2554 static int handle_triple_fault(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2555 {
2556         kvm_run->exit_reason = KVM_EXIT_SHUTDOWN;
2557         return 0;
2558 }
2559
2560 static int handle_io(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2561 {
2562         unsigned long exit_qualification;
2563         int size, down, in, string, rep;
2564         unsigned port;
2565
2566         ++vcpu->stat.io_exits;
2567         exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
2568         string = (exit_qualification & 16) != 0;
2569
2570         if (string) {
2571                 if (emulate_instruction(vcpu,
2572                                         kvm_run, 0, 0, 0) == EMULATE_DO_MMIO)
2573                         return 0;
2574                 return 1;
2575         }
2576
2577         size = (exit_qualification & 7) + 1;
2578         in = (exit_qualification & 8) != 0;
2579         down = (vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_DF) != 0;
2580         rep = (exit_qualification & 32) != 0;
2581         port = exit_qualification >> 16;
2582
2583         return kvm_emulate_pio(vcpu, kvm_run, in, size, port);
2584 }
2585
2586 static void
2587 vmx_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
2588 {
2589         /*
2590          * Patch in the VMCALL instruction:
2591          */
2592         hypercall[0] = 0x0f;
2593         hypercall[1] = 0x01;
2594         hypercall[2] = 0xc1;
2595 }
2596
2597 static int handle_cr(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2598 {
2599         unsigned long exit_qualification;
2600         int cr;
2601         int reg;
2602
2603         exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
2604         cr = exit_qualification & 15;
2605         reg = (exit_qualification >> 8) & 15;
2606         switch ((exit_qualification >> 4) & 3) {
2607         case 0: /* mov to cr */
2608                 KVMTRACE_3D(CR_WRITE, vcpu, (u32)cr,
2609                             (u32)kvm_register_read(vcpu, reg),
2610                             (u32)((u64)kvm_register_read(vcpu, reg) >> 32),
2611                             handler);
2612                 switch (cr) {
2613                 case 0:
2614                         kvm_set_cr0(vcpu, kvm_register_read(vcpu, reg));
2615                         skip_emulated_instruction(vcpu);
2616                         return 1;
2617                 case 3:
2618                         kvm_set_cr3(vcpu, kvm_register_read(vcpu, reg));
2619                         skip_emulated_instruction(vcpu);
2620                         return 1;
2621                 case 4:
2622                         kvm_set_cr4(vcpu, kvm_register_read(vcpu, reg));
2623                         skip_emulated_instruction(vcpu);
2624                         return 1;
2625                 case 8:
2626                         kvm_set_cr8(vcpu, kvm_register_read(vcpu, reg));
2627                         skip_emulated_instruction(vcpu);
2628                         if (irqchip_in_kernel(vcpu->kvm))
2629                                 return 1;
2630                         kvm_run->exit_reason = KVM_EXIT_SET_TPR;
2631                         return 0;
2632                 };
2633                 break;
2634         case 2: /* clts */
2635                 vmx_fpu_deactivate(vcpu);
2636                 vcpu->arch.cr0 &= ~X86_CR0_TS;
2637                 vmcs_writel(CR0_READ_SHADOW, vcpu->arch.cr0);
2638                 vmx_fpu_activate(vcpu);
2639                 KVMTRACE_0D(CLTS, vcpu, handler);
2640                 skip_emulated_instruction(vcpu);
2641                 return 1;
2642         case 1: /*mov from cr*/
2643                 switch (cr) {
2644                 case 3:
2645                         kvm_register_write(vcpu, reg, vcpu->arch.cr3);
2646                         KVMTRACE_3D(CR_READ, vcpu, (u32)cr,
2647                                     (u32)kvm_register_read(vcpu, reg),
2648                                     (u32)((u64)kvm_register_read(vcpu, reg) >> 32),
2649                                     handler);
2650                         skip_emulated_instruction(vcpu);
2651                         return 1;
2652                 case 8:
2653                         kvm_register_write(vcpu, reg, kvm_get_cr8(vcpu));
2654                         KVMTRACE_2D(CR_READ, vcpu, (u32)cr,
2655                                     (u32)kvm_register_read(vcpu, reg), handler);
2656                         skip_emulated_instruction(vcpu);
2657                         return 1;
2658                 }
2659                 break;
2660         case 3: /* lmsw */
2661                 kvm_lmsw(vcpu, (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f);
2662
2663                 skip_emulated_instruction(vcpu);
2664                 return 1;
2665         default:
2666                 break;
2667         }
2668         kvm_run->exit_reason = 0;
2669         pr_unimpl(vcpu, "unhandled control register: op %d cr %d\n",
2670                (int)(exit_qualification >> 4) & 3, cr);
2671         return 0;
2672 }
2673
2674 static int handle_dr(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2675 {
2676         unsigned long exit_qualification;
2677         unsigned long val;
2678         int dr, reg;
2679
2680         /*
2681          * FIXME: this code assumes the host is debugging the guest.
2682          *        need to deal with guest debugging itself too.
2683          */
2684         exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
2685         dr = exit_qualification & 7;
2686         reg = (exit_qualification >> 8) & 15;
2687         if (exit_qualification & 16) {
2688                 /* mov from dr */
2689                 switch (dr) {
2690                 case 6:
2691                         val = 0xffff0ff0;
2692                         break;
2693                 case 7:
2694                         val = 0x400;
2695                         break;
2696                 default:
2697                         val = 0;
2698                 }
2699                 kvm_register_write(vcpu, reg, val);
2700                 KVMTRACE_2D(DR_READ, vcpu, (u32)dr, (u32)val, handler);
2701         } else {
2702                 /* mov to dr */
2703         }
2704         skip_emulated_instruction(vcpu);
2705         return 1;
2706 }
2707
2708 static int handle_cpuid(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2709 {
2710         kvm_emulate_cpuid(vcpu);
2711         return 1;
2712 }
2713
2714 static int handle_rdmsr(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2715 {
2716         u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
2717         u64 data;
2718
2719         if (vmx_get_msr(vcpu, ecx, &data)) {
2720                 kvm_inject_gp(vcpu, 0);
2721                 return 1;
2722         }
2723
2724         KVMTRACE_3D(MSR_READ, vcpu, ecx, (u32)data, (u32)(data >> 32),
2725                     handler);
2726
2727         /* FIXME: handling of bits 32:63 of rax, rdx */
2728         vcpu->arch.regs[VCPU_REGS_RAX] = data & -1u;
2729         vcpu->arch.regs[VCPU_REGS_RDX] = (data >> 32) & -1u;
2730         skip_emulated_instruction(vcpu);
2731         return 1;
2732 }
2733
2734 static int handle_wrmsr(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2735 {
2736         u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
2737         u64 data = (vcpu->arch.regs[VCPU_REGS_RAX] & -1u)
2738                 | ((u64)(vcpu->arch.regs[VCPU_REGS_RDX] & -1u) << 32);
2739
2740         KVMTRACE_3D(MSR_WRITE, vcpu, ecx, (u32)data, (u32)(data >> 32),
2741                     handler);
2742
2743         if (vmx_set_msr(vcpu, ecx, data) != 0) {
2744                 kvm_inject_gp(vcpu, 0);
2745                 return 1;
2746         }
2747
2748         skip_emulated_instruction(vcpu);
2749         return 1;
2750 }
2751
2752 static int handle_tpr_below_threshold(struct kvm_vcpu *vcpu,
2753                                       struct kvm_run *kvm_run)
2754 {
2755         return 1;
2756 }
2757
2758 static int handle_interrupt_window(struct kvm_vcpu *vcpu,
2759                                    struct kvm_run *kvm_run)
2760 {
2761         u32 cpu_based_vm_exec_control;
2762
2763         /* clear pending irq */
2764         cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
2765         cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_INTR_PENDING;
2766         vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
2767
2768         KVMTRACE_0D(PEND_INTR, vcpu, handler);
2769
2770         /*
2771          * If the user space waits to inject interrupts, exit as soon as
2772          * possible
2773          */
2774         if (kvm_run->request_interrupt_window &&
2775             !vcpu->arch.irq_summary) {
2776                 kvm_run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
2777                 ++vcpu->stat.irq_window_exits;
2778                 return 0;
2779         }
2780         return 1;
2781 }
2782
2783 static int handle_halt(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2784 {
2785         skip_emulated_instruction(vcpu);
2786         return kvm_emulate_halt(vcpu);
2787 }
2788
2789 static int handle_vmcall(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2790 {
2791         skip_emulated_instruction(vcpu);
2792         kvm_emulate_hypercall(vcpu);
2793         return 1;
2794 }
2795
2796 static int handle_wbinvd(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2797 {
2798         skip_emulated_instruction(vcpu);
2799         /* TODO: Add support for VT-d/pass-through device */
2800         return 1;
2801 }
2802
2803 static int handle_apic_access(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2804 {
2805         u64 exit_qualification;
2806         enum emulation_result er;
2807         unsigned long offset;
2808
2809         exit_qualification = vmcs_read64(EXIT_QUALIFICATION);
2810         offset = exit_qualification & 0xffful;
2811
2812         er = emulate_instruction(vcpu, kvm_run, 0, 0, 0);
2813
2814         if (er !=  EMULATE_DONE) {
2815                 printk(KERN_ERR
2816                        "Fail to handle apic access vmexit! Offset is 0x%lx\n",
2817                        offset);
2818                 return -ENOTSUPP;
2819         }
2820         return 1;
2821 }
2822
2823 static int handle_task_switch(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2824 {
2825         unsigned long exit_qualification;
2826         u16 tss_selector;
2827         int reason;
2828
2829         exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
2830
2831         reason = (u32)exit_qualification >> 30;
2832         tss_selector = exit_qualification;
2833
2834         return kvm_task_switch(vcpu, tss_selector, reason);
2835 }
2836
2837 static int handle_ept_violation(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2838 {
2839         u64 exit_qualification;
2840         enum emulation_result er;
2841         gpa_t gpa;
2842         unsigned long hva;
2843         int gla_validity;
2844         int r;
2845
2846         exit_qualification = vmcs_read64(EXIT_QUALIFICATION);
2847
2848         if (exit_qualification & (1 << 6)) {
2849                 printk(KERN_ERR "EPT: GPA exceeds GAW!\n");
2850                 return -ENOTSUPP;
2851         }
2852
2853         gla_validity = (exit_qualification >> 7) & 0x3;
2854         if (gla_validity != 0x3 && gla_validity != 0x1 && gla_validity != 0) {
2855                 printk(KERN_ERR "EPT: Handling EPT violation failed!\n");
2856                 printk(KERN_ERR "EPT: GPA: 0x%lx, GVA: 0x%lx\n",
2857                         (long unsigned int)vmcs_read64(GUEST_PHYSICAL_ADDRESS),
2858                         (long unsigned int)vmcs_read64(GUEST_LINEAR_ADDRESS));
2859                 printk(KERN_ERR "EPT: Exit qualification is 0x%lx\n",
2860                         (long unsigned int)exit_qualification);
2861                 kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
2862                 kvm_run->hw.hardware_exit_reason = 0;
2863                 return -ENOTSUPP;
2864         }
2865
2866         gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
2867         hva = gfn_to_hva(vcpu->kvm, gpa >> PAGE_SHIFT);
2868         if (!kvm_is_error_hva(hva)) {
2869                 r = kvm_mmu_page_fault(vcpu, gpa & PAGE_MASK, 0);
2870                 if (r < 0) {
2871                         printk(KERN_ERR "EPT: Not enough memory!\n");
2872                         return -ENOMEM;
2873                 }
2874                 return 1;
2875         } else {
2876                 /* must be MMIO */
2877                 er = emulate_instruction(vcpu, kvm_run, 0, 0, 0);
2878
2879                 if (er == EMULATE_FAIL) {
2880                         printk(KERN_ERR
2881                          "EPT: Fail to handle EPT violation vmexit!er is %d\n",
2882                          er);
2883                         printk(KERN_ERR "EPT: GPA: 0x%lx, GVA: 0x%lx\n",
2884                          (long unsigned int)vmcs_read64(GUEST_PHYSICAL_ADDRESS),
2885                          (long unsigned int)vmcs_read64(GUEST_LINEAR_ADDRESS));
2886                         printk(KERN_ERR "EPT: Exit qualification is 0x%lx\n",
2887                                 (long unsigned int)exit_qualification);
2888                         return -ENOTSUPP;
2889                 } else if (er == EMULATE_DO_MMIO)
2890                         return 0;
2891         }
2892         return 1;
2893 }
2894
2895 static int handle_nmi_window(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2896 {
2897         u32 cpu_based_vm_exec_control;
2898
2899         /* clear pending NMI */
2900         cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
2901         cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_NMI_PENDING;
2902         vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
2903         ++vcpu->stat.nmi_window_exits;
2904
2905         return 1;
2906 }
2907
2908 static void handle_invalid_guest_state(struct kvm_vcpu *vcpu,
2909                                 struct kvm_run *kvm_run)
2910 {
2911         struct vcpu_vmx *vmx = to_vmx(vcpu);
2912         int err;
2913
2914         preempt_enable();
2915         local_irq_enable();
2916
2917         while (!guest_state_valid(vcpu)) {
2918                 err = emulate_instruction(vcpu, kvm_run, 0, 0, 0);
2919
2920                 switch (err) {
2921                         case EMULATE_DONE:
2922                                 break;
2923                         case EMULATE_DO_MMIO:
2924                                 kvm_report_emulation_failure(vcpu, "mmio");
2925                                 /* TODO: Handle MMIO */
2926                                 return;
2927                         default:
2928                                 kvm_report_emulation_failure(vcpu, "emulation failure");
2929                                 return;
2930                 }
2931
2932                 if (signal_pending(current))
2933                         break;
2934                 if (need_resched())
2935                         schedule();
2936         }
2937
2938         local_irq_disable();
2939         preempt_disable();
2940
2941         /* Guest state should be valid now, no more emulation should be needed */
2942         vmx->emulation_required = 0;
2943 }
2944
2945 /*
2946  * The exit handlers return 1 if the exit was handled fully and guest execution
2947  * may resume.  Otherwise they set the kvm_run parameter to indicate what needs
2948  * to be done to userspace and return 0.
2949  */
2950 static int (*kvm_vmx_exit_handlers[])(struct kvm_vcpu *vcpu,
2951                                       struct kvm_run *kvm_run) = {
2952         [EXIT_REASON_EXCEPTION_NMI]           = handle_exception,
2953         [EXIT_REASON_EXTERNAL_INTERRUPT]      = handle_external_interrupt,
2954         [EXIT_REASON_TRIPLE_FAULT]            = handle_triple_fault,
2955         [EXIT_REASON_NMI_WINDOW]              = handle_nmi_window,
2956         [EXIT_REASON_IO_INSTRUCTION]          = handle_io,
2957         [EXIT_REASON_CR_ACCESS]               = handle_cr,
2958         [EXIT_REASON_DR_ACCESS]               = handle_dr,
2959         [EXIT_REASON_CPUID]                   = handle_cpuid,
2960         [EXIT_REASON_MSR_READ]                = handle_rdmsr,
2961         [EXIT_REASON_MSR_WRITE]               = handle_wrmsr,
2962         [EXIT_REASON_PENDING_INTERRUPT]       = handle_interrupt_window,
2963         [EXIT_REASON_HLT]                     = handle_halt,
2964         [EXIT_REASON_VMCALL]                  = handle_vmcall,
2965         [EXIT_REASON_TPR_BELOW_THRESHOLD]     = handle_tpr_below_threshold,
2966         [EXIT_REASON_APIC_ACCESS]             = handle_apic_access,
2967         [EXIT_REASON_WBINVD]                  = handle_wbinvd,
2968         [EXIT_REASON_TASK_SWITCH]             = handle_task_switch,
2969         [EXIT_REASON_EPT_VIOLATION]           = handle_ept_violation,
2970 };
2971
2972 static const int kvm_vmx_max_exit_handlers =
2973         ARRAY_SIZE(kvm_vmx_exit_handlers);
2974
2975 /*
2976  * The guest has exited.  See if we can fix it or if we need userspace
2977  * assistance.
2978  */
2979 static int kvm_handle_exit(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
2980 {
2981         u32 exit_reason = vmcs_read32(VM_EXIT_REASON);
2982         struct vcpu_vmx *vmx = to_vmx(vcpu);
2983         u32 vectoring_info = vmx->idt_vectoring_info;
2984
2985         KVMTRACE_3D(VMEXIT, vcpu, exit_reason, (u32)kvm_rip_read(vcpu),
2986                     (u32)((u64)kvm_rip_read(vcpu) >> 32), entryexit);
2987
2988         /* Access CR3 don't cause VMExit in paging mode, so we need
2989          * to sync with guest real CR3. */
2990         if (vm_need_ept() && is_paging(vcpu)) {
2991                 vcpu->arch.cr3 = vmcs_readl(GUEST_CR3);
2992                 ept_load_pdptrs(vcpu);
2993         }
2994
2995         if (unlikely(vmx->fail)) {
2996                 kvm_run->exit_reason = KVM_EXIT_FAIL_ENTRY;
2997                 kvm_run->fail_entry.hardware_entry_failure_reason
2998                         = vmcs_read32(VM_INSTRUCTION_ERROR);
2999                 return 0;
3000         }
3001
3002         if ((vectoring_info & VECTORING_INFO_VALID_MASK) &&
3003                         (exit_reason != EXIT_REASON_EXCEPTION_NMI &&
3004                         exit_reason != EXIT_REASON_EPT_VIOLATION))
3005                 printk(KERN_WARNING "%s: unexpected, valid vectoring info and "
3006                        "exit reason is 0x%x\n", __func__, exit_reason);
3007         if (exit_reason < kvm_vmx_max_exit_handlers
3008             && kvm_vmx_exit_handlers[exit_reason])
3009                 return kvm_vmx_exit_handlers[exit_reason](vcpu, kvm_run);
3010         else {
3011                 kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
3012                 kvm_run->hw.hardware_exit_reason = exit_reason;
3013         }
3014         return 0;
3015 }
3016
3017 static void update_tpr_threshold(struct kvm_vcpu *vcpu)
3018 {
3019         int max_irr, tpr;
3020
3021         if (!vm_need_tpr_shadow(vcpu->kvm))
3022                 return;
3023
3024         if (!kvm_lapic_enabled(vcpu) ||
3025             ((max_irr = kvm_lapic_find_highest_irr(vcpu)) == -1)) {
3026                 vmcs_write32(TPR_THRESHOLD, 0);
3027                 return;
3028         }
3029
3030         tpr = (kvm_lapic_get_cr8(vcpu) & 0x0f) << 4;
3031         vmcs_write32(TPR_THRESHOLD, (max_irr > tpr) ? tpr >> 4 : max_irr >> 4);
3032 }
3033
3034 static void enable_irq_window(struct kvm_vcpu *vcpu)
3035 {
3036         u32 cpu_based_vm_exec_control;
3037
3038         cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
3039         cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_INTR_PENDING;
3040         vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
3041 }
3042
3043 static void enable_nmi_window(struct kvm_vcpu *vcpu)
3044 {
3045         u32 cpu_based_vm_exec_control;
3046
3047         if (!cpu_has_virtual_nmis())
3048                 return;
3049
3050         cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
3051         cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_NMI_PENDING;
3052         vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
3053 }
3054
3055 static int vmx_nmi_enabled(struct kvm_vcpu *vcpu)
3056 {
3057         u32 guest_intr = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
3058         return !(guest_intr & (GUEST_INTR_STATE_NMI |
3059                                GUEST_INTR_STATE_MOV_SS |
3060                                GUEST_INTR_STATE_STI));
3061 }
3062
3063 static int vmx_irq_enabled(struct kvm_vcpu *vcpu)
3064 {
3065         u32 guest_intr = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
3066         return (!(guest_intr & (GUEST_INTR_STATE_MOV_SS |
3067                                GUEST_INTR_STATE_STI)) &&
3068                 (vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_IF));
3069 }
3070
3071 static void enable_intr_window(struct kvm_vcpu *vcpu)
3072 {
3073         if (vcpu->arch.nmi_pending)
3074                 enable_nmi_window(vcpu);
3075         else if (kvm_cpu_has_interrupt(vcpu))
3076                 enable_irq_window(vcpu);
3077 }
3078
3079 static void vmx_complete_interrupts(struct vcpu_vmx *vmx)
3080 {
3081         u32 exit_intr_info;
3082         u32 idt_vectoring_info;
3083         bool unblock_nmi;
3084         u8 vector;
3085         int type;
3086         bool idtv_info_valid;
3087         u32 error;
3088
3089         exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
3090         if (cpu_has_virtual_nmis()) {
3091                 unblock_nmi = (exit_intr_info & INTR_INFO_UNBLOCK_NMI) != 0;
3092                 vector = exit_intr_info & INTR_INFO_VECTOR_MASK;
3093                 /*
3094                  * SDM 3: 25.7.1.2
3095                  * Re-set bit "block by NMI" before VM entry if vmexit caused by
3096                  * a guest IRET fault.
3097                  */
3098                 if (unblock_nmi && vector != DF_VECTOR)
3099                         vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
3100                                       GUEST_INTR_STATE_NMI);
3101         }
3102
3103         idt_vectoring_info = vmx->idt_vectoring_info;
3104         idtv_info_valid = idt_vectoring_info & VECTORING_INFO_VALID_MASK;
3105         vector = idt_vectoring_info & VECTORING_INFO_VECTOR_MASK;
3106         type = idt_vectoring_info & VECTORING_INFO_TYPE_MASK;
3107         if (vmx->vcpu.arch.nmi_injected) {
3108                 /*
3109                  * SDM 3: 25.7.1.2
3110                  * Clear bit "block by NMI" before VM entry if a NMI delivery
3111                  * faulted.
3112                  */
3113                 if (idtv_info_valid && type == INTR_TYPE_NMI_INTR)
3114                         vmcs_clear_bits(GUEST_INTERRUPTIBILITY_INFO,
3115                                         GUEST_INTR_STATE_NMI);
3116                 else
3117                         vmx->vcpu.arch.nmi_injected = false;
3118         }
3119         kvm_clear_exception_queue(&vmx->vcpu);
3120         if (idtv_info_valid && type == INTR_TYPE_EXCEPTION) {
3121                 if (idt_vectoring_info & VECTORING_INFO_DELIVER_CODE_MASK) {
3122                         error = vmcs_read32(IDT_VECTORING_ERROR_CODE);
3123                         kvm_queue_exception_e(&vmx->vcpu, vector, error);
3124                 } else
3125                         kvm_queue_exception(&vmx->vcpu, vector);
3126                 vmx->idt_vectoring_info = 0;
3127         }
3128         kvm_clear_interrupt_queue(&vmx->vcpu);
3129         if (idtv_info_valid && type == INTR_TYPE_EXT_INTR) {
3130                 kvm_queue_interrupt(&vmx->vcpu, vector);
3131                 vmx->idt_vectoring_info = 0;
3132         }
3133 }
3134
3135 static void vmx_intr_assist(struct kvm_vcpu *vcpu)
3136 {
3137         u32 intr_info_field;
3138
3139         update_tpr_threshold(vcpu);
3140
3141         intr_info_field = vmcs_read32(VM_ENTRY_INTR_INFO_FIELD);
3142         if (cpu_has_virtual_nmis()) {
3143                 if (vcpu->arch.nmi_pending && !vcpu->arch.nmi_injected) {
3144                         if (vmx_nmi_enabled(vcpu)) {
3145                                 vcpu->arch.nmi_pending = false;
3146                                 vcpu->arch.nmi_injected = true;
3147                         } else {
3148                                 enable_intr_window(vcpu);
3149                                 return;
3150                         }
3151                 }
3152                 if (vcpu->arch.nmi_injected) {
3153                         vmx_inject_nmi(vcpu);
3154                         enable_intr_window(vcpu);
3155                         return;
3156                 }
3157         }
3158         if (!vcpu->arch.interrupt.pending && kvm_cpu_has_interrupt(vcpu)) {
3159                 if (vmx_irq_enabled(vcpu))
3160                         kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu));
3161                 else
3162                         enable_irq_window(vcpu);
3163         }
3164         if (vcpu->arch.interrupt.pending) {
3165                 vmx_inject_irq(vcpu, vcpu->arch.interrupt.nr);
3166                 kvm_timer_intr_post(vcpu, vcpu->arch.interrupt.nr);
3167         }
3168 }
3169
3170 /*
3171  * Failure to inject an interrupt should give us the information
3172  * in IDT_VECTORING_INFO_FIELD.  However, if the failure occurs
3173  * when fetching the interrupt redirection bitmap in the real-mode
3174  * tss, this doesn't happen.  So we do it ourselves.
3175  */
3176 static void fixup_rmode_irq(struct vcpu_vmx *vmx)
3177 {
3178         vmx->rmode.irq.pending = 0;
3179         if (kvm_rip_read(&vmx->vcpu) + 1 != vmx->rmode.irq.rip)
3180                 return;
3181         kvm_rip_write(&vmx->vcpu, vmx->rmode.irq.rip);
3182         if (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK) {
3183                 vmx->idt_vectoring_info &= ~VECTORING_INFO_TYPE_MASK;
3184                 vmx->idt_vectoring_info |= INTR_TYPE_EXT_INTR;
3185                 return;
3186         }
3187         vmx->idt_vectoring_info =
3188                 VECTORING_INFO_VALID_MASK
3189                 | INTR_TYPE_EXT_INTR
3190                 | vmx->rmode.irq.vector;
3191 }
3192
3193 #ifdef CONFIG_X86_64
3194 #define R "r"
3195 #define Q "q"
3196 #else
3197 #define R "e"
3198 #define Q "l"
3199 #endif
3200
3201 static void vmx_vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3202 {
3203         struct vcpu_vmx *vmx = to_vmx(vcpu);
3204         u32 intr_info;
3205
3206         /* Handle invalid guest state instead of entering VMX */
3207         if (vmx->emulation_required && emulate_invalid_guest_state) {
3208                 handle_invalid_guest_state(vcpu, kvm_run);
3209                 return;
3210         }
3211
3212         if (test_bit(VCPU_REGS_RSP, (unsigned long *)&vcpu->arch.regs_dirty))
3213                 vmcs_writel(GUEST_RSP, vcpu->arch.regs[VCPU_REGS_RSP]);
3214         if (test_bit(VCPU_REGS_RIP, (unsigned long *)&vcpu->arch.regs_dirty))
3215                 vmcs_writel(GUEST_RIP, vcpu->arch.regs[VCPU_REGS_RIP]);
3216
3217         /*
3218          * Loading guest fpu may have cleared host cr0.ts
3219          */
3220         vmcs_writel(HOST_CR0, read_cr0());
3221
3222         asm(
3223                 /* Store host registers */
3224                 "push %%"R"dx; push %%"R"bp;"
3225                 "push %%"R"cx \n\t"
3226                 "cmp %%"R"sp, %c[host_rsp](%0) \n\t"
3227                 "je 1f \n\t"
3228                 "mov %%"R"sp, %c[host_rsp](%0) \n\t"
3229                 __ex(ASM_VMX_VMWRITE_RSP_RDX) "\n\t"
3230                 "1: \n\t"
3231                 /* Check if vmlaunch of vmresume is needed */
3232                 "cmpl $0, %c[launched](%0) \n\t"
3233                 /* Load guest registers.  Don't clobber flags. */
3234                 "mov %c[cr2](%0), %%"R"ax \n\t"
3235                 "mov %%"R"ax, %%cr2 \n\t"
3236                 "mov %c[rax](%0), %%"R"ax \n\t"
3237                 "mov %c[rbx](%0), %%"R"bx \n\t"
3238                 "mov %c[rdx](%0), %%"R"dx \n\t"
3239                 "mov %c[rsi](%0), %%"R"si \n\t"
3240                 "mov %c[rdi](%0), %%"R"di \n\t"
3241                 "mov %c[rbp](%0), %%"R"bp \n\t"
3242 #ifdef CONFIG_X86_64
3243                 "mov %c[r8](%0),  %%r8  \n\t"
3244                 "mov %c[r9](%0),  %%r9  \n\t"
3245                 "mov %c[r10](%0), %%r10 \n\t"
3246                 "mov %c[r11](%0), %%r11 \n\t"
3247                 "mov %c[r12](%0), %%r12 \n\t"
3248                 "mov %c[r13](%0), %%r13 \n\t"
3249                 "mov %c[r14](%0), %%r14 \n\t"
3250                 "mov %c[r15](%0), %%r15 \n\t"
3251 #endif
3252                 "mov %c[rcx](%0), %%"R"cx \n\t" /* kills %0 (ecx) */
3253
3254                 /* Enter guest mode */
3255                 "jne .Llaunched \n\t"
3256                 __ex(ASM_VMX_VMLAUNCH) "\n\t"
3257                 "jmp .Lkvm_vmx_return \n\t"
3258                 ".Llaunched: " __ex(ASM_VMX_VMRESUME) "\n\t"
3259                 ".Lkvm_vmx_return: "
3260                 /* Save guest registers, load host registers, keep flags */
3261                 "xchg %0,     (%%"R"sp) \n\t"
3262                 "mov %%"R"ax, %c[rax](%0) \n\t"
3263                 "mov %%"R"bx, %c[rbx](%0) \n\t"
3264                 "push"Q" (%%"R"sp); pop"Q" %c[rcx](%0) \n\t"
3265                 "mov %%"R"dx, %c[rdx](%0) \n\t"
3266                 "mov %%"R"si, %c[rsi](%0) \n\t"
3267                 "mov %%"R"di, %c[rdi](%0) \n\t"
3268                 "mov %%"R"bp, %c[rbp](%0) \n\t"
3269 #ifdef CONFIG_X86_64
3270                 "mov %%r8,  %c[r8](%0) \n\t"
3271                 "mov %%r9,  %c[r9](%0) \n\t"
3272                 "mov %%r10, %c[r10](%0) \n\t"
3273                 "mov %%r11, %c[r11](%0) \n\t"
3274                 "mov %%r12, %c[r12](%0) \n\t"
3275                 "mov %%r13, %c[r13](%0) \n\t"
3276                 "mov %%r14, %c[r14](%0) \n\t"
3277                 "mov %%r15, %c[r15](%0) \n\t"
3278 #endif
3279                 "mov %%cr2, %%"R"ax   \n\t"
3280                 "mov %%"R"ax, %c[cr2](%0) \n\t"
3281
3282                 "pop  %%"R"bp; pop  %%"R"bp; pop  %%"R"dx \n\t"
3283                 "setbe %c[fail](%0) \n\t"
3284               : : "c"(vmx), "d"((unsigned long)HOST_RSP),
3285                 [launched]"i"(offsetof(struct vcpu_vmx, launched)),
3286                 [fail]"i"(offsetof(struct vcpu_vmx, fail)),
3287                 [host_rsp]"i"(offsetof(struct vcpu_vmx, host_rsp)),
3288                 [rax]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RAX])),
3289                 [rbx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBX])),
3290                 [rcx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RCX])),
3291                 [rdx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDX])),
3292                 [rsi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RSI])),
3293                 [rdi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDI])),
3294                 [rbp]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBP])),
3295 #ifdef CONFIG_X86_64
3296                 [r8]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R8])),
3297                 [r9]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R9])),
3298                 [r10]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R10])),
3299                 [r11]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R11])),
3300                 [r12]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R12])),
3301                 [r13]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R13])),
3302                 [r14]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R14])),
3303                 [r15]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R15])),
3304 #endif
3305                 [cr2]"i"(offsetof(struct vcpu_vmx, vcpu.arch.cr2))
3306               : "cc", "memory"
3307                 , R"bx", R"di", R"si"
3308 #ifdef CONFIG_X86_64
3309                 , "r8", "r9", "r10", "r11", "r12", "r13", "r14", "r15"
3310 #endif
3311               );
3312
3313         vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP));
3314         vcpu->arch.regs_dirty = 0;
3315
3316         vmx->idt_vectoring_info = vmcs_read32(IDT_VECTORING_INFO_FIELD);
3317         if (vmx->rmode.irq.pending)
3318                 fixup_rmode_irq(vmx);
3319
3320         vcpu->arch.interrupt_window_open =
3321                 (vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
3322                  (GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS)) == 0;
3323
3324         asm("mov %0, %%ds; mov %0, %%es" : : "r"(__USER_DS));
3325         vmx->launched = 1;
3326
3327         intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
3328
3329         /* We need to handle NMIs before interrupts are enabled */
3330         if ((intr_info & INTR_INFO_INTR_TYPE_MASK) == 0x200 &&
3331             (intr_info & INTR_INFO_VALID_MASK)) {
3332                 KVMTRACE_0D(NMI, vcpu, handler);
3333                 asm("int $2");
3334         }
3335
3336         vmx_complete_interrupts(vmx);
3337 }
3338
3339 #undef R
3340 #undef Q
3341
3342 static void vmx_free_vmcs(struct kvm_vcpu *vcpu)
3343 {
3344         struct vcpu_vmx *vmx = to_vmx(vcpu);
3345
3346         if (vmx->vmcs) {
3347                 vcpu_clear(vmx);
3348                 free_vmcs(vmx->vmcs);
3349                 vmx->vmcs = NULL;
3350         }
3351 }
3352
3353 static void vmx_free_vcpu(struct kvm_vcpu *vcpu)
3354 {
3355         struct vcpu_vmx *vmx = to_vmx(vcpu);
3356
3357         spin_lock(&vmx_vpid_lock);
3358         if (vmx->vpid != 0)
3359                 __clear_bit(vmx->vpid, vmx_vpid_bitmap);
3360         spin_unlock(&vmx_vpid_lock);
3361         vmx_free_vmcs(vcpu);
3362         kfree(vmx->host_msrs);
3363         kfree(vmx->guest_msrs);
3364         kvm_vcpu_uninit(vcpu);
3365         kmem_cache_free(kvm_vcpu_cache, vmx);
3366 }
3367
3368 static struct kvm_vcpu *vmx_create_vcpu(struct kvm *kvm, unsigned int id)
3369 {
3370         int err;
3371         struct vcpu_vmx *vmx = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
3372         int cpu;
3373
3374         if (!vmx)
3375                 return ERR_PTR(-ENOMEM);
3376
3377         allocate_vpid(vmx);
3378
3379         err = kvm_vcpu_init(&vmx->vcpu, kvm, id);
3380         if (err)
3381                 goto free_vcpu;
3382
3383         vmx->guest_msrs = kmalloc(PAGE_SIZE, GFP_KERNEL);
3384         if (!vmx->guest_msrs) {
3385                 err = -ENOMEM;
3386                 goto uninit_vcpu;
3387         }
3388
3389         vmx->host_msrs = kmalloc(PAGE_SIZE, GFP_KERNEL);
3390         if (!vmx->host_msrs)
3391                 goto free_guest_msrs;
3392
3393         vmx->vmcs = alloc_vmcs();
3394         if (!vmx->vmcs)
3395                 goto free_msrs;
3396
3397         vmcs_clear(vmx->vmcs);
3398
3399         cpu = get_cpu();
3400         vmx_vcpu_load(&vmx->vcpu, cpu);
3401         err = vmx_vcpu_setup(vmx);
3402         vmx_vcpu_put(&vmx->vcpu);
3403         put_cpu();
3404         if (err)
3405                 goto free_vmcs;
3406         if (vm_need_virtualize_apic_accesses(kvm))
3407                 if (alloc_apic_access_page(kvm) != 0)
3408                         goto free_vmcs;
3409
3410         if (vm_need_ept())
3411                 if (alloc_identity_pagetable(kvm) != 0)
3412                         goto free_vmcs;
3413
3414         return &vmx->vcpu;
3415
3416 free_vmcs:
3417         free_vmcs(vmx->vmcs);
3418 free_msrs:
3419         kfree(vmx->host_msrs);
3420 free_guest_msrs:
3421         kfree(vmx->guest_msrs);
3422 uninit_vcpu:
3423         kvm_vcpu_uninit(&vmx->vcpu);
3424 free_vcpu:
3425         kmem_cache_free(kvm_vcpu_cache, vmx);
3426         return ERR_PTR(err);
3427 }
3428
3429 static void __init vmx_check_processor_compat(void *rtn)
3430 {
3431         struct vmcs_config vmcs_conf;
3432
3433         *(int *)rtn = 0;
3434         if (setup_vmcs_config(&vmcs_conf) < 0)
3435                 *(int *)rtn = -EIO;
3436         if (memcmp(&vmcs_config, &vmcs_conf, sizeof(struct vmcs_config)) != 0) {
3437                 printk(KERN_ERR "kvm: CPU %d feature inconsistency!\n",
3438                                 smp_processor_id());
3439                 *(int *)rtn = -EIO;
3440         }
3441 }
3442
3443 static int get_ept_level(void)
3444 {
3445         return VMX_EPT_DEFAULT_GAW + 1;
3446 }
3447
3448 static struct kvm_x86_ops vmx_x86_ops = {
3449         .cpu_has_kvm_support = cpu_has_kvm_support,
3450         .disabled_by_bios = vmx_disabled_by_bios,
3451         .hardware_setup = hardware_setup,
3452         .hardware_unsetup = hardware_unsetup,
3453         .check_processor_compatibility = vmx_check_processor_compat,
3454         .hardware_enable = hardware_enable,
3455         .hardware_disable = hardware_disable,
3456         .cpu_has_accelerated_tpr = cpu_has_vmx_virtualize_apic_accesses,
3457
3458         .vcpu_create = vmx_create_vcpu,
3459         .vcpu_free = vmx_free_vcpu,
3460         .vcpu_reset = vmx_vcpu_reset,
3461
3462         .prepare_guest_switch = vmx_save_host_state,
3463         .vcpu_load = vmx_vcpu_load,
3464         .vcpu_put = vmx_vcpu_put,
3465
3466         .set_guest_debug = set_guest_debug,
3467         .guest_debug_pre = kvm_guest_debug_pre,
3468         .get_msr = vmx_get_msr,
3469         .set_msr = vmx_set_msr,
3470         .get_segment_base = vmx_get_segment_base,
3471         .get_segment = vmx_get_segment,
3472         .set_segment = vmx_set_segment,
3473         .get_cpl = vmx_get_cpl,
3474         .get_cs_db_l_bits = vmx_get_cs_db_l_bits,
3475         .decache_cr4_guest_bits = vmx_decache_cr4_guest_bits,
3476         .set_cr0 = vmx_set_cr0,
3477         .set_cr3 = vmx_set_cr3,
3478         .set_cr4 = vmx_set_cr4,
3479         .set_efer = vmx_set_efer,
3480         .get_idt = vmx_get_idt,
3481         .set_idt = vmx_set_idt,
3482         .get_gdt = vmx_get_gdt,
3483         .set_gdt = vmx_set_gdt,
3484         .cache_reg = vmx_cache_reg,
3485         .get_rflags = vmx_get_rflags,
3486         .set_rflags = vmx_set_rflags,
3487
3488         .tlb_flush = vmx_flush_tlb,
3489
3490         .run = vmx_vcpu_run,
3491         .handle_exit = kvm_handle_exit,
3492         .skip_emulated_instruction = skip_emulated_instruction,
3493         .patch_hypercall = vmx_patch_hypercall,
3494         .get_irq = vmx_get_irq,
3495         .set_irq = vmx_inject_irq,
3496         .queue_exception = vmx_queue_exception,
3497         .exception_injected = vmx_exception_injected,
3498         .inject_pending_irq = vmx_intr_assist,
3499         .inject_pending_vectors = do_interrupt_requests,
3500
3501         .set_tss_addr = vmx_set_tss_addr,
3502         .get_tdp_level = get_ept_level,
3503 };
3504
3505 static int __init vmx_init(void)
3506 {
3507         void *va;
3508         int r;
3509
3510         vmx_io_bitmap_a = alloc_page(GFP_KERNEL | __GFP_HIGHMEM);
3511         if (!vmx_io_bitmap_a)
3512                 return -ENOMEM;
3513
3514         vmx_io_bitmap_b = alloc_page(GFP_KERNEL | __GFP_HIGHMEM);
3515         if (!vmx_io_bitmap_b) {
3516                 r = -ENOMEM;
3517                 goto out;
3518         }
3519
3520         vmx_msr_bitmap = alloc_page(GFP_KERNEL | __GFP_HIGHMEM);
3521         if (!vmx_msr_bitmap) {
3522                 r = -ENOMEM;
3523                 goto out1;
3524         }
3525
3526         /*
3527          * Allow direct access to the PC debug port (it is often used for I/O
3528          * delays, but the vmexits simply slow things down).
3529          */
3530         va = kmap(vmx_io_bitmap_a);
3531         memset(va, 0xff, PAGE_SIZE);
3532         clear_bit(0x80, va);
3533         kunmap(vmx_io_bitmap_a);
3534
3535         va = kmap(vmx_io_bitmap_b);
3536         memset(va, 0xff, PAGE_SIZE);
3537         kunmap(vmx_io_bitmap_b);
3538
3539         va = kmap(vmx_msr_bitmap);
3540         memset(va, 0xff, PAGE_SIZE);
3541         kunmap(vmx_msr_bitmap);
3542
3543         set_bit(0, vmx_vpid_bitmap); /* 0 is reserved for host */
3544
3545         r = kvm_init(&vmx_x86_ops, sizeof(struct vcpu_vmx), THIS_MODULE);
3546         if (r)
3547                 goto out2;
3548
3549         vmx_disable_intercept_for_msr(vmx_msr_bitmap, MSR_FS_BASE);
3550         vmx_disable_intercept_for_msr(vmx_msr_bitmap, MSR_GS_BASE);
3551         vmx_disable_intercept_for_msr(vmx_msr_bitmap, MSR_IA32_SYSENTER_CS);
3552         vmx_disable_intercept_for_msr(vmx_msr_bitmap, MSR_IA32_SYSENTER_ESP);
3553         vmx_disable_intercept_for_msr(vmx_msr_bitmap, MSR_IA32_SYSENTER_EIP);
3554
3555         if (vm_need_ept()) {
3556                 bypass_guest_pf = 0;
3557                 kvm_mmu_set_base_ptes(VMX_EPT_READABLE_MASK |
3558                         VMX_EPT_WRITABLE_MASK |
3559                         VMX_EPT_DEFAULT_MT << VMX_EPT_MT_EPTE_SHIFT);
3560                 kvm_mmu_set_mask_ptes(0ull, 0ull, 0ull, 0ull,
3561                                 VMX_EPT_EXECUTABLE_MASK);
3562                 kvm_enable_tdp();
3563         } else
3564                 kvm_disable_tdp();
3565
3566         if (bypass_guest_pf)
3567                 kvm_mmu_set_nonpresent_ptes(~0xffeull, 0ull);
3568
3569         ept_sync_global();
3570
3571         return 0;
3572
3573 out2:
3574         __free_page(vmx_msr_bitmap);
3575 out1:
3576         __free_page(vmx_io_bitmap_b);
3577 out:
3578         __free_page(vmx_io_bitmap_a);
3579         return r;
3580 }
3581
3582 static void __exit vmx_exit(void)
3583 {
3584         __free_page(vmx_msr_bitmap);
3585         __free_page(vmx_io_bitmap_b);
3586         __free_page(vmx_io_bitmap_a);
3587
3588         kvm_exit();
3589 }
3590
3591 module_init(vmx_init)
3592 module_exit(vmx_exit)