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  * Copyright 2010 Red Hat, Inc. and/or its affiliates.
   9  *
  10  * Authors:
  11  *   Avi Kivity   <avi@qumranet.com>
  12  *   Yaniv Kamay  <yaniv@qumranet.com>
  13  *
  14  * This work is licensed under the terms of the GNU GPL, version 2.  See
  15  * the COPYING file in the top-level directory.
  16  *
  17  */
  18
  19 #include "irq.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 <linux/ftrace_event.h>
  30 #include <linux/slab.h>
  31 #include <linux/tboot.h>
  32 #include "kvm_cache_regs.h"
  33 #include "x86.h"
  34
  35 #include <asm/io.h>
  36 #include <asm/desc.h>
  37 #include <asm/vmx.h>
  38 #include <asm/virtext.h>
  39 #include <asm/mce.h>
  40 #include <asm/i387.h>
  41 #include <asm/xcr.h>
  42
  43 #include "trace.h"
  44
  45 #define __ex(x) __kvm_handle_fault_on_reboot(x)
  46
  47 MODULE_AUTHOR("Qumranet");
  48 MODULE_LICENSE("GPL");
  49
  50 static int __read_mostly bypass_guest_pf = 1;
  51 module_param(bypass_guest_pf, bool, S_IRUGO);
  52
  53 static int __read_mostly enable_vpid = 1;
  54 module_param_named(vpid, enable_vpid, bool, 0444);
  55
  56 static int __read_mostly flexpriority_enabled = 1;
  57 module_param_named(flexpriority, flexpriority_enabled, bool, S_IRUGO);
  58
  59 static int __read_mostly enable_ept = 1;
  60 module_param_named(ept, enable_ept, bool, S_IRUGO);
  61
  62 static int __read_mostly enable_unrestricted_guest = 1;
  63 module_param_named(unrestricted_guest,
  64                         enable_unrestricted_guest, bool, S_IRUGO);
  65
  66 static int __read_mostly emulate_invalid_guest_state = 0;
  67 module_param(emulate_invalid_guest_state, bool, S_IRUGO);
  68
  69 static int __read_mostly vmm_exclusive = 1;
  70 module_param(vmm_exclusive, bool, S_IRUGO);
  71
  72 static int __read_mostly yield_on_hlt = 1;
  73 module_param(yield_on_hlt, bool, S_IRUGO);
  74
  75 #define KVM_GUEST_CR0_MASK_UNRESTRICTED_GUEST                           \
  76         (X86_CR0_WP | X86_CR0_NE | X86_CR0_NW | X86_CR0_CD)
  77 #define KVM_GUEST_CR0_MASK                                              \
  78         (KVM_GUEST_CR0_MASK_UNRESTRICTED_GUEST | X86_CR0_PG | X86_CR0_PE)
  79 #define KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST                         \
  80         (X86_CR0_WP | X86_CR0_NE)
  81 #define KVM_VM_CR0_ALWAYS_ON                                            \
  82         (KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST | X86_CR0_PG | X86_CR0_PE)
  83 #define KVM_CR4_GUEST_OWNED_BITS                                      \
  84         (X86_CR4_PVI | X86_CR4_DE | X86_CR4_PCE | X86_CR4_OSFXSR      \
  85          | X86_CR4_OSXMMEXCPT)
  86
  87 #define KVM_PMODE_VM_CR4_ALWAYS_ON (X86_CR4_PAE | X86_CR4_VMXE)
  88 #define KVM_RMODE_VM_CR4_ALWAYS_ON (X86_CR4_VME | X86_CR4_PAE | X86_CR4_VMXE)
  89
  90 #define RMODE_GUEST_OWNED_EFLAGS_BITS (~(X86_EFLAGS_IOPL | X86_EFLAGS_VM))
  91
  92 /*
  93  * These 2 parameters are used to config the controls for Pause-Loop Exiting:
  94  * ple_gap:    upper bound on the amount of time between two successive
  95  *             executions of PAUSE in a loop. Also indicate if ple enabled.
  96  *             According to test, this time is usually small than 41 cycles.
  97  * ple_window: upper bound on the amount of time a guest is allowed to execute
  98  *             in a PAUSE loop. Tests indicate that most spinlocks are held for
  99  *             less than 2^12 cycles
 100  * Time is measured based on a counter that runs at the same rate as the TSC,
 101  * refer SDM volume 3b section 21.6.13 & 22.1.3.
 102  */
 103 #define KVM_VMX_DEFAULT_PLE_GAP    41
 104 #define KVM_VMX_DEFAULT_PLE_WINDOW 4096
 105 static int ple_gap = KVM_VMX_DEFAULT_PLE_GAP;
 106 module_param(ple_gap, int, S_IRUGO);
 107
 108 static int ple_window = KVM_VMX_DEFAULT_PLE_WINDOW;
 109 module_param(ple_window, int, S_IRUGO);
 110
 111 #define NR_AUTOLOAD_MSRS 1
 112
 113 struct vmcs {
 114         u32 revision_id;
 115         u32 abort;
 116         char data[0];
 117 };
 118
 119 struct shared_msr_entry {
 120         unsigned index;
 121         u64 data;
 122         u64 mask;
 123 };
 124
 125 struct vcpu_vmx {
 126         struct kvm_vcpu       vcpu;
 127         struct list_head      local_vcpus_link;
 128         unsigned long         host_rsp;
 129         int                   launched;
 130         u8                    fail;
 131         u32                   exit_intr_info;
 132         u32                   idt_vectoring_info;
 133         struct shared_msr_entry *guest_msrs;
 134         int                   nmsrs;
 135         int                   save_nmsrs;
 136 #ifdef CONFIG_X86_64
 137         u64                   msr_host_kernel_gs_base;
 138         u64                   msr_guest_kernel_gs_base;
 139 #endif
 140         struct vmcs          *vmcs;
 141         struct msr_autoload {
 142                 unsigned nr;
 143                 struct vmx_msr_entry guest[NR_AUTOLOAD_MSRS];
 144                 struct vmx_msr_entry host[NR_AUTOLOAD_MSRS];
 145         } msr_autoload;
 146         struct {
 147                 int           loaded;
 148                 u16           fs_sel, gs_sel, ldt_sel;
 149                 int           gs_ldt_reload_needed;
 150                 int           fs_reload_needed;
 151         } host_state;
 152         struct {
 153                 int vm86_active;
 154                 ulong save_rflags;
 155                 struct kvm_save_segment {
 156                         u16 selector;
 157                         unsigned long base;
 158                         u32 limit;
 159                         u32 ar;
 160                 } tr, es, ds, fs, gs;
 161         } rmode;
 162         int vpid;
 163         bool emulation_required;
 164
 165         /* Support for vnmi-less CPUs */
 166         int soft_vnmi_blocked;
 167         ktime_t entry_time;
 168         s64 vnmi_blocked_time;
 169         u32 exit_reason;
 170
 171         bool rdtscp_enabled;
 172 };
 173
 174 static inline struct vcpu_vmx *to_vmx(struct kvm_vcpu *vcpu)
 175 {
 176         return container_of(vcpu, struct vcpu_vmx, vcpu);
 177 }
 178
 179 static int init_rmode(struct kvm *kvm);
 180 static u64 construct_eptp(unsigned long root_hpa);
 181 static void kvm_cpu_vmxon(u64 addr);
 182 static void kvm_cpu_vmxoff(void);
 183
 184 static DEFINE_PER_CPU(struct vmcs *, vmxarea);
 185 static DEFINE_PER_CPU(struct vmcs *, current_vmcs);
 186 static DEFINE_PER_CPU(struct list_head, vcpus_on_cpu);
 187 static DEFINE_PER_CPU(struct desc_ptr, host_gdt);
 188
 189 static unsigned long *vmx_io_bitmap_a;
 190 static unsigned long *vmx_io_bitmap_b;
 191 static unsigned long *vmx_msr_bitmap_legacy;
 192 static unsigned long *vmx_msr_bitmap_longmode;
 193
 194 static DECLARE_BITMAP(vmx_vpid_bitmap, VMX_NR_VPIDS);
 195 static DEFINE_SPINLOCK(vmx_vpid_lock);
 196
 197 static struct vmcs_config {
 198         int size;
 199         int order;
 200         u32 revision_id;
 201         u32 pin_based_exec_ctrl;
 202         u32 cpu_based_exec_ctrl;
 203         u32 cpu_based_2nd_exec_ctrl;
 204         u32 vmexit_ctrl;
 205         u32 vmentry_ctrl;
 206 } vmcs_config;
 207
 208 static struct vmx_capability {
 209         u32 ept;
 210         u32 vpid;
 211 } vmx_capability;
 212
 213 #define VMX_SEGMENT_FIELD(seg)                                  \
 214         [VCPU_SREG_##seg] = {                                   \
 215                 .selector = GUEST_##seg##_SELECTOR,             \
 216                 .base = GUEST_##seg##_BASE,                     \
 217                 .limit = GUEST_##seg##_LIMIT,                   \
 218                 .ar_bytes = GUEST_##seg##_AR_BYTES,             \
 219         }
 220
 221 static struct kvm_vmx_segment_field {
 222         unsigned selector;
 223         unsigned base;
 224         unsigned limit;
 225         unsigned ar_bytes;
 226 } kvm_vmx_segment_fields[] = {
 227         VMX_SEGMENT_FIELD(CS),
 228         VMX_SEGMENT_FIELD(DS),
 229         VMX_SEGMENT_FIELD(ES),
 230         VMX_SEGMENT_FIELD(FS),
 231         VMX_SEGMENT_FIELD(GS),
 232         VMX_SEGMENT_FIELD(SS),
 233         VMX_SEGMENT_FIELD(TR),
 234         VMX_SEGMENT_FIELD(LDTR),
 235 };
 236
 237 static u64 host_efer;
 238
 239 static void ept_save_pdptrs(struct kvm_vcpu *vcpu);
 240
 241 /*
 242  * Keep MSR_STAR at the end, as setup_msrs() will try to optimize it
 243  * away by decrementing the array size.
 244  */
 245 static const u32 vmx_msr_index[] = {
 246 #ifdef CONFIG_X86_64
 247         MSR_SYSCALL_MASK, MSR_LSTAR, MSR_CSTAR,
 248 #endif
 249         MSR_EFER, MSR_TSC_AUX, MSR_STAR,
 250 };
 251 #define NR_VMX_MSR ARRAY_SIZE(vmx_msr_index)
 252
 253 static inline bool is_page_fault(u32 intr_info)
 254 {
 255         return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
 256                              INTR_INFO_VALID_MASK)) ==
 257                 (INTR_TYPE_HARD_EXCEPTION | PF_VECTOR | INTR_INFO_VALID_MASK);
 258 }
 259
 260 static inline bool is_no_device(u32 intr_info)
 261 {
 262         return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
 263                              INTR_INFO_VALID_MASK)) ==
 264                 (INTR_TYPE_HARD_EXCEPTION | NM_VECTOR | INTR_INFO_VALID_MASK);
 265 }
 266
 267 static inline bool is_invalid_opcode(u32 intr_info)
 268 {
 269         return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
 270                              INTR_INFO_VALID_MASK)) ==
 271                 (INTR_TYPE_HARD_EXCEPTION | UD_VECTOR | INTR_INFO_VALID_MASK);
 272 }
 273
 274 static inline bool is_external_interrupt(u32 intr_info)
 275 {
 276         return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VALID_MASK))
 277                 == (INTR_TYPE_EXT_INTR | INTR_INFO_VALID_MASK);
 278 }
 279
 280 static inline bool is_machine_check(u32 intr_info)
 281 {
 282         return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
 283                              INTR_INFO_VALID_MASK)) ==
 284                 (INTR_TYPE_HARD_EXCEPTION | MC_VECTOR | INTR_INFO_VALID_MASK);
 285 }
 286
 287 static inline bool cpu_has_vmx_msr_bitmap(void)
 288 {
 289         return vmcs_config.cpu_based_exec_ctrl & CPU_BASED_USE_MSR_BITMAPS;
 290 }
 291
 292 static inline bool cpu_has_vmx_tpr_shadow(void)
 293 {
 294         return vmcs_config.cpu_based_exec_ctrl & CPU_BASED_TPR_SHADOW;
 295 }
 296
 297 static inline bool vm_need_tpr_shadow(struct kvm *kvm)
 298 {
 299         return (cpu_has_vmx_tpr_shadow()) && (irqchip_in_kernel(kvm));
 300 }
 301
 302 static inline bool cpu_has_secondary_exec_ctrls(void)
 303 {
 304         return vmcs_config.cpu_based_exec_ctrl &
 305                 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
 306 }
 307
 308 static inline bool cpu_has_vmx_virtualize_apic_accesses(void)
 309 {
 310         return vmcs_config.cpu_based_2nd_exec_ctrl &
 311                 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
 312 }
 313
 314 static inline bool cpu_has_vmx_flexpriority(void)
 315 {
 316         return cpu_has_vmx_tpr_shadow() &&
 317                 cpu_has_vmx_virtualize_apic_accesses();
 318 }
 319
 320 static inline bool cpu_has_vmx_ept_execute_only(void)
 321 {
 322         return vmx_capability.ept & VMX_EPT_EXECUTE_ONLY_BIT;
 323 }
 324
 325 static inline bool cpu_has_vmx_eptp_uncacheable(void)
 326 {
 327         return vmx_capability.ept & VMX_EPTP_UC_BIT;
 328 }
 329
 330 static inline bool cpu_has_vmx_eptp_writeback(void)
 331 {
 332         return vmx_capability.ept & VMX_EPTP_WB_BIT;
 333 }
 334
 335 static inline bool cpu_has_vmx_ept_2m_page(void)
 336 {
 337         return vmx_capability.ept & VMX_EPT_2MB_PAGE_BIT;
 338 }
 339
 340 static inline bool cpu_has_vmx_ept_1g_page(void)
 341 {
 342         return vmx_capability.ept & VMX_EPT_1GB_PAGE_BIT;
 343 }
 344
 345 static inline bool cpu_has_vmx_ept_4levels(void)
 346 {
 347         return vmx_capability.ept & VMX_EPT_PAGE_WALK_4_BIT;
 348 }
 349
 350 static inline bool cpu_has_vmx_invept_individual_addr(void)
 351 {
 352         return vmx_capability.ept & VMX_EPT_EXTENT_INDIVIDUAL_BIT;
 353 }
 354
 355 static inline bool cpu_has_vmx_invept_context(void)
 356 {
 357         return vmx_capability.ept & VMX_EPT_EXTENT_CONTEXT_BIT;
 358 }
 359
 360 static inline bool cpu_has_vmx_invept_global(void)
 361 {
 362         return vmx_capability.ept & VMX_EPT_EXTENT_GLOBAL_BIT;
 363 }
 364
 365 static inline bool cpu_has_vmx_invvpid_single(void)
 366 {
 367         return vmx_capability.vpid & VMX_VPID_EXTENT_SINGLE_CONTEXT_BIT;
 368 }
 369
 370 static inline bool cpu_has_vmx_invvpid_global(void)
 371 {
 372         return vmx_capability.vpid & VMX_VPID_EXTENT_GLOBAL_CONTEXT_BIT;
 373 }
 374
 375 static inline bool cpu_has_vmx_ept(void)
 376 {
 377         return vmcs_config.cpu_based_2nd_exec_ctrl &
 378                 SECONDARY_EXEC_ENABLE_EPT;
 379 }
 380
 381 static inline bool cpu_has_vmx_unrestricted_guest(void)
 382 {
 383         return vmcs_config.cpu_based_2nd_exec_ctrl &
 384                 SECONDARY_EXEC_UNRESTRICTED_GUEST;
 385 }
 386
 387 static inline bool cpu_has_vmx_ple(void)
 388 {
 389         return vmcs_config.cpu_based_2nd_exec_ctrl &
 390                 SECONDARY_EXEC_PAUSE_LOOP_EXITING;
 391 }
 392
 393 static inline bool vm_need_virtualize_apic_accesses(struct kvm *kvm)
 394 {
 395         return flexpriority_enabled && irqchip_in_kernel(kvm);
 396 }
 397
 398 static inline bool cpu_has_vmx_vpid(void)
 399 {
 400         return vmcs_config.cpu_based_2nd_exec_ctrl &
 401                 SECONDARY_EXEC_ENABLE_VPID;
 402 }
 403
 404 static inline bool cpu_has_vmx_rdtscp(void)
 405 {
 406         return vmcs_config.cpu_based_2nd_exec_ctrl &
 407                 SECONDARY_EXEC_RDTSCP;
 408 }
 409
 410 static inline bool cpu_has_virtual_nmis(void)
 411 {
 412         return vmcs_config.pin_based_exec_ctrl & PIN_BASED_VIRTUAL_NMIS;
 413 }
 414
 415 static inline bool cpu_has_vmx_wbinvd_exit(void)
 416 {
 417         return vmcs_config.cpu_based_2nd_exec_ctrl &
 418                 SECONDARY_EXEC_WBINVD_EXITING;
 419 }
 420
 421 static inline bool report_flexpriority(void)
 422 {
 423         return flexpriority_enabled;
 424 }
 425
 426 static int __find_msr_index(struct vcpu_vmx *vmx, u32 msr)
 427 {
 428         int i;
 429
 430         for (i = 0; i < vmx->nmsrs; ++i)
 431                 if (vmx_msr_index[vmx->guest_msrs[i].index] == msr)
 432                         return i;
 433         return -1;
 434 }
 435
 436 static inline void __invvpid(int ext, u16 vpid, gva_t gva)
 437 {
 438     struct {
 439         u64 vpid : 16;
 440         u64 rsvd : 48;
 441         u64 gva;
 442     } operand = { vpid, 0, gva };
 443
 444     asm volatile (__ex(ASM_VMX_INVVPID)
 445                   /* CF==1 or ZF==1 --> rc = -1 */
 446                   "; ja 1f ; ud2 ; 1:"
 447                   : : "a"(&operand), "c"(ext) : "cc", "memory");
 448 }
 449
 450 static inline void __invept(int ext, u64 eptp, gpa_t gpa)
 451 {
 452         struct {
 453                 u64 eptp, gpa;
 454         } operand = {eptp, gpa};
 455
 456         asm volatile (__ex(ASM_VMX_INVEPT)
 457                         /* CF==1 or ZF==1 --> rc = -1 */
 458                         "; ja 1f ; ud2 ; 1:\n"
 459                         : : "a" (&operand), "c" (ext) : "cc", "memory");
 460 }
 461
 462 static struct shared_msr_entry *find_msr_entry(struct vcpu_vmx *vmx, u32 msr)
 463 {
 464         int i;
 465
 466         i = __find_msr_index(vmx, msr);
 467         if (i >= 0)
 468                 return &vmx->guest_msrs[i];
 469         return NULL;
 470 }
 471
 472 static void vmcs_clear(struct vmcs *vmcs)
 473 {
 474         u64 phys_addr = __pa(vmcs);
 475         u8 error;
 476
 477         asm volatile (__ex(ASM_VMX_VMCLEAR_RAX) "; setna %0"
 478                       : "=g"(error) : "a"(&phys_addr), "m"(phys_addr)
 479                       : "cc", "memory");
 480         if (error)
 481                 printk(KERN_ERR "kvm: vmclear fail: %p/%llx\n",
 482                        vmcs, phys_addr);
 483 }
 484
 485 static void vmcs_load(struct vmcs *vmcs)
 486 {
 487         u64 phys_addr = __pa(vmcs);
 488         u8 error;
 489
 490         asm volatile (__ex(ASM_VMX_VMPTRLD_RAX) "; setna %0"
 491                         : "=g"(error) : "a"(&phys_addr), "m"(phys_addr)
 492                         : "cc", "memory");
 493         if (error)
 494                 printk(KERN_ERR "kvm: vmptrld %p/%llx fail\n",
 495                        vmcs, phys_addr);
 496 }
 497
 498 static void __vcpu_clear(void *arg)
 499 {
 500         struct vcpu_vmx *vmx = arg;
 501         int cpu = raw_smp_processor_id();
 502
 503         if (vmx->vcpu.cpu == cpu)
 504                 vmcs_clear(vmx->vmcs);
 505         if (per_cpu(current_vmcs, cpu) == vmx->vmcs)
 506                 per_cpu(current_vmcs, cpu) = NULL;
 507         list_del(&vmx->local_vcpus_link);
 508         vmx->vcpu.cpu = -1;
 509         vmx->launched = 0;
 510 }
 511
 512 static void vcpu_clear(struct vcpu_vmx *vmx)
 513 {
 514         if (vmx->vcpu.cpu == -1)
 515                 return;
 516         smp_call_function_single(vmx->vcpu.cpu, __vcpu_clear, vmx, 1);
 517 }
 518
 519 static inline void vpid_sync_vcpu_single(struct vcpu_vmx *vmx)
 520 {
 521         if (vmx->vpid == 0)
 522                 return;
 523
 524         if (cpu_has_vmx_invvpid_single())
 525                 __invvpid(VMX_VPID_EXTENT_SINGLE_CONTEXT, vmx->vpid, 0);
 526 }
 527
 528 static inline void vpid_sync_vcpu_global(void)
 529 {
 530         if (cpu_has_vmx_invvpid_global())
 531                 __invvpid(VMX_VPID_EXTENT_ALL_CONTEXT, 0, 0);
 532 }
 533
 534 static inline void vpid_sync_context(struct vcpu_vmx *vmx)
 535 {
 536         if (cpu_has_vmx_invvpid_single())
 537                 vpid_sync_vcpu_single(vmx);
 538         else
 539                 vpid_sync_vcpu_global();
 540 }
 541
 542 static inline void ept_sync_global(void)
 543 {
 544         if (cpu_has_vmx_invept_global())
 545                 __invept(VMX_EPT_EXTENT_GLOBAL, 0, 0);
 546 }
 547
 548 static inline void ept_sync_context(u64 eptp)
 549 {
 550         if (enable_ept) {
 551                 if (cpu_has_vmx_invept_context())
 552                         __invept(VMX_EPT_EXTENT_CONTEXT, eptp, 0);
 553                 else
 554                         ept_sync_global();
 555         }
 556 }
 557
 558 static inline void ept_sync_individual_addr(u64 eptp, gpa_t gpa)
 559 {
 560         if (enable_ept) {
 561                 if (cpu_has_vmx_invept_individual_addr())
 562                         __invept(VMX_EPT_EXTENT_INDIVIDUAL_ADDR,
 563                                         eptp, gpa);
 564                 else
 565                         ept_sync_context(eptp);
 566         }
 567 }
 568
 569 static unsigned long vmcs_readl(unsigned long field)
 570 {
 571         unsigned long value = 0;
 572
 573         asm volatile (__ex(ASM_VMX_VMREAD_RDX_RAX)
 574                       : "+a"(value) : "d"(field) : "cc");
 575         return value;
 576 }
 577
 578 static u16 vmcs_read16(unsigned long field)
 579 {
 580         return vmcs_readl(field);
 581 }
 582
 583 static u32 vmcs_read32(unsigned long field)
 584 {
 585         return vmcs_readl(field);
 586 }
 587
 588 static u64 vmcs_read64(unsigned long field)
 589 {
 590 #ifdef CONFIG_X86_64
 591         return vmcs_readl(field);
 592 #else
 593         return vmcs_readl(field) | ((u64)vmcs_readl(field+1) << 32);
 594 #endif
 595 }
 596
 597 static noinline void vmwrite_error(unsigned long field, unsigned long value)
 598 {
 599         printk(KERN_ERR "vmwrite error: reg %lx value %lx (err %d)\n",
 600                field, value, vmcs_read32(VM_INSTRUCTION_ERROR));
 601         dump_stack();
 602 }
 603
 604 static void vmcs_writel(unsigned long field, unsigned long value)
 605 {
 606         u8 error;
 607
 608         asm volatile (__ex(ASM_VMX_VMWRITE_RAX_RDX) "; setna %0"
 609                        : "=q"(error) : "a"(value), "d"(field) : "cc");
 610         if (unlikely(error))
 611                 vmwrite_error(field, value);
 612 }
 613
 614 static void vmcs_write16(unsigned long field, u16 value)
 615 {
 616         vmcs_writel(field, value);
 617 }
 618
 619 static void vmcs_write32(unsigned long field, u32 value)
 620 {
 621         vmcs_writel(field, value);
 622 }
 623
 624 static void vmcs_write64(unsigned long field, u64 value)
 625 {
 626         vmcs_writel(field, value);
 627 #ifndef CONFIG_X86_64
 628         asm volatile ("");
 629         vmcs_writel(field+1, value >> 32);
 630 #endif
 631 }
 632
 633 static void vmcs_clear_bits(unsigned long field, u32 mask)
 634 {
 635         vmcs_writel(field, vmcs_readl(field) & ~mask);
 636 }
 637
 638 static void vmcs_set_bits(unsigned long field, u32 mask)
 639 {
 640         vmcs_writel(field, vmcs_readl(field) | mask);
 641 }
 642
 643 static void update_exception_bitmap(struct kvm_vcpu *vcpu)
 644 {
 645         u32 eb;
 646
 647         eb = (1u << PF_VECTOR) | (1u << UD_VECTOR) | (1u << MC_VECTOR) |
 648              (1u << NM_VECTOR) | (1u << DB_VECTOR);
 649         if ((vcpu->guest_debug &
 650              (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP)) ==
 651             (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP))
 652                 eb |= 1u << BP_VECTOR;
 653         if (to_vmx(vcpu)->rmode.vm86_active)
 654                 eb = ~0;
 655         if (enable_ept)
 656                 eb &= ~(1u << PF_VECTOR); /* bypass_guest_pf = 0 */
 657         if (vcpu->fpu_active)
 658                 eb &= ~(1u << NM_VECTOR);
 659         vmcs_write32(EXCEPTION_BITMAP, eb);
 660 }
 661
 662 static void clear_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr)
 663 {
 664         unsigned i;
 665         struct msr_autoload *m = &vmx->msr_autoload;
 666
 667         for (i = 0; i < m->nr; ++i)
 668                 if (m->guest[i].index == msr)
 669                         break;
 670
 671         if (i == m->nr)
 672                 return;
 673         --m->nr;
 674         m->guest[i] = m->guest[m->nr];
 675         m->host[i] = m->host[m->nr];
 676         vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->nr);
 677         vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->nr);
 678 }
 679
 680 static void add_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr,
 681                                   u64 guest_val, u64 host_val)
 682 {
 683         unsigned i;
 684         struct msr_autoload *m = &vmx->msr_autoload;
 685
 686         for (i = 0; i < m->nr; ++i)
 687                 if (m->guest[i].index == msr)
 688                         break;
 689
 690         if (i == m->nr) {
 691                 ++m->nr;
 692                 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->nr);
 693                 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->nr);
 694         }
 695
 696         m->guest[i].index = msr;
 697         m->guest[i].value = guest_val;
 698         m->host[i].index = msr;
 699         m->host[i].value = host_val;
 700 }
 701
 702 static void reload_tss(void)
 703 {
 704         /*
 705          * VT restores TR but not its size.  Useless.
 706          */
 707         struct desc_ptr *gdt = &__get_cpu_var(host_gdt);
 708         struct desc_struct *descs;
 709
 710         descs = (void *)gdt->address;
 711         descs[GDT_ENTRY_TSS].type = 9; /* available TSS */
 712         load_TR_desc();
 713 }
 714
 715 static bool update_transition_efer(struct vcpu_vmx *vmx, int efer_offset)
 716 {
 717         u64 guest_efer;
 718         u64 ignore_bits;
 719
 720         guest_efer = vmx->vcpu.arch.efer;
 721
 722         /*
 723          * NX is emulated; LMA and LME handled by hardware; SCE meaninless
 724          * outside long mode
 725          */
 726         ignore_bits = EFER_NX | EFER_SCE;
 727 #ifdef CONFIG_X86_64
 728         ignore_bits |= EFER_LMA | EFER_LME;
 729         /* SCE is meaningful only in long mode on Intel */
 730         if (guest_efer & EFER_LMA)
 731                 ignore_bits &= ~(u64)EFER_SCE;
 732 #endif
 733         guest_efer &= ~ignore_bits;
 734         guest_efer |= host_efer & ignore_bits;
 735         vmx->guest_msrs[efer_offset].data = guest_efer;
 736         vmx->guest_msrs[efer_offset].mask = ~ignore_bits;
 737
 738         clear_atomic_switch_msr(vmx, MSR_EFER);
 739         /* On ept, can't emulate nx, and must switch nx atomically */
 740         if (enable_ept && ((vmx->vcpu.arch.efer ^ host_efer) & EFER_NX)) {
 741                 guest_efer = vmx->vcpu.arch.efer;
 742                 if (!(guest_efer & EFER_LMA))
 743                         guest_efer &= ~EFER_LME;
 744                 add_atomic_switch_msr(vmx, MSR_EFER, guest_efer, host_efer);
 745                 return false;
 746         }
 747
 748         return true;
 749 }
 750
 751 static unsigned long segment_base(u16 selector)
 752 {
 753         struct desc_ptr *gdt = &__get_cpu_var(host_gdt);
 754         struct desc_struct *d;
 755         unsigned long table_base;
 756         unsigned long v;
 757
 758         if (!(selector & ~3))
 759                 return 0;
 760
 761         table_base = gdt->address;
 762
 763         if (selector & 4) {           /* from ldt */
 764                 u16 ldt_selector = kvm_read_ldt();
 765
 766                 if (!(ldt_selector & ~3))
 767                         return 0;
 768
 769                 table_base = segment_base(ldt_selector);
 770         }
 771         d = (struct desc_struct *)(table_base + (selector & ~7));
 772         v = get_desc_base(d);
 773 #ifdef CONFIG_X86_64
 774        if (d->s == 0 && (d->type == 2 || d->type == 9 || d->type == 11))
 775                v |= ((unsigned long)((struct ldttss_desc64 *)d)->base3) << 32;
 776 #endif
 777         return v;
 778 }
 779
 780 static inline unsigned long kvm_read_tr_base(void)
 781 {
 782         u16 tr;
 783         asm("str %0" : "=g"(tr));
 784         return segment_base(tr);
 785 }
 786
 787 static void vmx_save_host_state(struct kvm_vcpu *vcpu)
 788 {
 789         struct vcpu_vmx *vmx = to_vmx(vcpu);
 790         int i;
 791
 792         if (vmx->host_state.loaded)
 793                 return;
 794
 795         vmx->host_state.loaded = 1;
 796         /*
 797          * Set host fs and gs selectors.  Unfortunately, 22.2.3 does not
 798          * allow segment selectors with cpl > 0 or ti == 1.
 799          */
 800         vmx->host_state.ldt_sel = kvm_read_ldt();
 801         vmx->host_state.gs_ldt_reload_needed = vmx->host_state.ldt_sel;
 802         savesegment(fs, vmx->host_state.fs_sel);
 803         if (!(vmx->host_state.fs_sel & 7)) {
 804                 vmcs_write16(HOST_FS_SELECTOR, vmx->host_state.fs_sel);
 805                 vmx->host_state.fs_reload_needed = 0;
 806         } else {
 807                 vmcs_write16(HOST_FS_SELECTOR, 0);
 808                 vmx->host_state.fs_reload_needed = 1;
 809         }
 810         savesegment(gs, vmx->host_state.gs_sel);
 811         if (!(vmx->host_state.gs_sel & 7))
 812                 vmcs_write16(HOST_GS_SELECTOR, vmx->host_state.gs_sel);
 813         else {
 814                 vmcs_write16(HOST_GS_SELECTOR, 0);
 815                 vmx->host_state.gs_ldt_reload_needed = 1;
 816         }
 817
 818 #ifdef CONFIG_X86_64
 819         vmcs_writel(HOST_FS_BASE, read_msr(MSR_FS_BASE));
 820         vmcs_writel(HOST_GS_BASE, read_msr(MSR_GS_BASE));
 821 #else
 822         vmcs_writel(HOST_FS_BASE, segment_base(vmx->host_state.fs_sel));
 823         vmcs_writel(HOST_GS_BASE, segment_base(vmx->host_state.gs_sel));
 824 #endif
 825
 826 #ifdef CONFIG_X86_64
 827         rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
 828         if (is_long_mode(&vmx->vcpu))
 829                 wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
 830 #endif
 831         for (i = 0; i < vmx->save_nmsrs; ++i)
 832                 kvm_set_shared_msr(vmx->guest_msrs[i].index,
 833                                    vmx->guest_msrs[i].data,
 834                                    vmx->guest_msrs[i].mask);
 835 }
 836
 837 static void __vmx_load_host_state(struct vcpu_vmx *vmx)
 838 {
 839         if (!vmx->host_state.loaded)
 840                 return;
 841
 842         ++vmx->vcpu.stat.host_state_reload;
 843         vmx->host_state.loaded = 0;
 844 #ifdef CONFIG_X86_64
 845         if (is_long_mode(&vmx->vcpu))
 846                 rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
 847 #endif
 848         if (vmx->host_state.gs_ldt_reload_needed) {
 849                 kvm_load_ldt(vmx->host_state.ldt_sel);
 850 #ifdef CONFIG_X86_64
 851                 load_gs_index(vmx->host_state.gs_sel);
 852 #else
 853                 loadsegment(gs, vmx->host_state.gs_sel);
 854 #endif
 855         }
 856         if (vmx->host_state.fs_reload_needed)
 857                 loadsegment(fs, vmx->host_state.fs_sel);
 858         reload_tss();
 859 #ifdef CONFIG_X86_64
 860         wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
 861 #endif
 862         if (current_thread_info()->status & TS_USEDFPU)
 863                 clts();
 864         load_gdt(&__get_cpu_var(host_gdt));
 865 }
 866
 867 static void vmx_load_host_state(struct vcpu_vmx *vmx)
 868 {
 869         preempt_disable();
 870         __vmx_load_host_state(vmx);
 871         preempt_enable();
 872 }
 873
 874 /*
 875  * Switches to specified vcpu, until a matching vcpu_put(), but assumes
 876  * vcpu mutex is already taken.
 877  */
 878 static void vmx_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
 879 {
 880         struct vcpu_vmx *vmx = to_vmx(vcpu);
 881         u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
 882
 883         if (!vmm_exclusive)
 884                 kvm_cpu_vmxon(phys_addr);
 885         else if (vcpu->cpu != cpu)
 886                 vcpu_clear(vmx);
 887
 888         if (per_cpu(current_vmcs, cpu) != vmx->vmcs) {
 889                 per_cpu(current_vmcs, cpu) = vmx->vmcs;
 890                 vmcs_load(vmx->vmcs);
 891         }
 892
 893         if (vcpu->cpu != cpu) {
 894                 struct desc_ptr *gdt = &__get_cpu_var(host_gdt);
 895                 unsigned long sysenter_esp;
 896
 897                 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
 898                 local_irq_disable();
 899                 list_add(&vmx->local_vcpus_link,
 900                          &per_cpu(vcpus_on_cpu, cpu));
 901                 local_irq_enable();
 902
 903                 /*
 904                  * Linux uses per-cpu TSS and GDT, so set these when switching
 905                  * processors.
 906                  */
 907                 vmcs_writel(HOST_TR_BASE, kvm_read_tr_base()); /* 22.2.4 */
 908                 vmcs_writel(HOST_GDTR_BASE, gdt->address);   /* 22.2.4 */
 909
 910                 rdmsrl(MSR_IA32_SYSENTER_ESP, sysenter_esp);
 911                 vmcs_writel(HOST_IA32_SYSENTER_ESP, sysenter_esp); /* 22.2.3 */
 912         }
 913 }
 914
 915 static void vmx_vcpu_put(struct kvm_vcpu *vcpu)
 916 {
 917         __vmx_load_host_state(to_vmx(vcpu));
 918         if (!vmm_exclusive) {
 919                 __vcpu_clear(to_vmx(vcpu));
 920                 kvm_cpu_vmxoff();
 921         }
 922 }
 923
 924 static void vmx_fpu_activate(struct kvm_vcpu *vcpu)
 925 {
 926         ulong cr0;
 927
 928         if (vcpu->fpu_active)
 929                 return;
 930         vcpu->fpu_active = 1;
 931         cr0 = vmcs_readl(GUEST_CR0);
 932         cr0 &= ~(X86_CR0_TS | X86_CR0_MP);
 933         cr0 |= kvm_read_cr0_bits(vcpu, X86_CR0_TS | X86_CR0_MP);
 934         vmcs_writel(GUEST_CR0, cr0);
 935         update_exception_bitmap(vcpu);
 936         vcpu->arch.cr0_guest_owned_bits = X86_CR0_TS;
 937         vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
 938 }
 939
 940 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu);
 941
 942 static void vmx_fpu_deactivate(struct kvm_vcpu *vcpu)
 943 {
 944         vmx_decache_cr0_guest_bits(vcpu);
 945         vmcs_set_bits(GUEST_CR0, X86_CR0_TS | X86_CR0_MP);
 946         update_exception_bitmap(vcpu);
 947         vcpu->arch.cr0_guest_owned_bits = 0;
 948         vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
 949         vmcs_writel(CR0_READ_SHADOW, vcpu->arch.cr0);
 950 }
 951
 952 static unsigned long vmx_get_rflags(struct kvm_vcpu *vcpu)
 953 {
 954         unsigned long rflags, save_rflags;
 955
 956         rflags = vmcs_readl(GUEST_RFLAGS);
 957         if (to_vmx(vcpu)->rmode.vm86_active) {
 958                 rflags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
 959                 save_rflags = to_vmx(vcpu)->rmode.save_rflags;
 960                 rflags |= save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
 961         }
 962         return rflags;
 963 }
 964
 965 static void vmx_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
 966 {
 967         if (to_vmx(vcpu)->rmode.vm86_active) {
 968                 to_vmx(vcpu)->rmode.save_rflags = rflags;
 969                 rflags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
 970         }
 971         vmcs_writel(GUEST_RFLAGS, rflags);
 972 }
 973
 974 static u32 vmx_get_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
 975 {
 976         u32 interruptibility = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
 977         int ret = 0;
 978
 979         if (interruptibility & GUEST_INTR_STATE_STI)
 980                 ret |= KVM_X86_SHADOW_INT_STI;
 981         if (interruptibility & GUEST_INTR_STATE_MOV_SS)
 982                 ret |= KVM_X86_SHADOW_INT_MOV_SS;
 983
 984         return ret & mask;
 985 }
 986
 987 static void vmx_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
 988 {
 989         u32 interruptibility_old = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
 990         u32 interruptibility = interruptibility_old;
 991
 992         interruptibility &= ~(GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS);
 993
 994         if (mask & KVM_X86_SHADOW_INT_MOV_SS)
 995                 interruptibility |= GUEST_INTR_STATE_MOV_SS;
 996         else if (mask & KVM_X86_SHADOW_INT_STI)
 997                 interruptibility |= GUEST_INTR_STATE_STI;
 998
 999         if ((interruptibility != interruptibility_old))
1000                 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, interruptibility);
1001 }
1002
1003 static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
1004 {
1005         unsigned long rip;
1006
1007         rip = kvm_rip_read(vcpu);
1008         rip += vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
1009         kvm_rip_write(vcpu, rip);
1010
1011         /* skipping an emulated instruction also counts */
1012         vmx_set_interrupt_shadow(vcpu, 0);
1013 }
1014
1015 static void vmx_clear_hlt(struct kvm_vcpu *vcpu)
1016 {
1017         /* Ensure that we clear the HLT state in the VMCS.  We don't need to
1018          * explicitly skip the instruction because if the HLT state is set, then
1019          * the instruction is already executing and RIP has already been
1020          * advanced. */
1021         if (!yield_on_hlt &&
1022             vmcs_read32(GUEST_ACTIVITY_STATE) == GUEST_ACTIVITY_HLT)
1023                 vmcs_write32(GUEST_ACTIVITY_STATE, GUEST_ACTIVITY_ACTIVE);
1024 }
1025
1026 static void vmx_queue_exception(struct kvm_vcpu *vcpu, unsigned nr,
1027                                 bool has_error_code, u32 error_code,
1028                                 bool reinject)
1029 {
1030         struct vcpu_vmx *vmx = to_vmx(vcpu);
1031         u32 intr_info = nr | INTR_INFO_VALID_MASK;
1032
1033         if (has_error_code) {
1034                 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE, error_code);
1035                 intr_info |= INTR_INFO_DELIVER_CODE_MASK;
1036         }
1037
1038         if (vmx->rmode.vm86_active) {
1039                 if (kvm_inject_realmode_interrupt(vcpu, nr) != EMULATE_DONE)
1040                         kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
1041                 return;
1042         }
1043
1044         if (kvm_exception_is_soft(nr)) {
1045                 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
1046                              vmx->vcpu.arch.event_exit_inst_len);
1047                 intr_info |= INTR_TYPE_SOFT_EXCEPTION;
1048         } else
1049                 intr_info |= INTR_TYPE_HARD_EXCEPTION;
1050
1051         vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr_info);
1052         vmx_clear_hlt(vcpu);
1053 }
1054
1055 static bool vmx_rdtscp_supported(void)
1056 {
1057         return cpu_has_vmx_rdtscp();
1058 }
1059
1060 /*
1061  * Swap MSR entry in host/guest MSR entry array.
1062  */
1063 static void move_msr_up(struct vcpu_vmx *vmx, int from, int to)
1064 {
1065         struct shared_msr_entry tmp;
1066
1067         tmp = vmx->guest_msrs[to];
1068         vmx->guest_msrs[to] = vmx->guest_msrs[from];
1069         vmx->guest_msrs[from] = tmp;
1070 }
1071
1072 /*
1073  * Set up the vmcs to automatically save and restore system
1074  * msrs.  Don't touch the 64-bit msrs if the guest is in legacy
1075  * mode, as fiddling with msrs is very expensive.
1076  */
1077 static void setup_msrs(struct vcpu_vmx *vmx)
1078 {
1079         int save_nmsrs, index;
1080         unsigned long *msr_bitmap;
1081
1082         vmx_load_host_state(vmx);
1083         save_nmsrs = 0;
1084 #ifdef CONFIG_X86_64
1085         if (is_long_mode(&vmx->vcpu)) {
1086                 index = __find_msr_index(vmx, MSR_SYSCALL_MASK);
1087                 if (index >= 0)
1088                         move_msr_up(vmx, index, save_nmsrs++);
1089                 index = __find_msr_index(vmx, MSR_LSTAR);
1090                 if (index >= 0)
1091                         move_msr_up(vmx, index, save_nmsrs++);
1092                 index = __find_msr_index(vmx, MSR_CSTAR);
1093                 if (index >= 0)
1094                         move_msr_up(vmx, index, save_nmsrs++);
1095                 index = __find_msr_index(vmx, MSR_TSC_AUX);
1096                 if (index >= 0 && vmx->rdtscp_enabled)
1097                         move_msr_up(vmx, index, save_nmsrs++);
1098                 /*
1099                  * MSR_STAR is only needed on long mode guests, and only
1100                  * if efer.sce is enabled.
1101                  */
1102                 index = __find_msr_index(vmx, MSR_STAR);
1103                 if ((index >= 0) && (vmx->vcpu.arch.efer & EFER_SCE))
1104                         move_msr_up(vmx, index, save_nmsrs++);
1105         }
1106 #endif
1107         index = __find_msr_index(vmx, MSR_EFER);
1108         if (index >= 0 && update_transition_efer(vmx, index))
1109                 move_msr_up(vmx, index, save_nmsrs++);
1110
1111         vmx->save_nmsrs = save_nmsrs;
1112
1113         if (cpu_has_vmx_msr_bitmap()) {
1114                 if (is_long_mode(&vmx->vcpu))
1115                         msr_bitmap = vmx_msr_bitmap_longmode;
1116                 else
1117                         msr_bitmap = vmx_msr_bitmap_legacy;
1118
1119                 vmcs_write64(MSR_BITMAP, __pa(msr_bitmap));
1120         }
1121 }
1122
1123 /*
1124  * reads and returns guest's timestamp counter "register"
1125  * guest_tsc = host_tsc + tsc_offset    -- 21.3
1126  */
1127 static u64 guest_read_tsc(void)
1128 {
1129         u64 host_tsc, tsc_offset;
1130
1131         rdtscll(host_tsc);
1132         tsc_offset = vmcs_read64(TSC_OFFSET);
1133         return host_tsc + tsc_offset;
1134 }
1135
1136 /*
1137  * writes 'offset' into guest's timestamp counter offset register
1138  */
1139 static void vmx_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
1140 {
1141         vmcs_write64(TSC_OFFSET, offset);
1142 }
1143
1144 static void vmx_adjust_tsc_offset(struct kvm_vcpu *vcpu, s64 adjustment)
1145 {
1146         u64 offset = vmcs_read64(TSC_OFFSET);
1147         vmcs_write64(TSC_OFFSET, offset + adjustment);
1148 }
1149
1150 /*
1151  * Reads an msr value (of 'msr_index') into 'pdata'.
1152  * Returns 0 on success, non-0 otherwise.
1153  * Assumes vcpu_load() was already called.
1154  */
1155 static int vmx_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1156 {
1157         u64 data;
1158         struct shared_msr_entry *msr;
1159
1160         if (!pdata) {
1161                 printk(KERN_ERR "BUG: get_msr called with NULL pdata\n");
1162                 return -EINVAL;
1163         }
1164
1165         switch (msr_index) {
1166 #ifdef CONFIG_X86_64
1167         case MSR_FS_BASE:
1168                 data = vmcs_readl(GUEST_FS_BASE);
1169                 break;
1170         case MSR_GS_BASE:
1171                 data = vmcs_readl(GUEST_GS_BASE);
1172                 break;
1173         case MSR_KERNEL_GS_BASE:
1174                 vmx_load_host_state(to_vmx(vcpu));
1175                 data = to_vmx(vcpu)->msr_guest_kernel_gs_base;
1176                 break;
1177 #endif
1178         case MSR_EFER:
1179                 return kvm_get_msr_common(vcpu, msr_index, pdata);
1180         case MSR_IA32_TSC:
1181                 data = guest_read_tsc();
1182                 break;
1183         case MSR_IA32_SYSENTER_CS:
1184                 data = vmcs_read32(GUEST_SYSENTER_CS);
1185                 break;
1186         case MSR_IA32_SYSENTER_EIP:
1187                 data = vmcs_readl(GUEST_SYSENTER_EIP);
1188                 break;
1189         case MSR_IA32_SYSENTER_ESP:
1190                 data = vmcs_readl(GUEST_SYSENTER_ESP);
1191                 break;
1192         case MSR_TSC_AUX:
1193                 if (!to_vmx(vcpu)->rdtscp_enabled)
1194                         return 1;
1195                 /* Otherwise falls through */
1196         default:
1197                 vmx_load_host_state(to_vmx(vcpu));
1198                 msr = find_msr_entry(to_vmx(vcpu), msr_index);
1199                 if (msr) {
1200                         vmx_load_host_state(to_vmx(vcpu));
1201                         data = msr->data;
1202                         break;
1203                 }
1204                 return kvm_get_msr_common(vcpu, msr_index, pdata);
1205         }
1206
1207         *pdata = data;
1208         return 0;
1209 }
1210
1211 /*
1212  * Writes msr value into into the appropriate "register".
1213  * Returns 0 on success, non-0 otherwise.
1214  * Assumes vcpu_load() was already called.
1215  */
1216 static int vmx_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
1217 {
1218         struct vcpu_vmx *vmx = to_vmx(vcpu);
1219         struct shared_msr_entry *msr;
1220         int ret = 0;
1221
1222         switch (msr_index) {
1223         case MSR_EFER:
1224                 vmx_load_host_state(vmx);
1225                 ret = kvm_set_msr_common(vcpu, msr_index, data);
1226                 break;
1227 #ifdef CONFIG_X86_64
1228         case MSR_FS_BASE:
1229                 vmcs_writel(GUEST_FS_BASE, data);
1230                 break;
1231         case MSR_GS_BASE:
1232                 vmcs_writel(GUEST_GS_BASE, data);
1233                 break;
1234         case MSR_KERNEL_GS_BASE:
1235                 vmx_load_host_state(vmx);
1236                 vmx->msr_guest_kernel_gs_base = data;
1237                 break;
1238 #endif
1239         case MSR_IA32_SYSENTER_CS:
1240                 vmcs_write32(GUEST_SYSENTER_CS, data);
1241                 break;
1242         case MSR_IA32_SYSENTER_EIP:
1243                 vmcs_writel(GUEST_SYSENTER_EIP, data);
1244                 break;
1245         case MSR_IA32_SYSENTER_ESP:
1246                 vmcs_writel(GUEST_SYSENTER_ESP, data);
1247                 break;
1248         case MSR_IA32_TSC:
1249                 kvm_write_tsc(vcpu, data);
1250                 break;
1251         case MSR_IA32_CR_PAT:
1252                 if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
1253                         vmcs_write64(GUEST_IA32_PAT, data);
1254                         vcpu->arch.pat = data;
1255                         break;
1256                 }
1257                 ret = kvm_set_msr_common(vcpu, msr_index, data);
1258                 break;
1259         case MSR_TSC_AUX:
1260                 if (!vmx->rdtscp_enabled)
1261                         return 1;
1262                 /* Check reserved bit, higher 32 bits should be zero */
1263                 if ((data >> 32) != 0)
1264                         return 1;
1265                 /* Otherwise falls through */
1266         default:
1267                 msr = find_msr_entry(vmx, msr_index);
1268                 if (msr) {
1269                         vmx_load_host_state(vmx);
1270                         msr->data = data;
1271                         break;
1272                 }
1273                 ret = kvm_set_msr_common(vcpu, msr_index, data);
1274         }
1275
1276         return ret;
1277 }
1278
1279 static void vmx_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
1280 {
1281         __set_bit(reg, (unsigned long *)&vcpu->arch.regs_avail);
1282         switch (reg) {
1283         case VCPU_REGS_RSP:
1284                 vcpu->arch.regs[VCPU_REGS_RSP] = vmcs_readl(GUEST_RSP);
1285                 break;
1286         case VCPU_REGS_RIP:
1287                 vcpu->arch.regs[VCPU_REGS_RIP] = vmcs_readl(GUEST_RIP);
1288                 break;
1289         case VCPU_EXREG_PDPTR:
1290                 if (enable_ept)
1291                         ept_save_pdptrs(vcpu);
1292                 break;
1293         default:
1294                 break;
1295         }
1296 }
1297
1298 static void set_guest_debug(struct kvm_vcpu *vcpu, struct kvm_guest_debug *dbg)
1299 {
1300         if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1301                 vmcs_writel(GUEST_DR7, dbg->arch.debugreg[7]);
1302         else
1303                 vmcs_writel(GUEST_DR7, vcpu->arch.dr7);
1304
1305         update_exception_bitmap(vcpu);
1306 }
1307
1308 static __init int cpu_has_kvm_support(void)
1309 {
1310         return cpu_has_vmx();
1311 }
1312
1313 static __init int vmx_disabled_by_bios(void)
1314 {
1315         u64 msr;
1316
1317         rdmsrl(MSR_IA32_FEATURE_CONTROL, msr);
1318         if (msr & FEATURE_CONTROL_LOCKED) {
1319                 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX)
1320                         && tboot_enabled())
1321                         return 1;
1322                 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX)
1323                         && !tboot_enabled()) {
1324                         printk(KERN_WARNING "kvm: disable TXT in the BIOS or "
1325                                 " activate TXT before enabling KVM\n");
1326                         return 1;
1327                 }
1328         }
1329
1330         return 0;
1331         /* locked but not enabled */
1332 }
1333
1334 static void kvm_cpu_vmxon(u64 addr)
1335 {
1336         asm volatile (ASM_VMX_VMXON_RAX
1337                         : : "a"(&addr), "m"(addr)
1338                         : "memory", "cc");
1339 }
1340
1341 static int hardware_enable(void *garbage)
1342 {
1343         int cpu = raw_smp_processor_id();
1344         u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
1345         u64 old, test_bits;
1346
1347         if (read_cr4() & X86_CR4_VMXE)
1348                 return -EBUSY;
1349
1350         INIT_LIST_HEAD(&per_cpu(vcpus_on_cpu, cpu));
1351         rdmsrl(MSR_IA32_FEATURE_CONTROL, old);
1352
1353         test_bits = FEATURE_CONTROL_LOCKED;
1354         test_bits |= FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
1355         if (tboot_enabled())
1356                 test_bits |= FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX;
1357
1358         if ((old & test_bits) != test_bits) {
1359                 /* enable and lock */
1360                 wrmsrl(MSR_IA32_FEATURE_CONTROL, old | test_bits);
1361         }
1362         write_cr4(read_cr4() | X86_CR4_VMXE); /* FIXME: not cpu hotplug safe */
1363
1364         if (vmm_exclusive) {
1365                 kvm_cpu_vmxon(phys_addr);
1366                 ept_sync_global();
1367         }
1368
1369         store_gdt(&__get_cpu_var(host_gdt));
1370
1371         return 0;
1372 }
1373
1374 static void vmclear_local_vcpus(void)
1375 {
1376         int cpu = raw_smp_processor_id();
1377         struct vcpu_vmx *vmx, *n;
1378
1379         list_for_each_entry_safe(vmx, n, &per_cpu(vcpus_on_cpu, cpu),
1380                                  local_vcpus_link)
1381                 __vcpu_clear(vmx);
1382 }
1383
1384
1385 /* Just like cpu_vmxoff(), but with the __kvm_handle_fault_on_reboot()
1386  * tricks.
1387  */
1388 static void kvm_cpu_vmxoff(void)
1389 {
1390         asm volatile (__ex(ASM_VMX_VMXOFF) : : : "cc");
1391 }
1392
1393 static void hardware_disable(void *garbage)
1394 {
1395         if (vmm_exclusive) {
1396                 vmclear_local_vcpus();
1397                 kvm_cpu_vmxoff();
1398         }
1399         write_cr4(read_cr4() & ~X86_CR4_VMXE);
1400 }
1401
1402 static __init int adjust_vmx_controls(u32 ctl_min, u32 ctl_opt,
1403                                       u32 msr, u32 *result)
1404 {
1405         u32 vmx_msr_low, vmx_msr_high;
1406         u32 ctl = ctl_min | ctl_opt;
1407
1408         rdmsr(msr, vmx_msr_low, vmx_msr_high);
1409
1410         ctl &= vmx_msr_high; /* bit == 0 in high word ==> must be zero */
1411         ctl |= vmx_msr_low;  /* bit == 1 in low word  ==> must be one  */
1412
1413         /* Ensure minimum (required) set of control bits are supported. */
1414         if (ctl_min & ~ctl)
1415                 return -EIO;
1416
1417         *result = ctl;
1418         return 0;
1419 }
1420
1421 static __init int setup_vmcs_config(struct vmcs_config *vmcs_conf)
1422 {
1423         u32 vmx_msr_low, vmx_msr_high;
1424         u32 min, opt, min2, opt2;
1425         u32 _pin_based_exec_control = 0;
1426         u32 _cpu_based_exec_control = 0;
1427         u32 _cpu_based_2nd_exec_control = 0;
1428         u32 _vmexit_control = 0;
1429         u32 _vmentry_control = 0;
1430
1431         min = PIN_BASED_EXT_INTR_MASK | PIN_BASED_NMI_EXITING;
1432         opt = PIN_BASED_VIRTUAL_NMIS;
1433         if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PINBASED_CTLS,
1434                                 &_pin_based_exec_control) < 0)
1435                 return -EIO;
1436
1437         min =
1438 #ifdef CONFIG_X86_64
1439               CPU_BASED_CR8_LOAD_EXITING |
1440               CPU_BASED_CR8_STORE_EXITING |
1441 #endif
1442               CPU_BASED_CR3_LOAD_EXITING |
1443               CPU_BASED_CR3_STORE_EXITING |
1444               CPU_BASED_USE_IO_BITMAPS |
1445               CPU_BASED_MOV_DR_EXITING |
1446               CPU_BASED_USE_TSC_OFFSETING |
1447               CPU_BASED_MWAIT_EXITING |
1448               CPU_BASED_MONITOR_EXITING |
1449               CPU_BASED_INVLPG_EXITING;
1450
1451         if (yield_on_hlt)
1452                 min |= CPU_BASED_HLT_EXITING;
1453
1454         opt = CPU_BASED_TPR_SHADOW |
1455               CPU_BASED_USE_MSR_BITMAPS |
1456               CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
1457         if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PROCBASED_CTLS,
1458                                 &_cpu_based_exec_control) < 0)
1459                 return -EIO;
1460 #ifdef CONFIG_X86_64
1461         if ((_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
1462                 _cpu_based_exec_control &= ~CPU_BASED_CR8_LOAD_EXITING &
1463                                            ~CPU_BASED_CR8_STORE_EXITING;
1464 #endif
1465         if (_cpu_based_exec_control & CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) {
1466                 min2 = 0;
1467                 opt2 = SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
1468                         SECONDARY_EXEC_WBINVD_EXITING |
1469                         SECONDARY_EXEC_ENABLE_VPID |
1470                         SECONDARY_EXEC_ENABLE_EPT |
1471                         SECONDARY_EXEC_UNRESTRICTED_GUEST |
1472                         SECONDARY_EXEC_PAUSE_LOOP_EXITING |
1473                         SECONDARY_EXEC_RDTSCP;
1474                 if (adjust_vmx_controls(min2, opt2,
1475                                         MSR_IA32_VMX_PROCBASED_CTLS2,
1476                                         &_cpu_based_2nd_exec_control) < 0)
1477                         return -EIO;
1478         }
1479 #ifndef CONFIG_X86_64
1480         if (!(_cpu_based_2nd_exec_control &
1481                                 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
1482                 _cpu_based_exec_control &= ~CPU_BASED_TPR_SHADOW;
1483 #endif
1484         if (_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_EPT) {
1485                 /* CR3 accesses and invlpg don't need to cause VM Exits when EPT
1486                    enabled */
1487                 _cpu_based_exec_control &= ~(CPU_BASED_CR3_LOAD_EXITING |
1488                                              CPU_BASED_CR3_STORE_EXITING |
1489                                              CPU_BASED_INVLPG_EXITING);
1490                 rdmsr(MSR_IA32_VMX_EPT_VPID_CAP,
1491                       vmx_capability.ept, vmx_capability.vpid);
1492         }
1493
1494         min = 0;
1495 #ifdef CONFIG_X86_64
1496         min |= VM_EXIT_HOST_ADDR_SPACE_SIZE;
1497 #endif
1498         opt = VM_EXIT_SAVE_IA32_PAT | VM_EXIT_LOAD_IA32_PAT;
1499         if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_EXIT_CTLS,
1500                                 &_vmexit_control) < 0)
1501                 return -EIO;
1502
1503         min = 0;
1504         opt = VM_ENTRY_LOAD_IA32_PAT;
1505         if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_ENTRY_CTLS,
1506                                 &_vmentry_control) < 0)
1507                 return -EIO;
1508
1509         rdmsr(MSR_IA32_VMX_BASIC, vmx_msr_low, vmx_msr_high);
1510
1511         /* IA-32 SDM Vol 3B: VMCS size is never greater than 4kB. */
1512         if ((vmx_msr_high & 0x1fff) > PAGE_SIZE)
1513                 return -EIO;
1514
1515 #ifdef CONFIG_X86_64
1516         /* IA-32 SDM Vol 3B: 64-bit CPUs always have VMX_BASIC_MSR[48]==0. */
1517         if (vmx_msr_high & (1u<<16))
1518                 return -EIO;
1519 #endif
1520
1521         /* Require Write-Back (WB) memory type for VMCS accesses. */
1522         if (((vmx_msr_high >> 18) & 15) != 6)
1523                 return -EIO;
1524
1525         vmcs_conf->size = vmx_msr_high & 0x1fff;
1526         vmcs_conf->order = get_order(vmcs_config.size);
1527         vmcs_conf->revision_id = vmx_msr_low;
1528
1529         vmcs_conf->pin_based_exec_ctrl = _pin_based_exec_control;
1530         vmcs_conf->cpu_based_exec_ctrl = _cpu_based_exec_control;
1531         vmcs_conf->cpu_based_2nd_exec_ctrl = _cpu_based_2nd_exec_control;
1532         vmcs_conf->vmexit_ctrl         = _vmexit_control;
1533         vmcs_conf->vmentry_ctrl        = _vmentry_control;
1534
1535         return 0;
1536 }
1537
1538 static struct vmcs *alloc_vmcs_cpu(int cpu)
1539 {
1540         int node = cpu_to_node(cpu);
1541         struct page *pages;
1542         struct vmcs *vmcs;
1543
1544         pages = alloc_pages_exact_node(node, GFP_KERNEL, vmcs_config.order);
1545         if (!pages)
1546                 return NULL;
1547         vmcs = page_address(pages);
1548         memset(vmcs, 0, vmcs_config.size);
1549         vmcs->revision_id = vmcs_config.revision_id; /* vmcs revision id */
1550         return vmcs;
1551 }
1552
1553 static struct vmcs *alloc_vmcs(void)
1554 {
1555         return alloc_vmcs_cpu(raw_smp_processor_id());
1556 }
1557
1558 static void free_vmcs(struct vmcs *vmcs)
1559 {
1560         free_pages((unsigned long)vmcs, vmcs_config.order);
1561 }
1562
1563 static void free_kvm_area(void)
1564 {
1565         int cpu;
1566
1567         for_each_possible_cpu(cpu) {
1568                 free_vmcs(per_cpu(vmxarea, cpu));
1569                 per_cpu(vmxarea, cpu) = NULL;
1570         }
1571 }
1572
1573 static __init int alloc_kvm_area(void)
1574 {
1575         int cpu;
1576
1577         for_each_possible_cpu(cpu) {
1578                 struct vmcs *vmcs;
1579
1580                 vmcs = alloc_vmcs_cpu(cpu);
1581                 if (!vmcs) {
1582                         free_kvm_area();
1583                         return -ENOMEM;
1584                 }
1585
1586                 per_cpu(vmxarea, cpu) = vmcs;
1587         }
1588         return 0;
1589 }
1590
1591 static __init int hardware_setup(void)
1592 {
1593         if (setup_vmcs_config(&vmcs_config) < 0)
1594                 return -EIO;
1595
1596         if (boot_cpu_has(X86_FEATURE_NX))
1597                 kvm_enable_efer_bits(EFER_NX);
1598
1599         if (!cpu_has_vmx_vpid())
1600                 enable_vpid = 0;
1601
1602         if (!cpu_has_vmx_ept() ||
1603             !cpu_has_vmx_ept_4levels()) {
1604                 enable_ept = 0;
1605                 enable_unrestricted_guest = 0;
1606         }
1607
1608         if (!cpu_has_vmx_unrestricted_guest())
1609                 enable_unrestricted_guest = 0;
1610
1611         if (!cpu_has_vmx_flexpriority())
1612                 flexpriority_enabled = 0;
1613
1614         if (!cpu_has_vmx_tpr_shadow())
1615                 kvm_x86_ops->update_cr8_intercept = NULL;
1616
1617         if (enable_ept && !cpu_has_vmx_ept_2m_page())
1618                 kvm_disable_largepages();
1619
1620         if (!cpu_has_vmx_ple())
1621                 ple_gap = 0;
1622
1623         return alloc_kvm_area();
1624 }
1625
1626 static __exit void hardware_unsetup(void)
1627 {
1628         free_kvm_area();
1629 }
1630
1631 static void fix_pmode_dataseg(int seg, struct kvm_save_segment *save)
1632 {
1633         struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1634
1635         if (vmcs_readl(sf->base) == save->base && (save->base & AR_S_MASK)) {
1636                 vmcs_write16(sf->selector, save->selector);
1637                 vmcs_writel(sf->base, save->base);
1638                 vmcs_write32(sf->limit, save->limit);
1639                 vmcs_write32(sf->ar_bytes, save->ar);
1640         } else {
1641                 u32 dpl = (vmcs_read16(sf->selector) & SELECTOR_RPL_MASK)
1642                         << AR_DPL_SHIFT;
1643                 vmcs_write32(sf->ar_bytes, 0x93 | dpl);
1644         }
1645 }
1646
1647 static void enter_pmode(struct kvm_vcpu *vcpu)
1648 {
1649         unsigned long flags;
1650         struct vcpu_vmx *vmx = to_vmx(vcpu);
1651
1652         vmx->emulation_required = 1;
1653         vmx->rmode.vm86_active = 0;
1654
1655         vmcs_writel(GUEST_TR_BASE, vmx->rmode.tr.base);
1656         vmcs_write32(GUEST_TR_LIMIT, vmx->rmode.tr.limit);
1657         vmcs_write32(GUEST_TR_AR_BYTES, vmx->rmode.tr.ar);
1658
1659         flags = vmcs_readl(GUEST_RFLAGS);
1660         flags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
1661         flags |= vmx->rmode.save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
1662         vmcs_writel(GUEST_RFLAGS, flags);
1663
1664         vmcs_writel(GUEST_CR4, (vmcs_readl(GUEST_CR4) & ~X86_CR4_VME) |
1665                         (vmcs_readl(CR4_READ_SHADOW) & X86_CR4_VME));
1666
1667         update_exception_bitmap(vcpu);
1668
1669         if (emulate_invalid_guest_state)
1670                 return;
1671
1672         fix_pmode_dataseg(VCPU_SREG_ES, &vmx->rmode.es);
1673         fix_pmode_dataseg(VCPU_SREG_DS, &vmx->rmode.ds);
1674         fix_pmode_dataseg(VCPU_SREG_GS, &vmx->rmode.gs);
1675         fix_pmode_dataseg(VCPU_SREG_FS, &vmx->rmode.fs);
1676
1677         vmcs_write16(GUEST_SS_SELECTOR, 0);
1678         vmcs_write32(GUEST_SS_AR_BYTES, 0x93);
1679
1680         vmcs_write16(GUEST_CS_SELECTOR,
1681                      vmcs_read16(GUEST_CS_SELECTOR) & ~SELECTOR_RPL_MASK);
1682         vmcs_write32(GUEST_CS_AR_BYTES, 0x9b);
1683 }
1684
1685 static gva_t rmode_tss_base(struct kvm *kvm)
1686 {
1687         if (!kvm->arch.tss_addr) {
1688                 struct kvm_memslots *slots;
1689                 gfn_t base_gfn;
1690
1691                 slots = kvm_memslots(kvm);
1692                 base_gfn = slots->memslots[0].base_gfn +
1693                                  kvm->memslots->memslots[0].npages - 3;
1694                 return base_gfn << PAGE_SHIFT;
1695         }
1696         return kvm->arch.tss_addr;
1697 }
1698
1699 static void fix_rmode_seg(int seg, struct kvm_save_segment *save)
1700 {
1701         struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1702
1703         save->selector = vmcs_read16(sf->selector);
1704         save->base = vmcs_readl(sf->base);
1705         save->limit = vmcs_read32(sf->limit);
1706         save->ar = vmcs_read32(sf->ar_bytes);
1707         vmcs_write16(sf->selector, save->base >> 4);
1708         vmcs_write32(sf->base, save->base & 0xfffff);
1709         vmcs_write32(sf->limit, 0xffff);
1710         vmcs_write32(sf->ar_bytes, 0xf3);
1711 }
1712
1713 static void enter_rmode(struct kvm_vcpu *vcpu)
1714 {
1715         unsigned long flags;
1716         struct vcpu_vmx *vmx = to_vmx(vcpu);
1717
1718         if (enable_unrestricted_guest)
1719                 return;
1720
1721         vmx->emulation_required = 1;
1722         vmx->rmode.vm86_active = 1;
1723
1724         vmx->rmode.tr.base = vmcs_readl(GUEST_TR_BASE);
1725         vmcs_writel(GUEST_TR_BASE, rmode_tss_base(vcpu->kvm));
1726
1727         vmx->rmode.tr.limit = vmcs_read32(GUEST_TR_LIMIT);
1728         vmcs_write32(GUEST_TR_LIMIT, RMODE_TSS_SIZE - 1);
1729
1730         vmx->rmode.tr.ar = vmcs_read32(GUEST_TR_AR_BYTES);
1731         vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
1732
1733         flags = vmcs_readl(GUEST_RFLAGS);
1734         vmx->rmode.save_rflags = flags;
1735
1736         flags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
1737
1738         vmcs_writel(GUEST_RFLAGS, flags);
1739         vmcs_writel(GUEST_CR4, vmcs_readl(GUEST_CR4) | X86_CR4_VME);
1740         update_exception_bitmap(vcpu);
1741
1742         if (emulate_invalid_guest_state)
1743                 goto continue_rmode;
1744
1745         vmcs_write16(GUEST_SS_SELECTOR, vmcs_readl(GUEST_SS_BASE) >> 4);
1746         vmcs_write32(GUEST_SS_LIMIT, 0xffff);
1747         vmcs_write32(GUEST_SS_AR_BYTES, 0xf3);
1748
1749         vmcs_write32(GUEST_CS_AR_BYTES, 0xf3);
1750         vmcs_write32(GUEST_CS_LIMIT, 0xffff);
1751         if (vmcs_readl(GUEST_CS_BASE) == 0xffff0000)
1752                 vmcs_writel(GUEST_CS_BASE, 0xf0000);
1753         vmcs_write16(GUEST_CS_SELECTOR, vmcs_readl(GUEST_CS_BASE) >> 4);
1754
1755         fix_rmode_seg(VCPU_SREG_ES, &vmx->rmode.es);
1756         fix_rmode_seg(VCPU_SREG_DS, &vmx->rmode.ds);
1757         fix_rmode_seg(VCPU_SREG_GS, &vmx->rmode.gs);
1758         fix_rmode_seg(VCPU_SREG_FS, &vmx->rmode.fs);
1759
1760 continue_rmode:
1761         kvm_mmu_reset_context(vcpu);
1762         init_rmode(vcpu->kvm);
1763 }
1764
1765 static void vmx_set_efer(struct kvm_vcpu *vcpu, u64 efer)
1766 {
1767         struct vcpu_vmx *vmx = to_vmx(vcpu);
1768         struct shared_msr_entry *msr = find_msr_entry(vmx, MSR_EFER);
1769
1770         if (!msr)
1771                 return;
1772
1773         /*
1774          * Force kernel_gs_base reloading before EFER changes, as control
1775          * of this msr depends on is_long_mode().
1776          */
1777         vmx_load_host_state(to_vmx(vcpu));
1778         vcpu->arch.efer = efer;
1779         if (efer & EFER_LMA) {
1780                 vmcs_write32(VM_ENTRY_CONTROLS,
1781                              vmcs_read32(VM_ENTRY_CONTROLS) |
1782                              VM_ENTRY_IA32E_MODE);
1783                 msr->data = efer;
1784         } else {
1785                 vmcs_write32(VM_ENTRY_CONTROLS,
1786                              vmcs_read32(VM_ENTRY_CONTROLS) &
1787                              ~VM_ENTRY_IA32E_MODE);
1788
1789                 msr->data = efer & ~EFER_LME;
1790         }
1791         setup_msrs(vmx);
1792 }
1793
1794 #ifdef CONFIG_X86_64
1795
1796 static void enter_lmode(struct kvm_vcpu *vcpu)
1797 {
1798         u32 guest_tr_ar;
1799
1800         guest_tr_ar = vmcs_read32(GUEST_TR_AR_BYTES);
1801         if ((guest_tr_ar & AR_TYPE_MASK) != AR_TYPE_BUSY_64_TSS) {
1802                 printk(KERN_DEBUG "%s: tss fixup for long mode. \n",
1803                        __func__);
1804                 vmcs_write32(GUEST_TR_AR_BYTES,
1805                              (guest_tr_ar & ~AR_TYPE_MASK)
1806                              | AR_TYPE_BUSY_64_TSS);
1807         }
1808         vmx_set_efer(vcpu, vcpu->arch.efer | EFER_LMA);
1809 }
1810
1811 static void exit_lmode(struct kvm_vcpu *vcpu)
1812 {
1813         vmcs_write32(VM_ENTRY_CONTROLS,
1814                      vmcs_read32(VM_ENTRY_CONTROLS)
1815                      & ~VM_ENTRY_IA32E_MODE);
1816         vmx_set_efer(vcpu, vcpu->arch.efer & ~EFER_LMA);
1817 }
1818
1819 #endif
1820
1821 static void vmx_flush_tlb(struct kvm_vcpu *vcpu)
1822 {
1823         vpid_sync_context(to_vmx(vcpu));
1824         if (enable_ept) {
1825                 if (!VALID_PAGE(vcpu->arch.mmu.root_hpa))
1826                         return;
1827                 ept_sync_context(construct_eptp(vcpu->arch.mmu.root_hpa));
1828         }
1829 }
1830
1831 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu)
1832 {
1833         ulong cr0_guest_owned_bits = vcpu->arch.cr0_guest_owned_bits;
1834
1835         vcpu->arch.cr0 &= ~cr0_guest_owned_bits;
1836         vcpu->arch.cr0 |= vmcs_readl(GUEST_CR0) & cr0_guest_owned_bits;
1837 }
1838
1839 static void vmx_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
1840 {
1841         ulong cr4_guest_owned_bits = vcpu->arch.cr4_guest_owned_bits;
1842
1843         vcpu->arch.cr4 &= ~cr4_guest_owned_bits;
1844         vcpu->arch.cr4 |= vmcs_readl(GUEST_CR4) & cr4_guest_owned_bits;
1845 }
1846
1847 static void ept_load_pdptrs(struct kvm_vcpu *vcpu)
1848 {
1849         if (!test_bit(VCPU_EXREG_PDPTR,
1850                       (unsigned long *)&vcpu->arch.regs_dirty))
1851                 return;
1852
1853         if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
1854                 vmcs_write64(GUEST_PDPTR0, vcpu->arch.mmu.pdptrs[0]);
1855                 vmcs_write64(GUEST_PDPTR1, vcpu->arch.mmu.pdptrs[1]);
1856                 vmcs_write64(GUEST_PDPTR2, vcpu->arch.mmu.pdptrs[2]);
1857                 vmcs_write64(GUEST_PDPTR3, vcpu->arch.mmu.pdptrs[3]);
1858         }
1859 }
1860
1861 static void ept_save_pdptrs(struct kvm_vcpu *vcpu)
1862 {
1863         if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
1864                 vcpu->arch.mmu.pdptrs[0] = vmcs_read64(GUEST_PDPTR0);
1865                 vcpu->arch.mmu.pdptrs[1] = vmcs_read64(GUEST_PDPTR1);
1866                 vcpu->arch.mmu.pdptrs[2] = vmcs_read64(GUEST_PDPTR2);
1867                 vcpu->arch.mmu.pdptrs[3] = vmcs_read64(GUEST_PDPTR3);
1868         }
1869
1870         __set_bit(VCPU_EXREG_PDPTR,
1871                   (unsigned long *)&vcpu->arch.regs_avail);
1872         __set_bit(VCPU_EXREG_PDPTR,
1873                   (unsigned long *)&vcpu->arch.regs_dirty);
1874 }
1875
1876 static void vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4);
1877
1878 static void ept_update_paging_mode_cr0(unsigned long *hw_cr0,
1879                                         unsigned long cr0,
1880                                         struct kvm_vcpu *vcpu)
1881 {
1882         if (!(cr0 & X86_CR0_PG)) {
1883                 /* From paging/starting to nonpaging */
1884                 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
1885                              vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) |
1886                              (CPU_BASED_CR3_LOAD_EXITING |
1887                               CPU_BASED_CR3_STORE_EXITING));
1888                 vcpu->arch.cr0 = cr0;
1889                 vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
1890         } else if (!is_paging(vcpu)) {
1891                 /* From nonpaging to paging */
1892                 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
1893                              vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) &
1894                              ~(CPU_BASED_CR3_LOAD_EXITING |
1895                                CPU_BASED_CR3_STORE_EXITING));
1896                 vcpu->arch.cr0 = cr0;
1897                 vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
1898         }
1899
1900         if (!(cr0 & X86_CR0_WP))
1901                 *hw_cr0 &= ~X86_CR0_WP;
1902 }
1903
1904 static void vmx_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
1905 {
1906         struct vcpu_vmx *vmx = to_vmx(vcpu);
1907         unsigned long hw_cr0;
1908
1909         if (enable_unrestricted_guest)
1910                 hw_cr0 = (cr0 & ~KVM_GUEST_CR0_MASK_UNRESTRICTED_GUEST)
1911                         | KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST;
1912         else
1913                 hw_cr0 = (cr0 & ~KVM_GUEST_CR0_MASK) | KVM_VM_CR0_ALWAYS_ON;
1914
1915         if (vmx->rmode.vm86_active && (cr0 & X86_CR0_PE))
1916                 enter_pmode(vcpu);
1917
1918         if (!vmx->rmode.vm86_active && !(cr0 & X86_CR0_PE))
1919                 enter_rmode(vcpu);
1920
1921 #ifdef CONFIG_X86_64
1922         if (vcpu->arch.efer & EFER_LME) {
1923                 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG))
1924                         enter_lmode(vcpu);
1925                 if (is_paging(vcpu) && !(cr0 & X86_CR0_PG))
1926                         exit_lmode(vcpu);
1927         }
1928 #endif
1929
1930         if (enable_ept)
1931                 ept_update_paging_mode_cr0(&hw_cr0, cr0, vcpu);
1932
1933         if (!vcpu->fpu_active)
1934                 hw_cr0 |= X86_CR0_TS | X86_CR0_MP;
1935
1936         vmcs_writel(CR0_READ_SHADOW, cr0);
1937         vmcs_writel(GUEST_CR0, hw_cr0);
1938         vcpu->arch.cr0 = cr0;
1939 }
1940
1941 static u64 construct_eptp(unsigned long root_hpa)
1942 {
1943         u64 eptp;
1944
1945         /* TODO write the value reading from MSR */
1946         eptp = VMX_EPT_DEFAULT_MT |
1947                 VMX_EPT_DEFAULT_GAW << VMX_EPT_GAW_EPTP_SHIFT;
1948         eptp |= (root_hpa & PAGE_MASK);
1949
1950         return eptp;
1951 }
1952
1953 static void vmx_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
1954 {
1955         unsigned long guest_cr3;
1956         u64 eptp;
1957
1958         guest_cr3 = cr3;
1959         if (enable_ept) {
1960                 eptp = construct_eptp(cr3);
1961                 vmcs_write64(EPT_POINTER, eptp);
1962                 guest_cr3 = is_paging(vcpu) ? vcpu->arch.cr3 :
1963                         vcpu->kvm->arch.ept_identity_map_addr;
1964                 ept_load_pdptrs(vcpu);
1965         }
1966
1967         vmx_flush_tlb(vcpu);
1968         vmcs_writel(GUEST_CR3, guest_cr3);
1969 }
1970
1971 static void vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
1972 {
1973         unsigned long hw_cr4 = cr4 | (to_vmx(vcpu)->rmode.vm86_active ?
1974                     KVM_RMODE_VM_CR4_ALWAYS_ON : KVM_PMODE_VM_CR4_ALWAYS_ON);
1975
1976         vcpu->arch.cr4 = cr4;
1977         if (enable_ept) {
1978                 if (!is_paging(vcpu)) {
1979                         hw_cr4 &= ~X86_CR4_PAE;
1980                         hw_cr4 |= X86_CR4_PSE;
1981                 } else if (!(cr4 & X86_CR4_PAE)) {
1982                         hw_cr4 &= ~X86_CR4_PAE;
1983                 }
1984         }
1985
1986         vmcs_writel(CR4_READ_SHADOW, cr4);
1987         vmcs_writel(GUEST_CR4, hw_cr4);
1988 }
1989
1990 static u64 vmx_get_segment_base(struct kvm_vcpu *vcpu, int seg)
1991 {
1992         struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1993
1994         return vmcs_readl(sf->base);
1995 }
1996
1997 static void vmx_get_segment(struct kvm_vcpu *vcpu,
1998                             struct kvm_segment *var, int seg)
1999 {
2000         struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
2001         u32 ar;
2002
2003         var->base = vmcs_readl(sf->base);
2004         var->limit = vmcs_read32(sf->limit);
2005         var->selector = vmcs_read16(sf->selector);
2006         ar = vmcs_read32(sf->ar_bytes);
2007         if ((ar & AR_UNUSABLE_MASK) && !emulate_invalid_guest_state)
2008                 ar = 0;
2009         var->type = ar & 15;
2010         var->s = (ar >> 4) & 1;
2011         var->dpl = (ar >> 5) & 3;
2012         var->present = (ar >> 7) & 1;
2013         var->avl = (ar >> 12) & 1;
2014         var->l = (ar >> 13) & 1;
2015         var->db = (ar >> 14) & 1;
2016         var->g = (ar >> 15) & 1;
2017         var->unusable = (ar >> 16) & 1;
2018 }
2019
2020 static int vmx_get_cpl(struct kvm_vcpu *vcpu)
2021 {
2022         if (!is_protmode(vcpu))
2023                 return 0;
2024
2025         if (vmx_get_rflags(vcpu) & X86_EFLAGS_VM) /* if virtual 8086 */
2026                 return 3;
2027
2028         return vmcs_read16(GUEST_CS_SELECTOR) & 3;
2029 }
2030
2031 static u32 vmx_segment_access_rights(struct kvm_segment *var)
2032 {
2033         u32 ar;
2034
2035         if (var->unusable)
2036                 ar = 1 << 16;
2037         else {
2038                 ar = var->type & 15;
2039                 ar |= (var->s & 1) << 4;
2040                 ar |= (var->dpl & 3) << 5;
2041                 ar |= (var->present & 1) << 7;
2042                 ar |= (var->avl & 1) << 12;
2043                 ar |= (var->l & 1) << 13;
2044                 ar |= (var->db & 1) << 14;
2045                 ar |= (var->g & 1) << 15;
2046         }
2047         if (ar == 0) /* a 0 value means unusable */
2048                 ar = AR_UNUSABLE_MASK;
2049
2050         return ar;
2051 }
2052
2053 static void vmx_set_segment(struct kvm_vcpu *vcpu,
2054                             struct kvm_segment *var, int seg)
2055 {
2056         struct vcpu_vmx *vmx = to_vmx(vcpu);
2057         struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
2058         u32 ar;
2059
2060         if (vmx->rmode.vm86_active && seg == VCPU_SREG_TR) {
2061                 vmx->rmode.tr.selector = var->selector;
2062                 vmx->rmode.tr.base = var->base;
2063                 vmx->rmode.tr.limit = var->limit;
2064                 vmx->rmode.tr.ar = vmx_segment_access_rights(var);
2065                 return;
2066         }
2067         vmcs_writel(sf->base, var->base);
2068         vmcs_write32(sf->limit, var->limit);
2069         vmcs_write16(sf->selector, var->selector);
2070         if (vmx->rmode.vm86_active && var->s) {
2071                 /*
2072                  * Hack real-mode segments into vm86 compatibility.
2073                  */
2074                 if (var->base == 0xffff0000 && var->selector == 0xf000)
2075                         vmcs_writel(sf->base, 0xf0000);
2076                 ar = 0xf3;
2077         } else
2078                 ar = vmx_segment_access_rights(var);
2079
2080         /*
2081          *   Fix the "Accessed" bit in AR field of segment registers for older
2082          * qemu binaries.
2083          *   IA32 arch specifies that at the time of processor reset the
2084          * "Accessed" bit in the AR field of segment registers is 1. And qemu
2085          * is setting it to 0 in the usedland code. This causes invalid guest
2086          * state vmexit when "unrestricted guest" mode is turned on.
2087          *    Fix for this setup issue in cpu_reset is being pushed in the qemu
2088          * tree. Newer qemu binaries with that qemu fix would not need this
2089          * kvm hack.
2090          */
2091         if (enable_unrestricted_guest && (seg != VCPU_SREG_LDTR))
2092                 ar |= 0x1; /* Accessed */
2093
2094         vmcs_write32(sf->ar_bytes, ar);
2095 }
2096
2097 static void vmx_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
2098 {
2099         u32 ar = vmcs_read32(GUEST_CS_AR_BYTES);
2100
2101         *db = (ar >> 14) & 1;
2102         *l = (ar >> 13) & 1;
2103 }
2104
2105 static void vmx_get_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
2106 {
2107         dt->size = vmcs_read32(GUEST_IDTR_LIMIT);
2108         dt->address = vmcs_readl(GUEST_IDTR_BASE);
2109 }
2110
2111 static void vmx_set_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
2112 {
2113         vmcs_write32(GUEST_IDTR_LIMIT, dt->size);
2114         vmcs_writel(GUEST_IDTR_BASE, dt->address);
2115 }
2116
2117 static void vmx_get_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
2118 {
2119         dt->size = vmcs_read32(GUEST_GDTR_LIMIT);
2120         dt->address = vmcs_readl(GUEST_GDTR_BASE);
2121 }
2122
2123 static void vmx_set_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
2124 {
2125         vmcs_write32(GUEST_GDTR_LIMIT, dt->size);
2126         vmcs_writel(GUEST_GDTR_BASE, dt->address);
2127 }
2128
2129 static bool rmode_segment_valid(struct kvm_vcpu *vcpu, int seg)
2130 {
2131         struct kvm_segment var;
2132         u32 ar;
2133
2134         vmx_get_segment(vcpu, &var, seg);
2135         ar = vmx_segment_access_rights(&var);
2136
2137         if (var.base != (var.selector << 4))
2138                 return false;
2139         if (var.limit != 0xffff)
2140                 return false;
2141         if (ar != 0xf3)
2142                 return false;
2143
2144         return true;
2145 }
2146
2147 static bool code_segment_valid(struct kvm_vcpu *vcpu)
2148 {
2149         struct kvm_segment cs;
2150         unsigned int cs_rpl;
2151
2152         vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
2153         cs_rpl = cs.selector & SELECTOR_RPL_MASK;
2154
2155         if (cs.unusable)
2156                 return false;
2157         if (~cs.type & (AR_TYPE_CODE_MASK|AR_TYPE_ACCESSES_MASK))
2158                 return false;
2159         if (!cs.s)
2160                 return false;
2161         if (cs.type & AR_TYPE_WRITEABLE_MASK) {
2162                 if (cs.dpl > cs_rpl)
2163                         return false;
2164         } else {
2165                 if (cs.dpl != cs_rpl)
2166                         return false;
2167         }
2168         if (!cs.present)
2169                 return false;
2170
2171         /* TODO: Add Reserved field check, this'll require a new member in the kvm_segment_field structure */
2172         return true;
2173 }
2174
2175 static bool stack_segment_valid(struct kvm_vcpu *vcpu)
2176 {
2177         struct kvm_segment ss;
2178         unsigned int ss_rpl;
2179
2180         vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
2181         ss_rpl = ss.selector & SELECTOR_RPL_MASK;
2182
2183         if (ss.unusable)
2184                 return true;
2185         if (ss.type != 3 && ss.type != 7)
2186                 return false;
2187         if (!ss.s)
2188                 return false;
2189         if (ss.dpl != ss_rpl) /* DPL != RPL */
2190                 return false;
2191         if (!ss.present)
2192                 return false;
2193
2194         return true;
2195 }
2196
2197 static bool data_segment_valid(struct kvm_vcpu *vcpu, int seg)
2198 {
2199         struct kvm_segment var;
2200         unsigned int rpl;
2201
2202         vmx_get_segment(vcpu, &var, seg);
2203         rpl = var.selector & SELECTOR_RPL_MASK;
2204
2205         if (var.unusable)
2206                 return true;
2207         if (!var.s)
2208                 return false;
2209         if (!var.present)
2210                 return false;
2211         if (~var.type & (AR_TYPE_CODE_MASK|AR_TYPE_WRITEABLE_MASK)) {
2212                 if (var.dpl < rpl) /* DPL < RPL */
2213                         return false;
2214         }
2215
2216         /* TODO: Add other members to kvm_segment_field to allow checking for other access
2217          * rights flags
2218          */
2219         return true;
2220 }
2221
2222 static bool tr_valid(struct kvm_vcpu *vcpu)
2223 {
2224         struct kvm_segment tr;
2225
2226         vmx_get_segment(vcpu, &tr, VCPU_SREG_TR);
2227
2228         if (tr.unusable)
2229                 return false;
2230         if (tr.selector & SELECTOR_TI_MASK)     /* TI = 1 */
2231                 return false;
2232         if (tr.type != 3 && tr.type != 11) /* TODO: Check if guest is in IA32e mode */
2233                 return false;
2234         if (!tr.present)
2235                 return false;
2236
2237         return true;
2238 }
2239
2240 static bool ldtr_valid(struct kvm_vcpu *vcpu)
2241 {
2242         struct kvm_segment ldtr;
2243
2244         vmx_get_segment(vcpu, &ldtr, VCPU_SREG_LDTR);
2245
2246         if (ldtr.unusable)
2247                 return true;
2248         if (ldtr.selector & SELECTOR_TI_MASK)   /* TI = 1 */
2249                 return false;
2250         if (ldtr.type != 2)
2251                 return false;
2252         if (!ldtr.present)
2253                 return false;
2254
2255         return true;
2256 }
2257
2258 static bool cs_ss_rpl_check(struct kvm_vcpu *vcpu)
2259 {
2260         struct kvm_segment cs, ss;
2261
2262         vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
2263         vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
2264
2265         return ((cs.selector & SELECTOR_RPL_MASK) ==
2266                  (ss.selector & SELECTOR_RPL_MASK));
2267 }
2268
2269 /*
2270  * Check if guest state is valid. Returns true if valid, false if
2271  * not.
2272  * We assume that registers are always usable
2273  */
2274 static bool guest_state_valid(struct kvm_vcpu *vcpu)
2275 {
2276         /* real mode guest state checks */
2277         if (!is_protmode(vcpu)) {
2278                 if (!rmode_segment_valid(vcpu, VCPU_SREG_CS))
2279                         return false;
2280                 if (!rmode_segment_valid(vcpu, VCPU_SREG_SS))
2281                         return false;
2282                 if (!rmode_segment_valid(vcpu, VCPU_SREG_DS))
2283                         return false;
2284                 if (!rmode_segment_valid(vcpu, VCPU_SREG_ES))
2285                         return false;
2286                 if (!rmode_segment_valid(vcpu, VCPU_SREG_FS))
2287                         return false;
2288                 if (!rmode_segment_valid(vcpu, VCPU_SREG_GS))
2289                         return false;
2290         } else {
2291         /* protected mode guest state checks */
2292                 if (!cs_ss_rpl_check(vcpu))
2293                         return false;
2294                 if (!code_segment_valid(vcpu))
2295                         return false;
2296                 if (!stack_segment_valid(vcpu))
2297                         return false;
2298                 if (!data_segment_valid(vcpu, VCPU_SREG_DS))
2299                         return false;
2300                 if (!data_segment_valid(vcpu, VCPU_SREG_ES))
2301                         return false;
2302                 if (!data_segment_valid(vcpu, VCPU_SREG_FS))
2303                         return false;
2304                 if (!data_segment_valid(vcpu, VCPU_SREG_GS))
2305                         return false;
2306                 if (!tr_valid(vcpu))
2307                         return false;
2308                 if (!ldtr_valid(vcpu))
2309                         return false;
2310         }
2311         /* TODO:
2312          * - Add checks on RIP
2313          * - Add checks on RFLAGS
2314          */
2315
2316         return true;
2317 }
2318
2319 static int init_rmode_tss(struct kvm *kvm)
2320 {
2321         gfn_t fn = rmode_tss_base(kvm) >> PAGE_SHIFT;
2322         u16 data = 0;
2323         int ret = 0;
2324         int r;
2325
2326         r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
2327         if (r < 0)
2328                 goto out;
2329         data = TSS_BASE_SIZE + TSS_REDIRECTION_SIZE;
2330         r = kvm_write_guest_page(kvm, fn++, &data,
2331                         TSS_IOPB_BASE_OFFSET, sizeof(u16));
2332         if (r < 0)
2333                 goto out;
2334         r = kvm_clear_guest_page(kvm, fn++, 0, PAGE_SIZE);
2335         if (r < 0)
2336                 goto out;
2337         r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
2338         if (r < 0)
2339                 goto out;
2340         data = ~0;
2341         r = kvm_write_guest_page(kvm, fn, &data,
2342                                  RMODE_TSS_SIZE - 2 * PAGE_SIZE - 1,
2343                                  sizeof(u8));
2344         if (r < 0)
2345                 goto out;
2346
2347         ret = 1;
2348 out:
2349         return ret;
2350 }
2351
2352 static int init_rmode_identity_map(struct kvm *kvm)
2353 {
2354         int i, r, ret;
2355         pfn_t identity_map_pfn;
2356         u32 tmp;
2357
2358         if (!enable_ept)
2359                 return 1;
2360         if (unlikely(!kvm->arch.ept_identity_pagetable)) {
2361                 printk(KERN_ERR "EPT: identity-mapping pagetable "
2362                         "haven't been allocated!\n");
2363                 return 0;
2364         }
2365         if (likely(kvm->arch.ept_identity_pagetable_done))
2366                 return 1;
2367         ret = 0;
2368         identity_map_pfn = kvm->arch.ept_identity_map_addr >> PAGE_SHIFT;
2369         r = kvm_clear_guest_page(kvm, identity_map_pfn, 0, PAGE_SIZE);
2370         if (r < 0)
2371                 goto out;
2372         /* Set up identity-mapping pagetable for EPT in real mode */
2373         for (i = 0; i < PT32_ENT_PER_PAGE; i++) {
2374                 tmp = (i << 22) + (_PAGE_PRESENT | _PAGE_RW | _PAGE_USER |
2375                         _PAGE_ACCESSED | _PAGE_DIRTY | _PAGE_PSE);
2376                 r = kvm_write_guest_page(kvm, identity_map_pfn,
2377                                 &tmp, i * sizeof(tmp), sizeof(tmp));
2378                 if (r < 0)
2379                         goto out;
2380         }
2381         kvm->arch.ept_identity_pagetable_done = true;
2382         ret = 1;
2383 out:
2384         return ret;
2385 }
2386
2387 static void seg_setup(int seg)
2388 {
2389         struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
2390         unsigned int ar;
2391
2392         vmcs_write16(sf->selector, 0);
2393         vmcs_writel(sf->base, 0);
2394         vmcs_write32(sf->limit, 0xffff);
2395         if (enable_unrestricted_guest) {
2396                 ar = 0x93;
2397                 if (seg == VCPU_SREG_CS)
2398                         ar |= 0x08; /* code segment */
2399         } else
2400                 ar = 0xf3;
2401
2402         vmcs_write32(sf->ar_bytes, ar);
2403 }
2404
2405 static int alloc_apic_access_page(struct kvm *kvm)
2406 {
2407         struct kvm_userspace_memory_region kvm_userspace_mem;
2408         int r = 0;
2409
2410         mutex_lock(&kvm->slots_lock);
2411         if (kvm->arch.apic_access_page)
2412                 goto out;
2413         kvm_userspace_mem.slot = APIC_ACCESS_PAGE_PRIVATE_MEMSLOT;
2414         kvm_userspace_mem.flags = 0;
2415         kvm_userspace_mem.guest_phys_addr = 0xfee00000ULL;
2416         kvm_userspace_mem.memory_size = PAGE_SIZE;
2417         r = __kvm_set_memory_region(kvm, &kvm_userspace_mem, 0);
2418         if (r)
2419                 goto out;
2420
2421         kvm->arch.apic_access_page = gfn_to_page(kvm, 0xfee00);
2422 out:
2423         mutex_unlock(&kvm->slots_lock);
2424         return r;
2425 }
2426
2427 static int alloc_identity_pagetable(struct kvm *kvm)
2428 {
2429         struct kvm_userspace_memory_region kvm_userspace_mem;
2430         int r = 0;
2431
2432         mutex_lock(&kvm->slots_lock);
2433         if (kvm->arch.ept_identity_pagetable)
2434                 goto out;
2435         kvm_userspace_mem.slot = IDENTITY_PAGETABLE_PRIVATE_MEMSLOT;
2436         kvm_userspace_mem.flags = 0;
2437         kvm_userspace_mem.guest_phys_addr =
2438                 kvm->arch.ept_identity_map_addr;
2439         kvm_userspace_mem.memory_size = PAGE_SIZE;
2440         r = __kvm_set_memory_region(kvm, &kvm_userspace_mem, 0);
2441         if (r)
2442                 goto out;
2443
2444         kvm->arch.ept_identity_pagetable = gfn_to_page(kvm,
2445                         kvm->arch.ept_identity_map_addr >> PAGE_SHIFT);
2446 out:
2447         mutex_unlock(&kvm->slots_lock);
2448         return r;
2449 }
2450
2451 static void allocate_vpid(struct vcpu_vmx *vmx)
2452 {
2453         int vpid;
2454
2455         vmx->vpid = 0;
2456         if (!enable_vpid)
2457                 return;
2458         spin_lock(&vmx_vpid_lock);
2459         vpid = find_first_zero_bit(vmx_vpid_bitmap, VMX_NR_VPIDS);
2460         if (vpid < VMX_NR_VPIDS) {
2461                 vmx->vpid = vpid;
2462                 __set_bit(vpid, vmx_vpid_bitmap);
2463         }
2464         spin_unlock(&vmx_vpid_lock);
2465 }
2466
2467 static void free_vpid(struct vcpu_vmx *vmx)
2468 {
2469         if (!enable_vpid)
2470                 return;
2471         spin_lock(&vmx_vpid_lock);
2472         if (vmx->vpid != 0)
2473                 __clear_bit(vmx->vpid, vmx_vpid_bitmap);
2474         spin_unlock(&vmx_vpid_lock);
2475 }
2476
2477 static void __vmx_disable_intercept_for_msr(unsigned long *msr_bitmap, u32 msr)
2478 {
2479         int f = sizeof(unsigned long);
2480
2481         if (!cpu_has_vmx_msr_bitmap())
2482                 return;
2483
2484         /*
2485          * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
2486          * have the write-low and read-high bitmap offsets the wrong way round.
2487          * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
2488          */
2489         if (msr <= 0x1fff) {
2490                 __clear_bit(msr, msr_bitmap + 0x000 / f); /* read-low */
2491                 __clear_bit(msr, msr_bitmap + 0x800 / f); /* write-low */
2492         } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
2493                 msr &= 0x1fff;
2494                 __clear_bit(msr, msr_bitmap + 0x400 / f); /* read-high */
2495                 __clear_bit(msr, msr_bitmap + 0xc00 / f); /* write-high */
2496         }
2497 }
2498
2499 static void vmx_disable_intercept_for_msr(u32 msr, bool longmode_only)
2500 {
2501         if (!longmode_only)
2502                 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_legacy, msr);
2503         __vmx_disable_intercept_for_msr(vmx_msr_bitmap_longmode, msr);
2504 }
2505
2506 /*
2507  * Sets up the vmcs for emulated real mode.
2508  */
2509 static int vmx_vcpu_setup(struct vcpu_vmx *vmx)
2510 {
2511         u32 host_sysenter_cs, msr_low, msr_high;
2512         u32 junk;
2513         u64 host_pat;
2514         unsigned long a;
2515         struct desc_ptr dt;
2516         int i;
2517         unsigned long kvm_vmx_return;
2518         u32 exec_control;
2519
2520         /* I/O */
2521         vmcs_write64(IO_BITMAP_A, __pa(vmx_io_bitmap_a));
2522         vmcs_write64(IO_BITMAP_B, __pa(vmx_io_bitmap_b));
2523
2524         if (cpu_has_vmx_msr_bitmap())
2525                 vmcs_write64(MSR_BITMAP, __pa(vmx_msr_bitmap_legacy));
2526
2527         vmcs_write64(VMCS_LINK_POINTER, -1ull); /* 22.3.1.5 */
2528
2529         /* Control */
2530         vmcs_write32(PIN_BASED_VM_EXEC_CONTROL,
2531                 vmcs_config.pin_based_exec_ctrl);
2532
2533         exec_control = vmcs_config.cpu_based_exec_ctrl;
2534         if (!vm_need_tpr_shadow(vmx->vcpu.kvm)) {
2535                 exec_control &= ~CPU_BASED_TPR_SHADOW;
2536 #ifdef CONFIG_X86_64
2537                 exec_control |= CPU_BASED_CR8_STORE_EXITING |
2538                                 CPU_BASED_CR8_LOAD_EXITING;
2539 #endif
2540         }
2541         if (!enable_ept)
2542                 exec_control |= CPU_BASED_CR3_STORE_EXITING |
2543                                 CPU_BASED_CR3_LOAD_EXITING  |
2544                                 CPU_BASED_INVLPG_EXITING;
2545         vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, exec_control);
2546
2547         if (cpu_has_secondary_exec_ctrls()) {
2548                 exec_control = vmcs_config.cpu_based_2nd_exec_ctrl;
2549                 if (!vm_need_virtualize_apic_accesses(vmx->vcpu.kvm))
2550                         exec_control &=
2551                                 ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
2552                 if (vmx->vpid == 0)
2553                         exec_control &= ~SECONDARY_EXEC_ENABLE_VPID;
2554                 if (!enable_ept) {
2555                         exec_control &= ~SECONDARY_EXEC_ENABLE_EPT;
2556                         enable_unrestricted_guest = 0;
2557                 }
2558                 if (!enable_unrestricted_guest)
2559                         exec_control &= ~SECONDARY_EXEC_UNRESTRICTED_GUEST;
2560                 if (!ple_gap)
2561                         exec_control &= ~SECONDARY_EXEC_PAUSE_LOOP_EXITING;
2562                 vmcs_write32(SECONDARY_VM_EXEC_CONTROL, exec_control);
2563         }
2564
2565         if (ple_gap) {
2566                 vmcs_write32(PLE_GAP, ple_gap);
2567                 vmcs_write32(PLE_WINDOW, ple_window);
2568         }
2569
2570         vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, !!bypass_guest_pf);
2571         vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, !!bypass_guest_pf);
2572         vmcs_write32(CR3_TARGET_COUNT, 0);           /* 22.2.1 */
2573
2574         vmcs_writel(HOST_CR0, read_cr0() | X86_CR0_TS);  /* 22.2.3 */
2575         vmcs_writel(HOST_CR4, read_cr4());  /* 22.2.3, 22.2.5 */
2576         vmcs_writel(HOST_CR3, read_cr3());  /* 22.2.3  FIXME: shadow tables */
2577
2578         vmcs_write16(HOST_CS_SELECTOR, __KERNEL_CS);  /* 22.2.4 */
2579         vmcs_write16(HOST_DS_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
2580         vmcs_write16(HOST_ES_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
2581         vmcs_write16(HOST_FS_SELECTOR, 0);            /* 22.2.4 */
2582         vmcs_write16(HOST_GS_SELECTOR, 0);            /* 22.2.4 */
2583         vmcs_write16(HOST_SS_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
2584 #ifdef CONFIG_X86_64
2585         rdmsrl(MSR_FS_BASE, a);
2586         vmcs_writel(HOST_FS_BASE, a); /* 22.2.4 */
2587         rdmsrl(MSR_GS_BASE, a);
2588         vmcs_writel(HOST_GS_BASE, a); /* 22.2.4 */
2589 #else
2590         vmcs_writel(HOST_FS_BASE, 0); /* 22.2.4 */
2591         vmcs_writel(HOST_GS_BASE, 0); /* 22.2.4 */
2592 #endif
2593
2594         vmcs_write16(HOST_TR_SELECTOR, GDT_ENTRY_TSS*8);  /* 22.2.4 */
2595
2596         native_store_idt(&dt);
2597         vmcs_writel(HOST_IDTR_BASE, dt.address);   /* 22.2.4 */
2598
2599         asm("mov $.Lkvm_vmx_return, %0" : "=r"(kvm_vmx_return));
2600         vmcs_writel(HOST_RIP, kvm_vmx_return); /* 22.2.5 */
2601         vmcs_write32(VM_EXIT_MSR_STORE_COUNT, 0);
2602         vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, 0);
2603         vmcs_write64(VM_EXIT_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.host));
2604         vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, 0);
2605         vmcs_write64(VM_ENTRY_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.guest));
2606
2607         rdmsr(MSR_IA32_SYSENTER_CS, host_sysenter_cs, junk);
2608         vmcs_write32(HOST_IA32_SYSENTER_CS, host_sysenter_cs);
2609         rdmsrl(MSR_IA32_SYSENTER_ESP, a);
2610         vmcs_writel(HOST_IA32_SYSENTER_ESP, a);   /* 22.2.3 */
2611         rdmsrl(MSR_IA32_SYSENTER_EIP, a);
2612         vmcs_writel(HOST_IA32_SYSENTER_EIP, a);   /* 22.2.3 */
2613
2614         if (vmcs_config.vmexit_ctrl & VM_EXIT_LOAD_IA32_PAT) {
2615                 rdmsr(MSR_IA32_CR_PAT, msr_low, msr_high);
2616                 host_pat = msr_low | ((u64) msr_high << 32);
2617                 vmcs_write64(HOST_IA32_PAT, host_pat);
2618         }
2619         if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
2620                 rdmsr(MSR_IA32_CR_PAT, msr_low, msr_high);
2621                 host_pat = msr_low | ((u64) msr_high << 32);
2622                 /* Write the default value follow host pat */
2623                 vmcs_write64(GUEST_IA32_PAT, host_pat);
2624                 /* Keep arch.pat sync with GUEST_IA32_PAT */
2625                 vmx->vcpu.arch.pat = host_pat;
2626         }
2627
2628         for (i = 0; i < NR_VMX_MSR; ++i) {
2629                 u32 index = vmx_msr_index[i];
2630                 u32 data_low, data_high;
2631                 int j = vmx->nmsrs;
2632
2633                 if (rdmsr_safe(index, &data_low, &data_high) < 0)
2634                         continue;
2635                 if (wrmsr_safe(index, data_low, data_high) < 0)
2636                         continue;
2637                 vmx->guest_msrs[j].index = i;
2638                 vmx->guest_msrs[j].data = 0;
2639                 vmx->guest_msrs[j].mask = -1ull;
2640                 ++vmx->nmsrs;
2641         }
2642
2643         vmcs_write32(VM_EXIT_CONTROLS, vmcs_config.vmexit_ctrl);
2644
2645         /* 22.2.1, 20.8.1 */
2646         vmcs_write32(VM_ENTRY_CONTROLS, vmcs_config.vmentry_ctrl);
2647
2648         vmcs_writel(CR0_GUEST_HOST_MASK, ~0UL);
2649         vmx->vcpu.arch.cr4_guest_owned_bits = KVM_CR4_GUEST_OWNED_BITS;
2650         if (enable_ept)
2651                 vmx->vcpu.arch.cr4_guest_owned_bits |= X86_CR4_PGE;
2652         vmcs_writel(CR4_GUEST_HOST_MASK, ~vmx->vcpu.arch.cr4_guest_owned_bits);
2653
2654         kvm_write_tsc(&vmx->vcpu, 0);
2655
2656         return 0;
2657 }
2658
2659 static int init_rmode(struct kvm *kvm)
2660 {
2661         int idx, ret = 0;
2662
2663         idx = srcu_read_lock(&kvm->srcu);
2664         if (!init_rmode_tss(kvm))
2665                 goto exit;
2666         if (!init_rmode_identity_map(kvm))
2667                 goto exit;
2668
2669         ret = 1;
2670 exit:
2671         srcu_read_unlock(&kvm->srcu, idx);
2672         return ret;
2673 }
2674
2675 static int vmx_vcpu_reset(struct kvm_vcpu *vcpu)
2676 {
2677         struct vcpu_vmx *vmx = to_vmx(vcpu);
2678         u64 msr;
2679         int ret;
2680
2681         vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP));
2682         if (!init_rmode(vmx->vcpu.kvm)) {
2683                 ret = -ENOMEM;
2684                 goto out;
2685         }
2686
2687         vmx->rmode.vm86_active = 0;
2688
2689         vmx->soft_vnmi_blocked = 0;
2690
2691         vmx->vcpu.arch.regs[VCPU_REGS_RDX] = get_rdx_init_val();
2692         kvm_set_cr8(&vmx->vcpu, 0);
2693         msr = 0xfee00000 | MSR_IA32_APICBASE_ENABLE;
2694         if (kvm_vcpu_is_bsp(&vmx->vcpu))
2695                 msr |= MSR_IA32_APICBASE_BSP;
2696         kvm_set_apic_base(&vmx->vcpu, msr);
2697
2698         ret = fx_init(&vmx->vcpu);
2699         if (ret != 0)
2700                 goto out;
2701
2702         seg_setup(VCPU_SREG_CS);
2703         /*
2704          * GUEST_CS_BASE should really be 0xffff0000, but VT vm86 mode
2705          * insists on having GUEST_CS_BASE == GUEST_CS_SELECTOR << 4.  Sigh.
2706          */
2707         if (kvm_vcpu_is_bsp(&vmx->vcpu)) {
2708                 vmcs_write16(GUEST_CS_SELECTOR, 0xf000);
2709                 vmcs_writel(GUEST_CS_BASE, 0x000f0000);
2710         } else {
2711                 vmcs_write16(GUEST_CS_SELECTOR, vmx->vcpu.arch.sipi_vector << 8);
2712                 vmcs_writel(GUEST_CS_BASE, vmx->vcpu.arch.sipi_vector << 12);
2713         }
2714
2715         seg_setup(VCPU_SREG_DS);
2716         seg_setup(VCPU_SREG_ES);
2717         seg_setup(VCPU_SREG_FS);
2718         seg_setup(VCPU_SREG_GS);
2719         seg_setup(VCPU_SREG_SS);
2720
2721         vmcs_write16(GUEST_TR_SELECTOR, 0);
2722         vmcs_writel(GUEST_TR_BASE, 0);
2723         vmcs_write32(GUEST_TR_LIMIT, 0xffff);
2724         vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
2725
2726         vmcs_write16(GUEST_LDTR_SELECTOR, 0);
2727         vmcs_writel(GUEST_LDTR_BASE, 0);
2728         vmcs_write32(GUEST_LDTR_LIMIT, 0xffff);
2729         vmcs_write32(GUEST_LDTR_AR_BYTES, 0x00082);
2730
2731         vmcs_write32(GUEST_SYSENTER_CS, 0);
2732         vmcs_writel(GUEST_SYSENTER_ESP, 0);
2733         vmcs_writel(GUEST_SYSENTER_EIP, 0);
2734
2735         vmcs_writel(GUEST_RFLAGS, 0x02);
2736         if (kvm_vcpu_is_bsp(&vmx->vcpu))
2737                 kvm_rip_write(vcpu, 0xfff0);
2738         else
2739                 kvm_rip_write(vcpu, 0);
2740         kvm_register_write(vcpu, VCPU_REGS_RSP, 0);
2741
2742         vmcs_writel(GUEST_DR7, 0x400);
2743
2744         vmcs_writel(GUEST_GDTR_BASE, 0);
2745         vmcs_write32(GUEST_GDTR_LIMIT, 0xffff);
2746
2747         vmcs_writel(GUEST_IDTR_BASE, 0);
2748         vmcs_write32(GUEST_IDTR_LIMIT, 0xffff);
2749
2750         vmcs_write32(GUEST_ACTIVITY_STATE, GUEST_ACTIVITY_ACTIVE);
2751         vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, 0);
2752         vmcs_write32(GUEST_PENDING_DBG_EXCEPTIONS, 0);
2753
2754         /* Special registers */
2755         vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
2756
2757         setup_msrs(vmx);
2758
2759         vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);  /* 22.2.1 */
2760
2761         if (cpu_has_vmx_tpr_shadow()) {
2762                 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, 0);
2763                 if (vm_need_tpr_shadow(vmx->vcpu.kvm))
2764                         vmcs_write64(VIRTUAL_APIC_PAGE_ADDR,
2765                                 page_to_phys(vmx->vcpu.arch.apic->regs_page));
2766                 vmcs_write32(TPR_THRESHOLD, 0);
2767         }
2768
2769         if (vm_need_virtualize_apic_accesses(vmx->vcpu.kvm))
2770                 vmcs_write64(APIC_ACCESS_ADDR,
2771                              page_to_phys(vmx->vcpu.kvm->arch.apic_access_page));
2772
2773         if (vmx->vpid != 0)
2774                 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
2775
2776         vmx->vcpu.arch.cr0 = X86_CR0_NW | X86_CR0_CD | X86_CR0_ET;
2777         vmx_set_cr0(&vmx->vcpu, kvm_read_cr0(vcpu)); /* enter rmode */
2778         vmx_set_cr4(&vmx->vcpu, 0);
2779         vmx_set_efer(&vmx->vcpu, 0);
2780         vmx_fpu_activate(&vmx->vcpu);
2781         update_exception_bitmap(&vmx->vcpu);
2782
2783         vpid_sync_context(vmx);
2784
2785         ret = 0;
2786
2787         /* HACK: Don't enable emulation on guest boot/reset */
2788         vmx->emulation_required = 0;
2789
2790 out:
2791         return ret;
2792 }
2793
2794 static void enable_irq_window(struct kvm_vcpu *vcpu)
2795 {
2796         u32 cpu_based_vm_exec_control;
2797
2798         cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
2799         cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_INTR_PENDING;
2800         vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
2801 }
2802
2803 static void enable_nmi_window(struct kvm_vcpu *vcpu)
2804 {
2805         u32 cpu_based_vm_exec_control;
2806
2807         if (!cpu_has_virtual_nmis()) {
2808                 enable_irq_window(vcpu);
2809                 return;
2810         }
2811
2812         if (vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_STI) {
2813                 enable_irq_window(vcpu);
2814                 return;
2815         }
2816         cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
2817         cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_NMI_PENDING;
2818         vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
2819 }
2820
2821 static void vmx_inject_irq(struct kvm_vcpu *vcpu)
2822 {
2823         struct vcpu_vmx *vmx = to_vmx(vcpu);
2824         uint32_t intr;
2825         int irq = vcpu->arch.interrupt.nr;
2826
2827         trace_kvm_inj_virq(irq);
2828
2829         ++vcpu->stat.irq_injections;
2830         if (vmx->rmode.vm86_active) {
2831                 if (kvm_inject_realmode_interrupt(vcpu, irq) != EMULATE_DONE)
2832                         kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
2833                 return;
2834         }
2835         intr = irq | INTR_INFO_VALID_MASK;
2836         if (vcpu->arch.interrupt.soft) {
2837                 intr |= INTR_TYPE_SOFT_INTR;
2838                 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
2839                              vmx->vcpu.arch.event_exit_inst_len);
2840         } else
2841                 intr |= INTR_TYPE_EXT_INTR;
2842         vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr);
2843         vmx_clear_hlt(vcpu);
2844 }
2845
2846 static void vmx_inject_nmi(struct kvm_vcpu *vcpu)
2847 {
2848         struct vcpu_vmx *vmx = to_vmx(vcpu);
2849
2850         if (!cpu_has_virtual_nmis()) {
2851                 /*
2852                  * Tracking the NMI-blocked state in software is built upon
2853                  * finding the next open IRQ window. This, in turn, depends on
2854                  * well-behaving guests: They have to keep IRQs disabled at
2855                  * least as long as the NMI handler runs. Otherwise we may
2856                  * cause NMI nesting, maybe breaking the guest. But as this is
2857                  * highly unlikely, we can live with the residual risk.
2858                  */
2859                 vmx->soft_vnmi_blocked = 1;
2860                 vmx->vnmi_blocked_time = 0;
2861         }
2862
2863         ++vcpu->stat.nmi_injections;
2864         if (vmx->rmode.vm86_active) {
2865                 if (kvm_inject_realmode_interrupt(vcpu, NMI_VECTOR) != EMULATE_DONE)
2866                         kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
2867                 return;
2868         }
2869         vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
2870                         INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR);
2871         vmx_clear_hlt(vcpu);
2872 }
2873
2874 static int vmx_nmi_allowed(struct kvm_vcpu *vcpu)
2875 {
2876         if (!cpu_has_virtual_nmis() && to_vmx(vcpu)->soft_vnmi_blocked)
2877                 return 0;
2878
2879         return  !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
2880                   (GUEST_INTR_STATE_MOV_SS | GUEST_INTR_STATE_STI
2881                    | GUEST_INTR_STATE_NMI));
2882 }
2883
2884 static bool vmx_get_nmi_mask(struct kvm_vcpu *vcpu)
2885 {
2886         if (!cpu_has_virtual_nmis())
2887                 return to_vmx(vcpu)->soft_vnmi_blocked;
2888         return vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_NMI;
2889 }
2890
2891 static void vmx_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked)
2892 {
2893         struct vcpu_vmx *vmx = to_vmx(vcpu);
2894
2895         if (!cpu_has_virtual_nmis()) {
2896                 if (vmx->soft_vnmi_blocked != masked) {
2897                         vmx->soft_vnmi_blocked = masked;
2898                         vmx->vnmi_blocked_time = 0;
2899                 }
2900         } else {
2901                 if (masked)
2902                         vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
2903                                       GUEST_INTR_STATE_NMI);
2904                 else
2905                         vmcs_clear_bits(GUEST_INTERRUPTIBILITY_INFO,
2906                                         GUEST_INTR_STATE_NMI);
2907         }
2908 }
2909
2910 static int vmx_interrupt_allowed(struct kvm_vcpu *vcpu)
2911 {
2912         return (vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_IF) &&
2913                 !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
2914                         (GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS));
2915 }
2916
2917 static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr)
2918 {
2919         int ret;
2920         struct kvm_userspace_memory_region tss_mem = {
2921                 .slot = TSS_PRIVATE_MEMSLOT,
2922                 .guest_phys_addr = addr,
2923                 .memory_size = PAGE_SIZE * 3,
2924                 .flags = 0,
2925         };
2926
2927         ret = kvm_set_memory_region(kvm, &tss_mem, 0);
2928         if (ret)
2929                 return ret;
2930         kvm->arch.tss_addr = addr;
2931         return 0;
2932 }
2933
2934 static int handle_rmode_exception(struct kvm_vcpu *vcpu,
2935                                   int vec, u32 err_code)
2936 {
2937         /*
2938          * Instruction with address size override prefix opcode 0x67
2939          * Cause the #SS fault with 0 error code in VM86 mode.
2940          */
2941         if (((vec == GP_VECTOR) || (vec == SS_VECTOR)) && err_code == 0)
2942                 if (emulate_instruction(vcpu, 0, 0, 0) == EMULATE_DONE)
2943                         return 1;
2944         /*
2945          * Forward all other exceptions that are valid in real mode.
2946          * FIXME: Breaks guest debugging in real mode, needs to be fixed with
2947          *        the required debugging infrastructure rework.
2948          */
2949         switch (vec) {
2950         case DB_VECTOR:
2951                 if (vcpu->guest_debug &
2952                     (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
2953                         return 0;
2954                 kvm_queue_exception(vcpu, vec);
2955                 return 1;
2956         case BP_VECTOR:
2957                 /*
2958                  * Update instruction length as we may reinject the exception
2959                  * from user space while in guest debugging mode.
2960                  */
2961                 to_vmx(vcpu)->vcpu.arch.event_exit_inst_len =
2962                         vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
2963                 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
2964                         return 0;
2965                 /* fall through */
2966         case DE_VECTOR:
2967         case OF_VECTOR:
2968         case BR_VECTOR:
2969         case UD_VECTOR:
2970         case DF_VECTOR:
2971         case SS_VECTOR:
2972         case GP_VECTOR:
2973         case MF_VECTOR:
2974                 kvm_queue_exception(vcpu, vec);
2975                 return 1;
2976         }
2977         return 0;
2978 }
2979
2980 /*
2981  * Trigger machine check on the host. We assume all the MSRs are already set up
2982  * by the CPU and that we still run on the same CPU as the MCE occurred on.
2983  * We pass a fake environment to the machine check handler because we want
2984  * the guest to be always treated like user space, no matter what context
2985  * it used internally.
2986  */
2987 static void kvm_machine_check(void)
2988 {
2989 #if defined(CONFIG_X86_MCE) && defined(CONFIG_X86_64)
2990         struct pt_regs regs = {
2991                 .cs = 3, /* Fake ring 3 no matter what the guest ran on */
2992                 .flags = X86_EFLAGS_IF,
2993         };
2994
2995         do_machine_check(&regs, 0);
2996 #endif
2997 }
2998
2999 static int handle_machine_check(struct kvm_vcpu *vcpu)
3000 {
3001         /* already handled by vcpu_run */
3002         return 1;
3003 }
3004
3005 static int handle_exception(struct kvm_vcpu *vcpu)
3006 {
3007         struct vcpu_vmx *vmx = to_vmx(vcpu);
3008         struct kvm_run *kvm_run = vcpu->run;
3009         u32 intr_info, ex_no, error_code;
3010         unsigned long cr2, rip, dr6;
3011         u32 vect_info;
3012         enum emulation_result er;
3013
3014         vect_info = vmx->idt_vectoring_info;
3015         intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
3016
3017         if (is_machine_check(intr_info))
3018                 return handle_machine_check(vcpu);
3019
3020         if ((vect_info & VECTORING_INFO_VALID_MASK) &&
3021             !is_page_fault(intr_info)) {
3022                 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
3023                 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_SIMUL_EX;
3024                 vcpu->run->internal.ndata = 2;
3025                 vcpu->run->internal.data[0] = vect_info;
3026                 vcpu->run->internal.data[1] = intr_info;
3027                 return 0;
3028         }
3029
3030         if ((intr_info & INTR_INFO_INTR_TYPE_MASK) == INTR_TYPE_NMI_INTR)
3031                 return 1;  /* already handled by vmx_vcpu_run() */
3032
3033         if (is_no_device(intr_info)) {
3034                 vmx_fpu_activate(vcpu);
3035                 return 1;
3036         }
3037
3038         if (is_invalid_opcode(intr_info)) {
3039                 er = emulate_instruction(vcpu, 0, 0, EMULTYPE_TRAP_UD);
3040                 if (er != EMULATE_DONE)
3041                         kvm_queue_exception(vcpu, UD_VECTOR);
3042                 return 1;
3043         }
3044
3045         error_code = 0;
3046         rip = kvm_rip_read(vcpu);
3047         if (intr_info & INTR_INFO_DELIVER_CODE_MASK)
3048                 error_code = vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
3049         if (is_page_fault(intr_info)) {
3050                 /* EPT won't cause page fault directly */
3051                 if (enable_ept)
3052                         BUG();
3053                 cr2 = vmcs_readl(EXIT_QUALIFICATION);
3054                 trace_kvm_page_fault(cr2, error_code);
3055
3056                 if (kvm_event_needs_reinjection(vcpu))
3057                         kvm_mmu_unprotect_page_virt(vcpu, cr2);
3058                 return kvm_mmu_page_fault(vcpu, cr2, error_code);
3059         }
3060
3061         if (vmx->rmode.vm86_active &&
3062             handle_rmode_exception(vcpu, intr_info & INTR_INFO_VECTOR_MASK,
3063                                                                 error_code)) {
3064                 if (vcpu->arch.halt_request) {
3065                         vcpu->arch.halt_request = 0;
3066                         return kvm_emulate_halt(vcpu);
3067                 }
3068                 return 1;
3069         }
3070
3071         ex_no = intr_info & INTR_INFO_VECTOR_MASK;
3072         switch (ex_no) {
3073         case DB_VECTOR:
3074                 dr6 = vmcs_readl(EXIT_QUALIFICATION);
3075                 if (!(vcpu->guest_debug &
3076                       (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))) {
3077                         vcpu->arch.dr6 = dr6 | DR6_FIXED_1;
3078                         kvm_queue_exception(vcpu, DB_VECTOR);
3079                         return 1;
3080                 }
3081                 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1;
3082                 kvm_run->debug.arch.dr7 = vmcs_readl(GUEST_DR7);
3083                 /* fall through */
3084         case BP_VECTOR:
3085                 /*
3086                  * Update instruction length as we may reinject #BP from
3087                  * user space while in guest debugging mode. Reading it for
3088                  * #DB as well causes no harm, it is not used in that case.
3089                  */
3090                 vmx->vcpu.arch.event_exit_inst_len =
3091                         vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
3092                 kvm_run->exit_reason = KVM_EXIT_DEBUG;
3093                 kvm_run->debug.arch.pc = vmcs_readl(GUEST_CS_BASE) + rip;
3094                 kvm_run->debug.arch.exception = ex_no;
3095                 break;
3096         default:
3097                 kvm_run->exit_reason = KVM_EXIT_EXCEPTION;
3098                 kvm_run->ex.exception = ex_no;
3099                 kvm_run->ex.error_code = error_code;
3100                 break;
3101         }
3102         return 0;
3103 }
3104
3105 static int handle_external_interrupt(struct kvm_vcpu *vcpu)
3106 {
3107         ++vcpu->stat.irq_exits;
3108         return 1;
3109 }
3110
3111 static int handle_triple_fault(struct kvm_vcpu *vcpu)
3112 {
3113         vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
3114         return 0;
3115 }
3116
3117 static int handle_io(struct kvm_vcpu *vcpu)
3118 {
3119         unsigned long exit_qualification;
3120         int size, in, string;
3121         unsigned port;
3122
3123         exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
3124         string = (exit_qualification & 16) != 0;
3125         in = (exit_qualification & 8) != 0;
3126
3127         ++vcpu->stat.io_exits;
3128
3129         if (string || in)
3130                 return emulate_instruction(vcpu, 0, 0, 0) == EMULATE_DONE;
3131
3132         port = exit_qualification >> 16;
3133         size = (exit_qualification & 7) + 1;
3134         skip_emulated_instruction(vcpu);
3135
3136         return kvm_fast_pio_out(vcpu, size, port);
3137 }
3138
3139 static void
3140 vmx_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
3141 {
3142         /*
3143          * Patch in the VMCALL instruction:
3144          */
3145         hypercall[0] = 0x0f;
3146         hypercall[1] = 0x01;
3147         hypercall[2] = 0xc1;
3148 }
3149
3150 static void complete_insn_gp(struct kvm_vcpu *vcpu, int err)
3151 {
3152         if (err)
3153                 kvm_inject_gp(vcpu, 0);
3154         else
3155                 skip_emulated_instruction(vcpu);
3156 }
3157
3158 static int handle_cr(struct kvm_vcpu *vcpu)
3159 {
3160         unsigned long exit_qualification, val;
3161         int cr;
3162         int reg;
3163         int err;
3164
3165         exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
3166         cr = exit_qualification & 15;
3167         reg = (exit_qualification >> 8) & 15;
3168         switch ((exit_qualification >> 4) & 3) {
3169         case 0: /* mov to cr */
3170                 val = kvm_register_read(vcpu, reg);
3171                 trace_kvm_cr_write(cr, val);
3172                 switch (cr) {
3173                 case 0:
3174                         err = kvm_set_cr0(vcpu, val);
3175                         complete_insn_gp(vcpu, err);
3176                         return 1;
3177                 case 3:
3178                         err = kvm_set_cr3(vcpu, val);
3179                         complete_insn_gp(vcpu, err);
3180                         return 1;
3181                 case 4:
3182                         err = kvm_set_cr4(vcpu, val);
3183                         complete_insn_gp(vcpu, err);
3184                         return 1;
3185                 case 8: {
3186                                 u8 cr8_prev = kvm_get_cr8(vcpu);
3187                                 u8 cr8 = kvm_register_read(vcpu, reg);
3188                                 err = kvm_set_cr8(vcpu, cr8);
3189                                 complete_insn_gp(vcpu, err);
3190                                 if (irqchip_in_kernel(vcpu->kvm))
3191                                         return 1;
3192                                 if (cr8_prev <= cr8)
3193                                         return 1;
3194                                 vcpu->run->exit_reason = KVM_EXIT_SET_TPR;
3195                                 return 0;
3196                         }
3197                 };
3198                 break;
3199         case 2: /* clts */
3200                 vmx_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~X86_CR0_TS));
3201                 trace_kvm_cr_write(0, kvm_read_cr0(vcpu));
3202                 skip_emulated_instruction(vcpu);
3203                 vmx_fpu_activate(vcpu);
3204                 return 1;
3205         case 1: /*mov from cr*/
3206                 switch (cr) {
3207                 case 3:
3208                         kvm_register_write(vcpu, reg, vcpu->arch.cr3);
3209                         trace_kvm_cr_read(cr, vcpu->arch.cr3);
3210                         skip_emulated_instruction(vcpu);
3211                         return 1;
3212                 case 8:
3213                         val = kvm_get_cr8(vcpu);
3214                         kvm_register_write(vcpu, reg, val);
3215                         trace_kvm_cr_read(cr, val);
3216                         skip_emulated_instruction(vcpu);
3217                         return 1;
3218                 }
3219                 break;
3220         case 3: /* lmsw */
3221                 val = (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f;
3222                 trace_kvm_cr_write(0, (kvm_read_cr0(vcpu) & ~0xful) | val);
3223                 kvm_lmsw(vcpu, val);
3224
3225                 skip_emulated_instruction(vcpu);
3226                 return 1;
3227         default:
3228                 break;
3229         }
3230         vcpu->run->exit_reason = 0;
3231         pr_unimpl(vcpu, "unhandled control register: op %d cr %d\n",
3232                (int)(exit_qualification >> 4) & 3, cr);
3233         return 0;
3234 }
3235
3236 static int handle_dr(struct kvm_vcpu *vcpu)
3237 {
3238         unsigned long exit_qualification;
3239         int dr, reg;
3240
3241         /* Do not handle if the CPL > 0, will trigger GP on re-entry */
3242         if (!kvm_require_cpl(vcpu, 0))
3243                 return 1;
3244         dr = vmcs_readl(GUEST_DR7);
3245         if (dr & DR7_GD) {
3246                 /*
3247                  * As the vm-exit takes precedence over the debug trap, we
3248                  * need to emulate the latter, either for the host or the
3249                  * guest debugging itself.
3250                  */
3251                 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
3252                         vcpu->run->debug.arch.dr6 = vcpu->arch.dr6;
3253                         vcpu->run->debug.arch.dr7 = dr;
3254                         vcpu->run->debug.arch.pc =
3255                                 vmcs_readl(GUEST_CS_BASE) +
3256                                 vmcs_readl(GUEST_RIP);
3257                         vcpu->run->debug.arch.exception = DB_VECTOR;
3258                         vcpu->run->exit_reason = KVM_EXIT_DEBUG;
3259                         return 0;
3260                 } else {
3261                         vcpu->arch.dr7 &= ~DR7_GD;
3262                         vcpu->arch.dr6 |= DR6_BD;
3263                         vmcs_writel(GUEST_DR7, vcpu->arch.dr7);
3264                         kvm_queue_exception(vcpu, DB_VECTOR);
3265                         return 1;
3266                 }
3267         }
3268
3269         exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
3270         dr = exit_qualification & DEBUG_REG_ACCESS_NUM;
3271         reg = DEBUG_REG_ACCESS_REG(exit_qualification);
3272         if (exit_qualification & TYPE_MOV_FROM_DR) {
3273                 unsigned long val;
3274                 if (!kvm_get_dr(vcpu, dr, &val))
3275                         kvm_register_write(vcpu, reg, val);
3276         } else
3277                 kvm_set_dr(vcpu, dr, vcpu->arch.regs[reg]);
3278         skip_emulated_instruction(vcpu);
3279         return 1;
3280 }
3281
3282 static void vmx_set_dr7(struct kvm_vcpu *vcpu, unsigned long val)
3283 {
3284         vmcs_writel(GUEST_DR7, val);
3285 }
3286
3287 static int handle_cpuid(struct kvm_vcpu *vcpu)
3288 {
3289         kvm_emulate_cpuid(vcpu);
3290         return 1;
3291 }
3292
3293 static int handle_rdmsr(struct kvm_vcpu *vcpu)
3294 {
3295         u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
3296         u64 data;
3297
3298         if (vmx_get_msr(vcpu, ecx, &data)) {
3299                 trace_kvm_msr_read_ex(ecx);
3300                 kvm_inject_gp(vcpu, 0);
3301                 return 1;
3302         }
3303
3304         trace_kvm_msr_read(ecx, data);
3305
3306         /* FIXME: handling of bits 32:63 of rax, rdx */
3307         vcpu->arch.regs[VCPU_REGS_RAX] = data & -1u;
3308         vcpu->arch.regs[VCPU_REGS_RDX] = (data >> 32) & -1u;
3309         skip_emulated_instruction(vcpu);
3310         return 1;
3311 }
3312
3313 static int handle_wrmsr(struct kvm_vcpu *vcpu)
3314 {
3315         u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
3316         u64 data = (vcpu->arch.regs[VCPU_REGS_RAX] & -1u)
3317                 | ((u64)(vcpu->arch.regs[VCPU_REGS_RDX] & -1u) << 32);
3318
3319         if (vmx_set_msr(vcpu, ecx, data) != 0) {
3320                 trace_kvm_msr_write_ex(ecx, data);
3321                 kvm_inject_gp(vcpu, 0);
3322                 return 1;
3323         }
3324
3325         trace_kvm_msr_write(ecx, data);
3326         skip_emulated_instruction(vcpu);
3327         return 1;
3328 }
3329
3330 static int handle_tpr_below_threshold(struct kvm_vcpu *vcpu)
3331 {
3332         kvm_make_request(KVM_REQ_EVENT, vcpu);
3333         return 1;
3334 }
3335
3336 static int handle_interrupt_window(struct kvm_vcpu *vcpu)
3337 {
3338         u32 cpu_based_vm_exec_control;
3339
3340         /* clear pending irq */
3341         cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
3342         cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_INTR_PENDING;
3343         vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
3344
3345         kvm_make_request(KVM_REQ_EVENT, vcpu);
3346
3347         ++vcpu->stat.irq_window_exits;
3348
3349         /*
3350          * If the user space waits to inject interrupts, exit as soon as
3351          * possible
3352          */
3353         if (!irqchip_in_kernel(vcpu->kvm) &&
3354             vcpu->run->request_interrupt_window &&
3355             !kvm_cpu_has_interrupt(vcpu)) {
3356                 vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
3357                 return 0;
3358         }
3359         return 1;
3360 }
3361
3362 static int handle_halt(struct kvm_vcpu *vcpu)
3363 {
3364         skip_emulated_instruction(vcpu);
3365         return kvm_emulate_halt(vcpu);
3366 }
3367
3368 static int handle_vmcall(struct kvm_vcpu *vcpu)
3369 {
3370         skip_emulated_instruction(vcpu);
3371         kvm_emulate_hypercall(vcpu);
3372         return 1;
3373 }
3374
3375 static int handle_vmx_insn(struct kvm_vcpu *vcpu)
3376 {
3377         kvm_queue_exception(vcpu, UD_VECTOR);
3378         return 1;
3379 }
3380
3381 static int handle_invd(struct kvm_vcpu *vcpu)
3382 {
3383         return emulate_instruction(vcpu, 0, 0, 0) == EMULATE_DONE;
3384 }
3385
3386 static int handle_invlpg(struct kvm_vcpu *vcpu)
3387 {
3388         unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
3389
3390         kvm_mmu_invlpg(vcpu, exit_qualification);
3391         skip_emulated_instruction(vcpu);
3392         return 1;
3393 }
3394
3395 static int handle_wbinvd(struct kvm_vcpu *vcpu)
3396 {
3397         skip_emulated_instruction(vcpu);
3398         kvm_emulate_wbinvd(vcpu);
3399         return 1;
3400 }
3401
3402 static int handle_xsetbv(struct kvm_vcpu *vcpu)
3403 {
3404         u64 new_bv = kvm_read_edx_eax(vcpu);
3405         u32 index = kvm_register_read(vcpu, VCPU_REGS_RCX);
3406
3407         if (kvm_set_xcr(vcpu, index, new_bv) == 0)
3408                 skip_emulated_instruction(vcpu);
3409         return 1;
3410 }
3411
3412 static int handle_apic_access(struct kvm_vcpu *vcpu)
3413 {
3414         return emulate_instruction(vcpu, 0, 0, 0) == EMULATE_DONE;
3415 }
3416
3417 static int handle_task_switch(struct kvm_vcpu *vcpu)
3418 {
3419         struct vcpu_vmx *vmx = to_vmx(vcpu);
3420         unsigned long exit_qualification;
3421         bool has_error_code = false;
3422         u32 error_code = 0;
3423         u16 tss_selector;
3424         int reason, type, idt_v;
3425
3426         idt_v = (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK);
3427         type = (vmx->idt_vectoring_info & VECTORING_INFO_TYPE_MASK);
3428
3429         exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
3430
3431         reason = (u32)exit_qualification >> 30;
3432         if (reason == TASK_SWITCH_GATE && idt_v) {
3433                 switch (type) {
3434                 case INTR_TYPE_NMI_INTR:
3435                         vcpu->arch.nmi_injected = false;
3436                         if (cpu_has_virtual_nmis())
3437                                 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
3438                                               GUEST_INTR_STATE_NMI);
3439                         break;
3440                 case INTR_TYPE_EXT_INTR:
3441                 case INTR_TYPE_SOFT_INTR:
3442                         kvm_clear_interrupt_queue(vcpu);
3443                         break;
3444                 case INTR_TYPE_HARD_EXCEPTION:
3445                         if (vmx->idt_vectoring_info &
3446                             VECTORING_INFO_DELIVER_CODE_MASK) {
3447                                 has_error_code = true;
3448                                 error_code =
3449                                         vmcs_read32(IDT_VECTORING_ERROR_CODE);
3450                         }
3451                         /* fall through */
3452                 case INTR_TYPE_SOFT_EXCEPTION:
3453                         kvm_clear_exception_queue(vcpu);
3454                         break;
3455                 default:
3456                         break;
3457                 }
3458         }
3459         tss_selector = exit_qualification;
3460
3461         if (!idt_v || (type != INTR_TYPE_HARD_EXCEPTION &&
3462                        type != INTR_TYPE_EXT_INTR &&
3463                        type != INTR_TYPE_NMI_INTR))
3464                 skip_emulated_instruction(vcpu);
3465
3466         if (kvm_task_switch(vcpu, tss_selector, reason,
3467                                 has_error_code, error_code) == EMULATE_FAIL) {
3468                 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
3469                 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
3470                 vcpu->run->internal.ndata = 0;
3471                 return 0;
3472         }
3473
3474         /* clear all local breakpoint enable flags */
3475         vmcs_writel(GUEST_DR7, vmcs_readl(GUEST_DR7) & ~55);
3476
3477         /*
3478          * TODO: What about debug traps on tss switch?
3479          *       Are we supposed to inject them and update dr6?
3480          */
3481
3482         return 1;
3483 }
3484
3485 static int handle_ept_violation(struct kvm_vcpu *vcpu)
3486 {
3487         unsigned long exit_qualification;
3488         gpa_t gpa;
3489         int gla_validity;
3490
3491         exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
3492
3493         if (exit_qualification & (1 << 6)) {
3494                 printk(KERN_ERR "EPT: GPA exceeds GAW!\n");
3495                 return -EINVAL;
3496         }
3497
3498         gla_validity = (exit_qualification >> 7) & 0x3;
3499         if (gla_validity != 0x3 && gla_validity != 0x1 && gla_validity != 0) {
3500                 printk(KERN_ERR "EPT: Handling EPT violation failed!\n");
3501                 printk(KERN_ERR "EPT: GPA: 0x%lx, GVA: 0x%lx\n",
3502                         (long unsigned int)vmcs_read64(GUEST_PHYSICAL_ADDRESS),
3503                         vmcs_readl(GUEST_LINEAR_ADDRESS));
3504                 printk(KERN_ERR "EPT: Exit qualification is 0x%lx\n",
3505                         (long unsigned int)exit_qualification);
3506                 vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
3507                 vcpu->run->hw.hardware_exit_reason = EXIT_REASON_EPT_VIOLATION;
3508                 return 0;
3509         }
3510
3511         gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
3512         trace_kvm_page_fault(gpa, exit_qualification);
3513         return kvm_mmu_page_fault(vcpu, gpa, exit_qualification & 0x3);
3514 }
3515
3516 static u64 ept_rsvd_mask(u64 spte, int level)
3517 {
3518         int i;
3519         u64 mask = 0;
3520
3521         for (i = 51; i > boot_cpu_data.x86_phys_bits; i--)
3522                 mask |= (1ULL << i);
3523
3524         if (level > 2)
3525                 /* bits 7:3 reserved */
3526                 mask |= 0xf8;
3527         else if (level == 2) {
3528                 if (spte & (1ULL << 7))
3529                         /* 2MB ref, bits 20:12 reserved */
3530                         mask |= 0x1ff000;
3531                 else
3532                         /* bits 6:3 reserved */
3533                         mask |= 0x78;
3534         }
3535
3536         return mask;
3537 }
3538
3539 static void ept_misconfig_inspect_spte(struct kvm_vcpu *vcpu, u64 spte,
3540                                        int level)
3541 {
3542         printk(KERN_ERR "%s: spte 0x%llx level %d\n", __func__, spte, level);
3543
3544         /* 010b (write-only) */
3545         WARN_ON((spte & 0x7) == 0x2);
3546
3547         /* 110b (write/execute) */
3548         WARN_ON((spte & 0x7) == 0x6);
3549
3550         /* 100b (execute-only) and value not supported by logical processor */
3551         if (!cpu_has_vmx_ept_execute_only())
3552                 WARN_ON((spte & 0x7) == 0x4);
3553
3554         /* not 000b */
3555         if ((spte & 0x7)) {
3556                 u64 rsvd_bits = spte & ept_rsvd_mask(spte, level);
3557
3558                 if (rsvd_bits != 0) {
3559                         printk(KERN_ERR "%s: rsvd_bits = 0x%llx\n",
3560                                          __func__, rsvd_bits);
3561                         WARN_ON(1);
3562                 }
3563
3564                 if (level == 1 || (level == 2 && (spte & (1ULL << 7)))) {
3565                         u64 ept_mem_type = (spte & 0x38) >> 3;
3566
3567                         if (ept_mem_type == 2 || ept_mem_type == 3 ||
3568                             ept_mem_type == 7) {
3569                                 printk(KERN_ERR "%s: ept_mem_type=0x%llx\n",
3570                                                 __func__, ept_mem_type);
3571                                 WARN_ON(1);
3572                         }
3573                 }
3574         }
3575 }
3576
3577 static int handle_ept_misconfig(struct kvm_vcpu *vcpu)
3578 {
3579         u64 sptes[4];
3580         int nr_sptes, i;
3581         gpa_t gpa;
3582
3583         gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
3584
3585         printk(KERN_ERR "EPT: Misconfiguration.\n");
3586         printk(KERN_ERR "EPT: GPA: 0x%llx\n", gpa);
3587
3588         nr_sptes = kvm_mmu_get_spte_hierarchy(vcpu, gpa, sptes);
3589
3590         for (i = PT64_ROOT_LEVEL; i > PT64_ROOT_LEVEL - nr_sptes; --i)
3591                 ept_misconfig_inspect_spte(vcpu, sptes[i-1], i);
3592
3593         vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
3594         vcpu->run->hw.hardware_exit_reason = EXIT_REASON_EPT_MISCONFIG;
3595
3596         return 0;
3597 }
3598
3599 static int handle_nmi_window(struct kvm_vcpu *vcpu)
3600 {
3601         u32 cpu_based_vm_exec_control;
3602
3603         /* clear pending NMI */
3604         cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
3605         cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_NMI_PENDING;
3606         vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
3607         ++vcpu->stat.nmi_window_exits;
3608         kvm_make_request(KVM_REQ_EVENT, vcpu);
3609
3610         return 1;
3611 }
3612
3613 static int handle_invalid_guest_state(struct kvm_vcpu *vcpu)
3614 {
3615         struct vcpu_vmx *vmx = to_vmx(vcpu);
3616         enum emulation_result err = EMULATE_DONE;
3617         int ret = 1;
3618         u32 cpu_exec_ctrl;
3619         bool intr_window_requested;
3620
3621         cpu_exec_ctrl = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
3622         intr_window_requested = cpu_exec_ctrl & CPU_BASED_VIRTUAL_INTR_PENDING;
3623
3624         while (!guest_state_valid(vcpu)) {
3625                 if (intr_window_requested
3626                     && (kvm_get_rflags(&vmx->vcpu) & X86_EFLAGS_IF))
3627                         return handle_interrupt_window(&vmx->vcpu);
3628
3629                 err = emulate_instruction(vcpu, 0, 0, 0);
3630
3631                 if (err == EMULATE_DO_MMIO) {
3632                         ret = 0;
3633                         goto out;
3634                 }
3635
3636                 if (err != EMULATE_DONE)
3637                         return 0;
3638
3639                 if (signal_pending(current))
3640                         goto out;
3641                 if (need_resched())
3642                         schedule();
3643         }
3644
3645         vmx->emulation_required = 0;
3646 out:
3647         return ret;
3648 }
3649
3650 /*
3651  * Indicate a busy-waiting vcpu in spinlock. We do not enable the PAUSE
3652  * exiting, so only get here on cpu with PAUSE-Loop-Exiting.
3653  */
3654 static int handle_pause(struct kvm_vcpu *vcpu)
3655 {
3656         skip_emulated_instruction(vcpu);
3657         kvm_vcpu_on_spin(vcpu);
3658
3659         return 1;
3660 }
3661
3662 static int handle_invalid_op(struct kvm_vcpu *vcpu)
3663 {
3664         kvm_queue_exception(vcpu, UD_VECTOR);
3665         return 1;
3666 }
3667
3668 /*
3669  * The exit handlers return 1 if the exit was handled fully and guest execution
3670  * may resume.  Otherwise they set the kvm_run parameter to indicate what needs
3671  * to be done to userspace and return 0.
3672  */
3673 static int (*kvm_vmx_exit_handlers[])(struct kvm_vcpu *vcpu) = {
3674         [EXIT_REASON_EXCEPTION_NMI]           = handle_exception,
3675         [EXIT_REASON_EXTERNAL_INTERRUPT]      = handle_external_interrupt,
3676         [EXIT_REASON_TRIPLE_FAULT]            = handle_triple_fault,
3677         [EXIT_REASON_NMI_WINDOW]              = handle_nmi_window,
3678         [EXIT_REASON_IO_INSTRUCTION]          = handle_io,
3679         [EXIT_REASON_CR_ACCESS]               = handle_cr,
3680         [EXIT_REASON_DR_ACCESS]               = handle_dr,
3681         [EXIT_REASON_CPUID]                   = handle_cpuid,
3682         [EXIT_REASON_MSR_READ]                = handle_rdmsr,
3683         [EXIT_REASON_MSR_WRITE]               = handle_wrmsr,
3684         [EXIT_REASON_PENDING_INTERRUPT]       = handle_interrupt_window,
3685         [EXIT_REASON_HLT]                     = handle_halt,
3686         [EXIT_REASON_INVD]                    = handle_invd,
3687         [EXIT_REASON_INVLPG]                  = handle_invlpg,
3688         [EXIT_REASON_VMCALL]                  = handle_vmcall,
3689         [EXIT_REASON_VMCLEAR]                 = handle_vmx_insn,
3690         [EXIT_REASON_VMLAUNCH]                = handle_vmx_insn,
3691         [EXIT_REASON_VMPTRLD]                 = handle_vmx_insn,
3692         [EXIT_REASON_VMPTRST]                 = handle_vmx_insn,
3693         [EXIT_REASON_VMREAD]                  = handle_vmx_insn,
3694         [EXIT_REASON_VMRESUME]                = handle_vmx_insn,
3695         [EXIT_REASON_VMWRITE]                 = handle_vmx_insn,
3696         [EXIT_REASON_VMOFF]                   = handle_vmx_insn,
3697         [EXIT_REASON_VMON]                    = handle_vmx_insn,
3698         [EXIT_REASON_TPR_BELOW_THRESHOLD]     = handle_tpr_below_threshold,
3699         [EXIT_REASON_APIC_ACCESS]             = handle_apic_access,
3700         [EXIT_REASON_WBINVD]                  = handle_wbinvd,
3701         [EXIT_REASON_XSETBV]                  = handle_xsetbv,
3702         [EXIT_REASON_TASK_SWITCH]             = handle_task_switch,
3703         [EXIT_REASON_MCE_DURING_VMENTRY]      = handle_machine_check,
3704         [EXIT_REASON_EPT_VIOLATION]           = handle_ept_violation,
3705         [EXIT_REASON_EPT_MISCONFIG]           = handle_ept_misconfig,
3706         [EXIT_REASON_PAUSE_INSTRUCTION]       = handle_pause,
3707         [EXIT_REASON_MWAIT_INSTRUCTION]       = handle_invalid_op,
3708         [EXIT_REASON_MONITOR_INSTRUCTION]     = handle_invalid_op,
3709 };
3710
3711 static const int kvm_vmx_max_exit_handlers =
3712         ARRAY_SIZE(kvm_vmx_exit_handlers);
3713
3714 static void vmx_get_exit_info(struct kvm_vcpu *vcpu, u64 *info1, u64 *info2)
3715 {
3716         *info1 = vmcs_readl(EXIT_QUALIFICATION);
3717         *info2 = vmcs_read32(VM_EXIT_INTR_INFO);
3718 }
3719
3720 /*
3721  * The guest has exited.  See if we can fix it or if we need userspace
3722  * assistance.
3723  */
3724 static int vmx_handle_exit(struct kvm_vcpu *vcpu)
3725 {
3726         struct vcpu_vmx *vmx = to_vmx(vcpu);
3727         u32 exit_reason = vmx->exit_reason;
3728         u32 vectoring_info = vmx->idt_vectoring_info;
3729
3730         trace_kvm_exit(exit_reason, vcpu, KVM_ISA_VMX);
3731
3732         /* If guest state is invalid, start emulating */
3733         if (vmx->emulation_required && emulate_invalid_guest_state)
3734                 return handle_invalid_guest_state(vcpu);
3735
3736         /* Access CR3 don't cause VMExit in paging mode, so we need
3737          * to sync with guest real CR3. */
3738         if (enable_ept && is_paging(vcpu))
3739                 vcpu->arch.cr3 = vmcs_readl(GUEST_CR3);
3740
3741         if (exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY) {
3742                 vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
3743                 vcpu->run->fail_entry.hardware_entry_failure_reason
3744                         = exit_reason;
3745                 return 0;
3746         }
3747
3748         if (unlikely(vmx->fail)) {
3749                 vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
3750                 vcpu->run->fail_entry.hardware_entry_failure_reason
3751                         = vmcs_read32(VM_INSTRUCTION_ERROR);
3752                 return 0;
3753         }
3754
3755         if ((vectoring_info & VECTORING_INFO_VALID_MASK) &&
3756                         (exit_reason != EXIT_REASON_EXCEPTION_NMI &&
3757                         exit_reason != EXIT_REASON_EPT_VIOLATION &&
3758                         exit_reason != EXIT_REASON_TASK_SWITCH))
3759                 printk(KERN_WARNING "%s: unexpected, valid vectoring info "
3760                        "(0x%x) and exit reason is 0x%x\n",
3761                        __func__, vectoring_info, exit_reason);
3762
3763         if (unlikely(!cpu_has_virtual_nmis() && vmx->soft_vnmi_blocked)) {
3764                 if (vmx_interrupt_allowed(vcpu)) {
3765                         vmx->soft_vnmi_blocked = 0;
3766                 } else if (vmx->vnmi_blocked_time > 1000000000LL &&
3767                            vcpu->arch.nmi_pending) {
3768                         /*
3769                          * This CPU don't support us in finding the end of an
3770                          * NMI-blocked window if the guest runs with IRQs
3771                          * disabled. So we pull the trigger after 1 s of
3772                          * futile waiting, but inform the user about this.
3773                          */
3774                         printk(KERN_WARNING "%s: Breaking out of NMI-blocked "
3775                                "state on VCPU %d after 1 s timeout\n",
3776                                __func__, vcpu->vcpu_id);
3777                         vmx->soft_vnmi_blocked = 0;
3778                 }
3779         }
3780
3781         if (exit_reason < kvm_vmx_max_exit_handlers
3782             && kvm_vmx_exit_handlers[exit_reason])
3783                 return kvm_vmx_exit_handlers[exit_reason](vcpu);
3784         else {
3785                 vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
3786                 vcpu->run->hw.hardware_exit_reason = exit_reason;
3787         }
3788         return 0;
3789 }
3790
3791 static void update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
3792 {
3793         if (irr == -1 || tpr < irr) {
3794                 vmcs_write32(TPR_THRESHOLD, 0);
3795                 return;
3796         }
3797
3798         vmcs_write32(TPR_THRESHOLD, irr);
3799 }
3800
3801 static void vmx_complete_atomic_exit(struct vcpu_vmx *vmx)
3802 {
3803         u32 exit_intr_info = vmx->exit_intr_info;
3804
3805         /* Handle machine checks before interrupts are enabled */
3806         if ((vmx->exit_reason == EXIT_REASON_MCE_DURING_VMENTRY)
3807             || (vmx->exit_reason == EXIT_REASON_EXCEPTION_NMI
3808                 && is_machine_check(exit_intr_info)))
3809                 kvm_machine_check();
3810
3811         /* We need to handle NMIs before interrupts are enabled */
3812         if ((exit_intr_info & INTR_INFO_INTR_TYPE_MASK) == INTR_TYPE_NMI_INTR &&
3813             (exit_intr_info & INTR_INFO_VALID_MASK)) {
3814                 kvm_before_handle_nmi(&vmx->vcpu);
3815                 asm("int $2");
3816                 kvm_after_handle_nmi(&vmx->vcpu);
3817         }
3818 }
3819
3820 static void vmx_recover_nmi_blocking(struct vcpu_vmx *vmx)
3821 {
3822         u32 exit_intr_info = vmx->exit_intr_info;
3823         bool unblock_nmi;
3824         u8 vector;
3825         bool idtv_info_valid;
3826
3827         idtv_info_valid = vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK;
3828
3829         if (cpu_has_virtual_nmis()) {
3830                 unblock_nmi = (exit_intr_info & INTR_INFO_UNBLOCK_NMI) != 0;
3831                 vector = exit_intr_info & INTR_INFO_VECTOR_MASK;
3832                 /*
3833                  * SDM 3: 27.7.1.2 (September 2008)
3834                  * Re-set bit "block by NMI" before VM entry if vmexit caused by
3835                  * a guest IRET fault.
3836                  * SDM 3: 23.2.2 (September 2008)
3837                  * Bit 12 is undefined in any of the following cases:
3838                  *  If the VM exit sets the valid bit in the IDT-vectoring
3839                  *   information field.
3840                  *  If the VM exit is due to a double fault.
3841                  */
3842                 if ((exit_intr_info & INTR_INFO_VALID_MASK) && unblock_nmi &&
3843                     vector != DF_VECTOR && !idtv_info_valid)
3844                         vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
3845                                       GUEST_INTR_STATE_NMI);
3846         } else if (unlikely(vmx->soft_vnmi_blocked))
3847                 vmx->vnmi_blocked_time +=
3848                         ktime_to_ns(ktime_sub(ktime_get(), vmx->entry_time));
3849 }
3850
3851 static void __vmx_complete_interrupts(struct vcpu_vmx *vmx,
3852                                       u32 idt_vectoring_info,
3853                                       int instr_len_field,
3854                                       int error_code_field)
3855 {
3856         u8 vector;
3857         int type;
3858         bool idtv_info_valid;
3859
3860         idtv_info_valid = idt_vectoring_info & VECTORING_INFO_VALID_MASK;
3861
3862         vmx->vcpu.arch.nmi_injected = false;
3863         kvm_clear_exception_queue(&vmx->vcpu);
3864         kvm_clear_interrupt_queue(&vmx->vcpu);
3865
3866         if (!idtv_info_valid)
3867                 return;
3868
3869         kvm_make_request(KVM_REQ_EVENT, &vmx->vcpu);
3870
3871         vector = idt_vectoring_info & VECTORING_INFO_VECTOR_MASK;
3872         type = idt_vectoring_info & VECTORING_INFO_TYPE_MASK;
3873
3874         switch (type) {
3875         case INTR_TYPE_NMI_INTR:
3876                 vmx->vcpu.arch.nmi_injected = true;
3877                 /*
3878                  * SDM 3: 27.7.1.2 (September 2008)
3879                  * Clear bit "block by NMI" before VM entry if a NMI
3880                  * delivery faulted.
3881                  */
3882                 vmcs_clear_bits(GUEST_INTERRUPTIBILITY_INFO,
3883                                 GUEST_INTR_STATE_NMI);
3884                 break;
3885         case INTR_TYPE_SOFT_EXCEPTION:
3886                 vmx->vcpu.arch.event_exit_inst_len =
3887                         vmcs_read32(instr_len_field);
3888                 /* fall through */
3889         case INTR_TYPE_HARD_EXCEPTION:
3890                 if (idt_vectoring_info & VECTORING_INFO_DELIVER_CODE_MASK) {
3891                         u32 err = vmcs_read32(error_code_field);
3892                         kvm_queue_exception_e(&vmx->vcpu, vector, err);
3893                 } else
3894                         kvm_queue_exception(&vmx->vcpu, vector);
3895                 break;
3896         case INTR_TYPE_SOFT_INTR:
3897                 vmx->vcpu.arch.event_exit_inst_len =
3898                         vmcs_read32(instr_len_field);
3899                 /* fall through */
3900         case INTR_TYPE_EXT_INTR:
3901                 kvm_queue_interrupt(&vmx->vcpu, vector,
3902                         type == INTR_TYPE_SOFT_INTR);
3903                 break;
3904         default:
3905                 break;
3906         }
3907 }
3908
3909 static void vmx_complete_interrupts(struct vcpu_vmx *vmx)
3910 {
3911         __vmx_complete_interrupts(vmx, vmx->idt_vectoring_info,
3912                                   VM_EXIT_INSTRUCTION_LEN,
3913                                   IDT_VECTORING_ERROR_CODE);
3914 }
3915
3916 static void vmx_cancel_injection(struct kvm_vcpu *vcpu)
3917 {
3918         __vmx_complete_interrupts(to_vmx(vcpu),
3919                                   vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
3920                                   VM_ENTRY_INSTRUCTION_LEN,
3921                                   VM_ENTRY_EXCEPTION_ERROR_CODE);
3922
3923         vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);
3924 }
3925
3926 #ifdef CONFIG_X86_64
3927 #define R "r"
3928 #define Q "q"
3929 #else
3930 #define R "e"
3931 #define Q "l"
3932 #endif
3933
3934 static void vmx_vcpu_run(struct kvm_vcpu *vcpu)
3935 {
3936         struct vcpu_vmx *vmx = to_vmx(vcpu);
3937
3938         /* Record the guest's net vcpu time for enforced NMI injections. */
3939         if (unlikely(!cpu_has_virtual_nmis() && vmx->soft_vnmi_blocked))
3940                 vmx->entry_time = ktime_get();
3941
3942         /* Don't enter VMX if guest state is invalid, let the exit handler
3943            start emulation until we arrive back to a valid state */
3944         if (vmx->emulation_required && emulate_invalid_guest_state)
3945                 return;
3946
3947         if (test_bit(VCPU_REGS_RSP, (unsigned long *)&vcpu->arch.regs_dirty))
3948                 vmcs_writel(GUEST_RSP, vcpu->arch.regs[VCPU_REGS_RSP]);
3949         if (test_bit(VCPU_REGS_RIP, (unsigned long *)&vcpu->arch.regs_dirty))
3950                 vmcs_writel(GUEST_RIP, vcpu->arch.regs[VCPU_REGS_RIP]);
3951
3952         /* When single-stepping over STI and MOV SS, we must clear the
3953          * corresponding interruptibility bits in the guest state. Otherwise
3954          * vmentry fails as it then expects bit 14 (BS) in pending debug
3955          * exceptions being set, but that's not correct for the guest debugging
3956          * case. */
3957         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
3958                 vmx_set_interrupt_shadow(vcpu, 0);
3959
3960         asm(
3961                 /* Store host registers */
3962                 "push %%"R"dx; push %%"R"bp;"
3963                 "push %%"R"cx \n\t"
3964                 "cmp %%"R"sp, %c[host_rsp](%0) \n\t"
3965                 "je 1f \n\t"
3966                 "mov %%"R"sp, %c[host_rsp](%0) \n\t"
3967                 __ex(ASM_VMX_VMWRITE_RSP_RDX) "\n\t"
3968                 "1: \n\t"
3969                 /* Reload cr2 if changed */
3970                 "mov %c[cr2](%0), %%"R"ax \n\t"
3971                 "mov %%cr2, %%"R"dx \n\t"
3972                 "cmp %%"R"ax, %%"R"dx \n\t"
3973                 "je 2f \n\t"
3974                 "mov %%"R"ax, %%cr2 \n\t"
3975                 "2: \n\t"
3976                 /* Check if vmlaunch of vmresume is needed */
3977                 "cmpl $0, %c[launched](%0) \n\t"
3978                 /* Load guest registers.  Don't clobber flags. */
3979                 "mov %c[rax](%0), %%"R"ax \n\t"
3980                 "mov %c[rbx](%0), %%"R"bx \n\t"
3981                 "mov %c[rdx](%0), %%"R"dx \n\t"
3982                 "mov %c[rsi](%0), %%"R"si \n\t"
3983                 "mov %c[rdi](%0), %%"R"di \n\t"
3984                 "mov %c[rbp](%0), %%"R"bp \n\t"
3985 #ifdef CONFIG_X86_64
3986                 "mov %c[r8](%0),  %%r8  \n\t"
3987                 "mov %c[r9](%0),  %%r9  \n\t"
3988                 "mov %c[r10](%0), %%r10 \n\t"
3989                 "mov %c[r11](%0), %%r11 \n\t"
3990                 "mov %c[r12](%0), %%r12 \n\t"
3991                 "mov %c[r13](%0), %%r13 \n\t"
3992                 "mov %c[r14](%0), %%r14 \n\t"
3993                 "mov %c[r15](%0), %%r15 \n\t"
3994 #endif
3995                 "mov %c[rcx](%0), %%"R"cx \n\t" /* kills %0 (ecx) */
3996
3997                 /* Enter guest mode */
3998                 "jne .Llaunched \n\t"
3999                 __ex(ASM_VMX_VMLAUNCH) "\n\t"
4000                 "jmp .Lkvm_vmx_return \n\t"
4001                 ".Llaunched: " __ex(ASM_VMX_VMRESUME) "\n\t"
4002                 ".Lkvm_vmx_return: "
4003                 /* Save guest registers, load host registers, keep flags */
4004                 "xchg %0,     (%%"R"sp) \n\t"
4005                 "mov %%"R"ax, %c[rax](%0) \n\t"
4006                 "mov %%"R"bx, %c[rbx](%0) \n\t"
4007                 "push"Q" (%%"R"sp); pop"Q" %c[rcx](%0) \n\t"
4008                 "mov %%"R"dx, %c[rdx](%0) \n\t"
4009                 "mov %%"R"si, %c[rsi](%0) \n\t"
4010                 "mov %%"R"di, %c[rdi](%0) \n\t"
4011                 "mov %%"R"bp, %c[rbp](%0) \n\t"
4012 #ifdef CONFIG_X86_64
4013                 "mov %%r8,  %c[r8](%0) \n\t"
4014                 "mov %%r9,  %c[r9](%0) \n\t"
4015                 "mov %%r10, %c[r10](%0) \n\t"
4016                 "mov %%r11, %c[r11](%0) \n\t"
4017                 "mov %%r12, %c[r12](%0) \n\t"
4018                 "mov %%r13, %c[r13](%0) \n\t"
4019                 "mov %%r14, %c[r14](%0) \n\t"
4020                 "mov %%r15, %c[r15](%0) \n\t"
4021 #endif
4022                 "mov %%cr2, %%"R"ax   \n\t"
4023                 "mov %%"R"ax, %c[cr2](%0) \n\t"
4024
4025                 "pop  %%"R"bp; pop  %%"R"bp; pop  %%"R"dx \n\t"
4026                 "setbe %c[fail](%0) \n\t"
4027               : : "c"(vmx), "d"((unsigned long)HOST_RSP),
4028                 [launched]"i"(offsetof(struct vcpu_vmx, launched)),
4029                 [fail]"i"(offsetof(struct vcpu_vmx, fail)),
4030                 [host_rsp]"i"(offsetof(struct vcpu_vmx, host_rsp)),
4031                 [rax]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RAX])),
4032                 [rbx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBX])),
4033                 [rcx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RCX])),
4034                 [rdx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDX])),
4035                 [rsi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RSI])),
4036                 [rdi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDI])),
4037                 [rbp]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBP])),
4038 #ifdef CONFIG_X86_64
4039                 [r8]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R8])),
4040                 [r9]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R9])),
4041                 [r10]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R10])),
4042                 [r11]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R11])),
4043                 [r12]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R12])),
4044                 [r13]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R13])),
4045                 [r14]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R14])),
4046                 [r15]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R15])),
4047 #endif
4048                 [cr2]"i"(offsetof(struct vcpu_vmx, vcpu.arch.cr2))
4049               : "cc", "memory"
4050                 , R"ax", R"bx", R"di", R"si"
4051 #ifdef CONFIG_X86_64
4052                 , "r8", "r9", "r10", "r11", "r12", "r13", "r14", "r15"
4053 #endif
4054               );
4055
4056         vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP)
4057                                   | (1 << VCPU_EXREG_PDPTR));
4058         vcpu->arch.regs_dirty = 0;
4059
4060         vmx->idt_vectoring_info = vmcs_read32(IDT_VECTORING_INFO_FIELD);
4061
4062         asm("mov %0, %%ds; mov %0, %%es" : : "r"(__USER_DS));
4063         vmx->launched = 1;
4064
4065         vmx->exit_reason = vmcs_read32(VM_EXIT_REASON);
4066         vmx->exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
4067
4068         vmx_complete_atomic_exit(vmx);
4069         vmx_recover_nmi_blocking(vmx);
4070         vmx_complete_interrupts(vmx);
4071 }
4072
4073 #undef R
4074 #undef Q
4075
4076 static void vmx_free_vmcs(struct kvm_vcpu *vcpu)
4077 {
4078         struct vcpu_vmx *vmx = to_vmx(vcpu);
4079
4080         if (vmx->vmcs) {
4081                 vcpu_clear(vmx);
4082                 free_vmcs(vmx->vmcs);
4083                 vmx->vmcs = NULL;
4084         }
4085 }
4086
4087 static void vmx_free_vcpu(struct kvm_vcpu *vcpu)
4088 {
4089         struct vcpu_vmx *vmx = to_vmx(vcpu);
4090
4091         free_vpid(vmx);
4092         vmx_free_vmcs(vcpu);
4093         kfree(vmx->guest_msrs);
4094         kvm_vcpu_uninit(vcpu);
4095         kmem_cache_free(kvm_vcpu_cache, vmx);
4096 }
4097
4098 static inline void vmcs_init(struct vmcs *vmcs)
4099 {
4100         u64 phys_addr = __pa(per_cpu(vmxarea, raw_smp_processor_id()));
4101
4102         if (!vmm_exclusive)
4103                 kvm_cpu_vmxon(phys_addr);
4104
4105         vmcs_clear(vmcs);
4106
4107         if (!vmm_exclusive)
4108                 kvm_cpu_vmxoff();
4109 }
4110
4111 static struct kvm_vcpu *vmx_create_vcpu(struct kvm *kvm, unsigned int id)
4112 {
4113         int err;
4114         struct vcpu_vmx *vmx = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
4115         int cpu;
4116
4117         if (!vmx)
4118                 return ERR_PTR(-ENOMEM);
4119
4120         allocate_vpid(vmx);
4121
4122         err = kvm_vcpu_init(&vmx->vcpu, kvm, id);
4123         if (err)
4124                 goto free_vcpu;
4125
4126         vmx->guest_msrs = kmalloc(PAGE_SIZE, GFP_KERNEL);
4127         if (!vmx->guest_msrs) {
4128                 err = -ENOMEM;
4129                 goto uninit_vcpu;
4130         }
4131
4132         vmx->vmcs = alloc_vmcs();
4133         if (!vmx->vmcs)
4134                 goto free_msrs;
4135
4136         vmcs_init(vmx->vmcs);
4137
4138         cpu = get_cpu();
4139         vmx_vcpu_load(&vmx->vcpu, cpu);
4140         vmx->vcpu.cpu = cpu;
4141         err = vmx_vcpu_setup(vmx);
4142         vmx_vcpu_put(&vmx->vcpu);
4143         put_cpu();
4144         if (err)
4145                 goto free_vmcs;
4146         if (vm_need_virtualize_apic_accesses(kvm))
4147                 if (alloc_apic_access_page(kvm) != 0)
4148                         goto free_vmcs;
4149
4150         if (enable_ept) {
4151                 if (!kvm->arch.ept_identity_map_addr)
4152                         kvm->arch.ept_identity_map_addr =
4153                                 VMX_EPT_IDENTITY_PAGETABLE_ADDR;
4154                 if (alloc_identity_pagetable(kvm) != 0)
4155                         goto free_vmcs;
4156         }
4157
4158         return &vmx->vcpu;
4159
4160 free_vmcs:
4161         free_vmcs(vmx->vmcs);
4162 free_msrs:
4163         kfree(vmx->guest_msrs);
4164 uninit_vcpu:
4165         kvm_vcpu_uninit(&vmx->vcpu);
4166 free_vcpu:
4167         free_vpid(vmx);
4168         kmem_cache_free(kvm_vcpu_cache, vmx);
4169         return ERR_PTR(err);
4170 }
4171
4172 static void __init vmx_check_processor_compat(void *rtn)
4173 {
4174         struct vmcs_config vmcs_conf;
4175
4176         *(int *)rtn = 0;
4177         if (setup_vmcs_config(&vmcs_conf) < 0)
4178                 *(int *)rtn = -EIO;
4179         if (memcmp(&vmcs_config, &vmcs_conf, sizeof(struct vmcs_config)) != 0) {
4180                 printk(KERN_ERR "kvm: CPU %d feature inconsistency!\n",
4181                                 smp_processor_id());
4182                 *(int *)rtn = -EIO;
4183         }
4184 }
4185
4186 static int get_ept_level(void)
4187 {
4188         return VMX_EPT_DEFAULT_GAW + 1;
4189 }
4190
4191 static u64 vmx_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio)
4192 {
4193         u64 ret;
4194
4195         /* For VT-d and EPT combination
4196          * 1. MMIO: always map as UC
4197          * 2. EPT with VT-d:
4198          *   a. VT-d without snooping control feature: can't guarantee the
4199          *      result, try to trust guest.
4200          *   b. VT-d with snooping control feature: snooping control feature of
4201          *      VT-d engine can guarantee the cache correctness. Just set it
4202          *      to WB to keep consistent with host. So the same as item 3.
4203          * 3. EPT without VT-d: always map as WB and set IPAT=1 to keep
4204          *    consistent with host MTRR
4205          */
4206         if (is_mmio)
4207                 ret = MTRR_TYPE_UNCACHABLE << VMX_EPT_MT_EPTE_SHIFT;
4208         else if (vcpu->kvm->arch.iommu_domain &&
4209                 !(vcpu->kvm->arch.iommu_flags & KVM_IOMMU_CACHE_COHERENCY))
4210                 ret = kvm_get_guest_memory_type(vcpu, gfn) <<
4211                       VMX_EPT_MT_EPTE_SHIFT;
4212         else
4213                 ret = (MTRR_TYPE_WRBACK << VMX_EPT_MT_EPTE_SHIFT)
4214                         | VMX_EPT_IPAT_BIT;
4215
4216         return ret;
4217 }
4218
4219 #define _ER(x) { EXIT_REASON_##x, #x }
4220
4221 static const struct trace_print_flags vmx_exit_reasons_str[] = {
4222         _ER(EXCEPTION_NMI),
4223         _ER(EXTERNAL_INTERRUPT),
4224         _ER(TRIPLE_FAULT),
4225         _ER(PENDING_INTERRUPT),
4226         _ER(NMI_WINDOW),
4227         _ER(TASK_SWITCH),
4228         _ER(CPUID),
4229         _ER(HLT),
4230         _ER(INVLPG),
4231         _ER(RDPMC),
4232         _ER(RDTSC),
4233         _ER(VMCALL),
4234         _ER(VMCLEAR),
4235         _ER(VMLAUNCH),
4236         _ER(VMPTRLD),
4237         _ER(VMPTRST),
4238         _ER(VMREAD),
4239         _ER(VMRESUME),
4240         _ER(VMWRITE),
4241         _ER(VMOFF),
4242         _ER(VMON),
4243         _ER(CR_ACCESS),
4244         _ER(DR_ACCESS),
4245         _ER(IO_INSTRUCTION),
4246         _ER(MSR_READ),
4247         _ER(MSR_WRITE),
4248         _ER(MWAIT_INSTRUCTION),
4249         _ER(MONITOR_INSTRUCTION),
4250         _ER(PAUSE_INSTRUCTION),
4251         _ER(MCE_DURING_VMENTRY),
4252         _ER(TPR_BELOW_THRESHOLD),
4253         _ER(APIC_ACCESS),
4254         _ER(EPT_VIOLATION),
4255         _ER(EPT_MISCONFIG),
4256         _ER(WBINVD),
4257         { -1, NULL }
4258 };
4259
4260 #undef _ER
4261
4262 static int vmx_get_lpage_level(void)
4263 {
4264         if (enable_ept && !cpu_has_vmx_ept_1g_page())
4265                 return PT_DIRECTORY_LEVEL;
4266         else
4267                 /* For shadow and EPT supported 1GB page */
4268                 return PT_PDPE_LEVEL;
4269 }
4270
4271 static void vmx_cpuid_update(struct kvm_vcpu *vcpu)
4272 {
4273         struct kvm_cpuid_entry2 *best;
4274         struct vcpu_vmx *vmx = to_vmx(vcpu);
4275         u32 exec_control;
4276
4277         vmx->rdtscp_enabled = false;
4278         if (vmx_rdtscp_supported()) {
4279                 exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
4280                 if (exec_control & SECONDARY_EXEC_RDTSCP) {
4281                         best = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
4282                         if (best && (best->edx & bit(X86_FEATURE_RDTSCP)))
4283                                 vmx->rdtscp_enabled = true;
4284                         else {
4285                                 exec_control &= ~SECONDARY_EXEC_RDTSCP;
4286                                 vmcs_write32(SECONDARY_VM_EXEC_CONTROL,
4287                                                 exec_control);
4288                         }
4289                 }
4290         }
4291 }
4292
4293 static void vmx_set_supported_cpuid(u32 func, struct kvm_cpuid_entry2 *entry)
4294 {
4295 }
4296
4297 static struct kvm_x86_ops vmx_x86_ops = {
4298         .cpu_has_kvm_support = cpu_has_kvm_support,
4299         .disabled_by_bios = vmx_disabled_by_bios,
4300         .hardware_setup = hardware_setup,
4301         .hardware_unsetup = hardware_unsetup,
4302         .check_processor_compatibility = vmx_check_processor_compat,
4303         .hardware_enable = hardware_enable,
4304         .hardware_disable = hardware_disable,
4305         .cpu_has_accelerated_tpr = report_flexpriority,
4306
4307         .vcpu_create = vmx_create_vcpu,
4308         .vcpu_free = vmx_free_vcpu,
4309         .vcpu_reset = vmx_vcpu_reset,
4310
4311         .prepare_guest_switch = vmx_save_host_state,
4312         .vcpu_load = vmx_vcpu_load,
4313         .vcpu_put = vmx_vcpu_put,
4314
4315         .set_guest_debug = set_guest_debug,
4316         .get_msr = vmx_get_msr,
4317         .set_msr = vmx_set_msr,
4318         .get_segment_base = vmx_get_segment_base,
4319         .get_segment = vmx_get_segment,
4320         .set_segment = vmx_set_segment,
4321         .get_cpl = vmx_get_cpl,
4322         .get_cs_db_l_bits = vmx_get_cs_db_l_bits,
4323         .decache_cr0_guest_bits = vmx_decache_cr0_guest_bits,
4324         .decache_cr4_guest_bits = vmx_decache_cr4_guest_bits,
4325         .set_cr0 = vmx_set_cr0,
4326         .set_cr3 = vmx_set_cr3,
4327         .set_cr4 = vmx_set_cr4,
4328         .set_efer = vmx_set_efer,
4329         .get_idt = vmx_get_idt,
4330         .set_idt = vmx_set_idt,
4331         .get_gdt = vmx_get_gdt,
4332         .set_gdt = vmx_set_gdt,
4333         .set_dr7 = vmx_set_dr7,
4334         .cache_reg = vmx_cache_reg,
4335         .get_rflags = vmx_get_rflags,
4336         .set_rflags = vmx_set_rflags,
4337         .fpu_activate = vmx_fpu_activate,
4338         .fpu_deactivate = vmx_fpu_deactivate,
4339
4340         .tlb_flush = vmx_flush_tlb,
4341
4342         .run = vmx_vcpu_run,
4343         .handle_exit = vmx_handle_exit,
4344         .skip_emulated_instruction = skip_emulated_instruction,
4345         .set_interrupt_shadow = vmx_set_interrupt_shadow,
4346         .get_interrupt_shadow = vmx_get_interrupt_shadow,
4347         .patch_hypercall = vmx_patch_hypercall,
4348         .set_irq = vmx_inject_irq,
4349         .set_nmi = vmx_inject_nmi,
4350         .queue_exception = vmx_queue_exception,
4351         .cancel_injection = vmx_cancel_injection,
4352         .interrupt_allowed = vmx_interrupt_allowed,
4353         .nmi_allowed = vmx_nmi_allowed,
4354         .get_nmi_mask = vmx_get_nmi_mask,
4355         .set_nmi_mask = vmx_set_nmi_mask,
4356         .enable_nmi_window = enable_nmi_window,
4357         .enable_irq_window = enable_irq_window,
4358         .update_cr8_intercept = update_cr8_intercept,
4359
4360         .set_tss_addr = vmx_set_tss_addr,
4361         .get_tdp_level = get_ept_level,
4362         .get_mt_mask = vmx_get_mt_mask,
4363
4364         .get_exit_info = vmx_get_exit_info,
4365         .exit_reasons_str = vmx_exit_reasons_str,
4366
4367         .get_lpage_level = vmx_get_lpage_level,
4368
4369         .cpuid_update = vmx_cpuid_update,
4370
4371         .rdtscp_supported = vmx_rdtscp_supported,
4372
4373         .set_supported_cpuid = vmx_set_supported_cpuid,
4374
4375         .has_wbinvd_exit = cpu_has_vmx_wbinvd_exit,
4376
4377         .write_tsc_offset = vmx_write_tsc_offset,
4378         .adjust_tsc_offset = vmx_adjust_tsc_offset,
4379
4380         .set_tdp_cr3 = vmx_set_cr3,
4381 };
4382
4383 static int __init vmx_init(void)
4384 {
4385         int r, i;
4386
4387         rdmsrl_safe(MSR_EFER, &host_efer);
4388
4389         for (i = 0; i < NR_VMX_MSR; ++i)
4390                 kvm_define_shared_msr(i, vmx_msr_index[i]);
4391
4392         vmx_io_bitmap_a = (unsigned long *)__get_free_page(GFP_KERNEL);
4393         if (!vmx_io_bitmap_a)
4394                 return -ENOMEM;
4395
4396         vmx_io_bitmap_b = (unsigned long *)__get_free_page(GFP_KERNEL);
4397         if (!vmx_io_bitmap_b) {
4398                 r = -ENOMEM;
4399                 goto out;
4400         }
4401
4402         vmx_msr_bitmap_legacy = (unsigned long *)__get_free_page(GFP_KERNEL);
4403         if (!vmx_msr_bitmap_legacy) {
4404                 r = -ENOMEM;
4405                 goto out1;
4406         }
4407
4408         vmx_msr_bitmap_longmode = (unsigned long *)__get_free_page(GFP_KERNEL);
4409         if (!vmx_msr_bitmap_longmode) {
4410                 r = -ENOMEM;
4411                 goto out2;
4412         }
4413
4414         /*
4415          * Allow direct access to the PC debug port (it is often used for I/O
4416          * delays, but the vmexits simply slow things down).
4417          */
4418         memset(vmx_io_bitmap_a, 0xff, PAGE_SIZE);
4419         clear_bit(0x80, vmx_io_bitmap_a);
4420
4421         memset(vmx_io_bitmap_b, 0xff, PAGE_SIZE);
4422
4423         memset(vmx_msr_bitmap_legacy, 0xff, PAGE_SIZE);
4424         memset(vmx_msr_bitmap_longmode, 0xff, PAGE_SIZE);
4425
4426         set_bit(0, vmx_vpid_bitmap); /* 0 is reserved for host */
4427
4428         r = kvm_init(&vmx_x86_ops, sizeof(struct vcpu_vmx),
4429                      __alignof__(struct vcpu_vmx), THIS_MODULE);
4430         if (r)
4431                 goto out3;
4432
4433         vmx_disable_intercept_for_msr(MSR_FS_BASE, false);
4434         vmx_disable_intercept_for_msr(MSR_GS_BASE, false);
4435         vmx_disable_intercept_for_msr(MSR_KERNEL_GS_BASE, true);
4436         vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_CS, false);
4437         vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_ESP, false);
4438         vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_EIP, false);
4439
4440         if (enable_ept) {
4441                 bypass_guest_pf = 0;
4442                 kvm_mmu_set_mask_ptes(0ull, 0ull, 0ull, 0ull,
4443                                 VMX_EPT_EXECUTABLE_MASK);
4444                 kvm_enable_tdp();
4445         } else
4446                 kvm_disable_tdp();
4447
4448         if (bypass_guest_pf)
4449                 kvm_mmu_set_nonpresent_ptes(~0xffeull, 0ull);
4450
4451         return 0;
4452
4453 out3:
4454         free_page((unsigned long)vmx_msr_bitmap_longmode);
4455 out2:
4456         free_page((unsigned long)vmx_msr_bitmap_legacy);
4457 out1:
4458         free_page((unsigned long)vmx_io_bitmap_b);
4459 out:
4460         free_page((unsigned long)vmx_io_bitmap_a);
4461         return r;
4462 }
4463
4464 static void __exit vmx_exit(void)
4465 {
4466         free_page((unsigned long)vmx_msr_bitmap_legacy);
4467         free_page((unsigned long)vmx_msr_bitmap_longmode);
4468         free_page((unsigned long)vmx_io_bitmap_b);
4469         free_page((unsigned long)vmx_io_bitmap_a);
4470
4471         kvm_exit();
4472 }
4473
4474 module_init(vmx_init)
4475 module_exit(vmx_exit)