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 smaller than 128 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    128
 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         u8                    cpl;
 132         bool                  nmi_known_unmasked;
 133         u32                   exit_intr_info;
 134         u32                   idt_vectoring_info;
 135         ulong                 rflags;
 136         struct shared_msr_entry *guest_msrs;
 137         int                   nmsrs;
 138         int                   save_nmsrs;
 139 #ifdef CONFIG_X86_64
 140         u64                   msr_host_kernel_gs_base;
 141         u64                   msr_guest_kernel_gs_base;
 142 #endif
 143         struct vmcs          *vmcs;
 144         struct msr_autoload {
 145                 unsigned nr;
 146                 struct vmx_msr_entry guest[NR_AUTOLOAD_MSRS];
 147                 struct vmx_msr_entry host[NR_AUTOLOAD_MSRS];
 148         } msr_autoload;
 149         struct {
 150                 int           loaded;
 151                 u16           fs_sel, gs_sel, ldt_sel;
 152                 int           gs_ldt_reload_needed;
 153                 int           fs_reload_needed;
 154         } host_state;
 155         struct {
 156                 int vm86_active;
 157                 ulong save_rflags;
 158                 struct kvm_save_segment {
 159                         u16 selector;
 160                         unsigned long base;
 161                         u32 limit;
 162                         u32 ar;
 163                 } tr, es, ds, fs, gs;
 164         } rmode;
 165         int vpid;
 166         bool emulation_required;
 167
 168         /* Support for vnmi-less CPUs */
 169         int soft_vnmi_blocked;
 170         ktime_t entry_time;
 171         s64 vnmi_blocked_time;
 172         u32 exit_reason;
 173
 174         bool rdtscp_enabled;
 175 };
 176
 177 static inline struct vcpu_vmx *to_vmx(struct kvm_vcpu *vcpu)
 178 {
 179         return container_of(vcpu, struct vcpu_vmx, vcpu);
 180 }
 181
 182 static u64 construct_eptp(unsigned long root_hpa);
 183 static void kvm_cpu_vmxon(u64 addr);
 184 static void kvm_cpu_vmxoff(void);
 185 static void vmx_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3);
 186 static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr);
 187
 188 static DEFINE_PER_CPU(struct vmcs *, vmxarea);
 189 static DEFINE_PER_CPU(struct vmcs *, current_vmcs);
 190 static DEFINE_PER_CPU(struct list_head, vcpus_on_cpu);
 191 static DEFINE_PER_CPU(struct desc_ptr, host_gdt);
 192
 193 static unsigned long *vmx_io_bitmap_a;
 194 static unsigned long *vmx_io_bitmap_b;
 195 static unsigned long *vmx_msr_bitmap_legacy;
 196 static unsigned long *vmx_msr_bitmap_longmode;
 197
 198 static bool cpu_has_load_ia32_efer;
 199
 200 static DECLARE_BITMAP(vmx_vpid_bitmap, VMX_NR_VPIDS);
 201 static DEFINE_SPINLOCK(vmx_vpid_lock);
 202
 203 static struct vmcs_config {
 204         int size;
 205         int order;
 206         u32 revision_id;
 207         u32 pin_based_exec_ctrl;
 208         u32 cpu_based_exec_ctrl;
 209         u32 cpu_based_2nd_exec_ctrl;
 210         u32 vmexit_ctrl;
 211         u32 vmentry_ctrl;
 212 } vmcs_config;
 213
 214 static struct vmx_capability {
 215         u32 ept;
 216         u32 vpid;
 217 } vmx_capability;
 218
 219 #define VMX_SEGMENT_FIELD(seg)                                  \
 220         [VCPU_SREG_##seg] = {                                   \
 221                 .selector = GUEST_##seg##_SELECTOR,             \
 222                 .base = GUEST_##seg##_BASE,                     \
 223                 .limit = GUEST_##seg##_LIMIT,                   \
 224                 .ar_bytes = GUEST_##seg##_AR_BYTES,             \
 225         }
 226
 227 static struct kvm_vmx_segment_field {
 228         unsigned selector;
 229         unsigned base;
 230         unsigned limit;
 231         unsigned ar_bytes;
 232 } kvm_vmx_segment_fields[] = {
 233         VMX_SEGMENT_FIELD(CS),
 234         VMX_SEGMENT_FIELD(DS),
 235         VMX_SEGMENT_FIELD(ES),
 236         VMX_SEGMENT_FIELD(FS),
 237         VMX_SEGMENT_FIELD(GS),
 238         VMX_SEGMENT_FIELD(SS),
 239         VMX_SEGMENT_FIELD(TR),
 240         VMX_SEGMENT_FIELD(LDTR),
 241 };
 242
 243 static u64 host_efer;
 244
 245 static void ept_save_pdptrs(struct kvm_vcpu *vcpu);
 246
 247 /*
 248  * Keep MSR_STAR at the end, as setup_msrs() will try to optimize it
 249  * away by decrementing the array size.
 250  */
 251 static const u32 vmx_msr_index[] = {
 252 #ifdef CONFIG_X86_64
 253         MSR_SYSCALL_MASK, MSR_LSTAR, MSR_CSTAR,
 254 #endif
 255         MSR_EFER, MSR_TSC_AUX, MSR_STAR,
 256 };
 257 #define NR_VMX_MSR ARRAY_SIZE(vmx_msr_index)
 258
 259 static inline bool is_page_fault(u32 intr_info)
 260 {
 261         return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
 262                              INTR_INFO_VALID_MASK)) ==
 263                 (INTR_TYPE_HARD_EXCEPTION | PF_VECTOR | INTR_INFO_VALID_MASK);
 264 }
 265
 266 static inline bool is_no_device(u32 intr_info)
 267 {
 268         return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
 269                              INTR_INFO_VALID_MASK)) ==
 270                 (INTR_TYPE_HARD_EXCEPTION | NM_VECTOR | INTR_INFO_VALID_MASK);
 271 }
 272
 273 static inline bool is_invalid_opcode(u32 intr_info)
 274 {
 275         return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
 276                              INTR_INFO_VALID_MASK)) ==
 277                 (INTR_TYPE_HARD_EXCEPTION | UD_VECTOR | INTR_INFO_VALID_MASK);
 278 }
 279
 280 static inline bool is_external_interrupt(u32 intr_info)
 281 {
 282         return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VALID_MASK))
 283                 == (INTR_TYPE_EXT_INTR | INTR_INFO_VALID_MASK);
 284 }
 285
 286 static inline bool is_machine_check(u32 intr_info)
 287 {
 288         return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
 289                              INTR_INFO_VALID_MASK)) ==
 290                 (INTR_TYPE_HARD_EXCEPTION | MC_VECTOR | INTR_INFO_VALID_MASK);
 291 }
 292
 293 static inline bool cpu_has_vmx_msr_bitmap(void)
 294 {
 295         return vmcs_config.cpu_based_exec_ctrl & CPU_BASED_USE_MSR_BITMAPS;
 296 }
 297
 298 static inline bool cpu_has_vmx_tpr_shadow(void)
 299 {
 300         return vmcs_config.cpu_based_exec_ctrl & CPU_BASED_TPR_SHADOW;
 301 }
 302
 303 static inline bool vm_need_tpr_shadow(struct kvm *kvm)
 304 {
 305         return (cpu_has_vmx_tpr_shadow()) && (irqchip_in_kernel(kvm));
 306 }
 307
 308 static inline bool cpu_has_secondary_exec_ctrls(void)
 309 {
 310         return vmcs_config.cpu_based_exec_ctrl &
 311                 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
 312 }
 313
 314 static inline bool cpu_has_vmx_virtualize_apic_accesses(void)
 315 {
 316         return vmcs_config.cpu_based_2nd_exec_ctrl &
 317                 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
 318 }
 319
 320 static inline bool cpu_has_vmx_flexpriority(void)
 321 {
 322         return cpu_has_vmx_tpr_shadow() &&
 323                 cpu_has_vmx_virtualize_apic_accesses();
 324 }
 325
 326 static inline bool cpu_has_vmx_ept_execute_only(void)
 327 {
 328         return vmx_capability.ept & VMX_EPT_EXECUTE_ONLY_BIT;
 329 }
 330
 331 static inline bool cpu_has_vmx_eptp_uncacheable(void)
 332 {
 333         return vmx_capability.ept & VMX_EPTP_UC_BIT;
 334 }
 335
 336 static inline bool cpu_has_vmx_eptp_writeback(void)
 337 {
 338         return vmx_capability.ept & VMX_EPTP_WB_BIT;
 339 }
 340
 341 static inline bool cpu_has_vmx_ept_2m_page(void)
 342 {
 343         return vmx_capability.ept & VMX_EPT_2MB_PAGE_BIT;
 344 }
 345
 346 static inline bool cpu_has_vmx_ept_1g_page(void)
 347 {
 348         return vmx_capability.ept & VMX_EPT_1GB_PAGE_BIT;
 349 }
 350
 351 static inline bool cpu_has_vmx_ept_4levels(void)
 352 {
 353         return vmx_capability.ept & VMX_EPT_PAGE_WALK_4_BIT;
 354 }
 355
 356 static inline bool cpu_has_vmx_invept_individual_addr(void)
 357 {
 358         return vmx_capability.ept & VMX_EPT_EXTENT_INDIVIDUAL_BIT;
 359 }
 360
 361 static inline bool cpu_has_vmx_invept_context(void)
 362 {
 363         return vmx_capability.ept & VMX_EPT_EXTENT_CONTEXT_BIT;
 364 }
 365
 366 static inline bool cpu_has_vmx_invept_global(void)
 367 {
 368         return vmx_capability.ept & VMX_EPT_EXTENT_GLOBAL_BIT;
 369 }
 370
 371 static inline bool cpu_has_vmx_invvpid_single(void)
 372 {
 373         return vmx_capability.vpid & VMX_VPID_EXTENT_SINGLE_CONTEXT_BIT;
 374 }
 375
 376 static inline bool cpu_has_vmx_invvpid_global(void)
 377 {
 378         return vmx_capability.vpid & VMX_VPID_EXTENT_GLOBAL_CONTEXT_BIT;
 379 }
 380
 381 static inline bool cpu_has_vmx_ept(void)
 382 {
 383         return vmcs_config.cpu_based_2nd_exec_ctrl &
 384                 SECONDARY_EXEC_ENABLE_EPT;
 385 }
 386
 387 static inline bool cpu_has_vmx_unrestricted_guest(void)
 388 {
 389         return vmcs_config.cpu_based_2nd_exec_ctrl &
 390                 SECONDARY_EXEC_UNRESTRICTED_GUEST;
 391 }
 392
 393 static inline bool cpu_has_vmx_ple(void)
 394 {
 395         return vmcs_config.cpu_based_2nd_exec_ctrl &
 396                 SECONDARY_EXEC_PAUSE_LOOP_EXITING;
 397 }
 398
 399 static inline bool vm_need_virtualize_apic_accesses(struct kvm *kvm)
 400 {
 401         return flexpriority_enabled && irqchip_in_kernel(kvm);
 402 }
 403
 404 static inline bool cpu_has_vmx_vpid(void)
 405 {
 406         return vmcs_config.cpu_based_2nd_exec_ctrl &
 407                 SECONDARY_EXEC_ENABLE_VPID;
 408 }
 409
 410 static inline bool cpu_has_vmx_rdtscp(void)
 411 {
 412         return vmcs_config.cpu_based_2nd_exec_ctrl &
 413                 SECONDARY_EXEC_RDTSCP;
 414 }
 415
 416 static inline bool cpu_has_virtual_nmis(void)
 417 {
 418         return vmcs_config.pin_based_exec_ctrl & PIN_BASED_VIRTUAL_NMIS;
 419 }
 420
 421 static inline bool cpu_has_vmx_wbinvd_exit(void)
 422 {
 423         return vmcs_config.cpu_based_2nd_exec_ctrl &
 424                 SECONDARY_EXEC_WBINVD_EXITING;
 425 }
 426
 427 static inline bool report_flexpriority(void)
 428 {
 429         return flexpriority_enabled;
 430 }
 431
 432 static int __find_msr_index(struct vcpu_vmx *vmx, u32 msr)
 433 {
 434         int i;
 435
 436         for (i = 0; i < vmx->nmsrs; ++i)
 437                 if (vmx_msr_index[vmx->guest_msrs[i].index] == msr)
 438                         return i;
 439         return -1;
 440 }
 441
 442 static inline void __invvpid(int ext, u16 vpid, gva_t gva)
 443 {
 444     struct {
 445         u64 vpid : 16;
 446         u64 rsvd : 48;
 447         u64 gva;
 448     } operand = { vpid, 0, gva };
 449
 450     asm volatile (__ex(ASM_VMX_INVVPID)
 451                   /* CF==1 or ZF==1 --> rc = -1 */
 452                   "; ja 1f ; ud2 ; 1:"
 453                   : : "a"(&operand), "c"(ext) : "cc", "memory");
 454 }
 455
 456 static inline void __invept(int ext, u64 eptp, gpa_t gpa)
 457 {
 458         struct {
 459                 u64 eptp, gpa;
 460         } operand = {eptp, gpa};
 461
 462         asm volatile (__ex(ASM_VMX_INVEPT)
 463                         /* CF==1 or ZF==1 --> rc = -1 */
 464                         "; ja 1f ; ud2 ; 1:\n"
 465                         : : "a" (&operand), "c" (ext) : "cc", "memory");
 466 }
 467
 468 static struct shared_msr_entry *find_msr_entry(struct vcpu_vmx *vmx, u32 msr)
 469 {
 470         int i;
 471
 472         i = __find_msr_index(vmx, msr);
 473         if (i >= 0)
 474                 return &vmx->guest_msrs[i];
 475         return NULL;
 476 }
 477
 478 static void vmcs_clear(struct vmcs *vmcs)
 479 {
 480         u64 phys_addr = __pa(vmcs);
 481         u8 error;
 482
 483         asm volatile (__ex(ASM_VMX_VMCLEAR_RAX) "; setna %0"
 484                       : "=qm"(error) : "a"(&phys_addr), "m"(phys_addr)
 485                       : "cc", "memory");
 486         if (error)
 487                 printk(KERN_ERR "kvm: vmclear fail: %p/%llx\n",
 488                        vmcs, phys_addr);
 489 }
 490
 491 static void vmcs_load(struct vmcs *vmcs)
 492 {
 493         u64 phys_addr = __pa(vmcs);
 494         u8 error;
 495
 496         asm volatile (__ex(ASM_VMX_VMPTRLD_RAX) "; setna %0"
 497                         : "=qm"(error) : "a"(&phys_addr), "m"(phys_addr)
 498                         : "cc", "memory");
 499         if (error)
 500                 printk(KERN_ERR "kvm: vmptrld %p/%llx fail\n",
 501                        vmcs, phys_addr);
 502 }
 503
 504 static void __vcpu_clear(void *arg)
 505 {
 506         struct vcpu_vmx *vmx = arg;
 507         int cpu = raw_smp_processor_id();
 508
 509         if (vmx->vcpu.cpu == cpu)
 510                 vmcs_clear(vmx->vmcs);
 511         if (per_cpu(current_vmcs, cpu) == vmx->vmcs)
 512                 per_cpu(current_vmcs, cpu) = NULL;
 513         list_del(&vmx->local_vcpus_link);
 514         vmx->vcpu.cpu = -1;
 515         vmx->launched = 0;
 516 }
 517
 518 static void vcpu_clear(struct vcpu_vmx *vmx)
 519 {
 520         if (vmx->vcpu.cpu == -1)
 521                 return;
 522         smp_call_function_single(vmx->vcpu.cpu, __vcpu_clear, vmx, 1);
 523 }
 524
 525 static inline void vpid_sync_vcpu_single(struct vcpu_vmx *vmx)
 526 {
 527         if (vmx->vpid == 0)
 528                 return;
 529
 530         if (cpu_has_vmx_invvpid_single())
 531                 __invvpid(VMX_VPID_EXTENT_SINGLE_CONTEXT, vmx->vpid, 0);
 532 }
 533
 534 static inline void vpid_sync_vcpu_global(void)
 535 {
 536         if (cpu_has_vmx_invvpid_global())
 537                 __invvpid(VMX_VPID_EXTENT_ALL_CONTEXT, 0, 0);
 538 }
 539
 540 static inline void vpid_sync_context(struct vcpu_vmx *vmx)
 541 {
 542         if (cpu_has_vmx_invvpid_single())
 543                 vpid_sync_vcpu_single(vmx);
 544         else
 545                 vpid_sync_vcpu_global();
 546 }
 547
 548 static inline void ept_sync_global(void)
 549 {
 550         if (cpu_has_vmx_invept_global())
 551                 __invept(VMX_EPT_EXTENT_GLOBAL, 0, 0);
 552 }
 553
 554 static inline void ept_sync_context(u64 eptp)
 555 {
 556         if (enable_ept) {
 557                 if (cpu_has_vmx_invept_context())
 558                         __invept(VMX_EPT_EXTENT_CONTEXT, eptp, 0);
 559                 else
 560                         ept_sync_global();
 561         }
 562 }
 563
 564 static inline void ept_sync_individual_addr(u64 eptp, gpa_t gpa)
 565 {
 566         if (enable_ept) {
 567                 if (cpu_has_vmx_invept_individual_addr())
 568                         __invept(VMX_EPT_EXTENT_INDIVIDUAL_ADDR,
 569                                         eptp, gpa);
 570                 else
 571                         ept_sync_context(eptp);
 572         }
 573 }
 574
 575 static unsigned long vmcs_readl(unsigned long field)
 576 {
 577         unsigned long value = 0;
 578
 579         asm volatile (__ex(ASM_VMX_VMREAD_RDX_RAX)
 580                       : "+a"(value) : "d"(field) : "cc");
 581         return value;
 582 }
 583
 584 static u16 vmcs_read16(unsigned long field)
 585 {
 586         return vmcs_readl(field);
 587 }
 588
 589 static u32 vmcs_read32(unsigned long field)
 590 {
 591         return vmcs_readl(field);
 592 }
 593
 594 static u64 vmcs_read64(unsigned long field)
 595 {
 596 #ifdef CONFIG_X86_64
 597         return vmcs_readl(field);
 598 #else
 599         return vmcs_readl(field) | ((u64)vmcs_readl(field+1) << 32);
 600 #endif
 601 }
 602
 603 static noinline void vmwrite_error(unsigned long field, unsigned long value)
 604 {
 605         printk(KERN_ERR "vmwrite error: reg %lx value %lx (err %d)\n",
 606                field, value, vmcs_read32(VM_INSTRUCTION_ERROR));
 607         dump_stack();
 608 }
 609
 610 static void vmcs_writel(unsigned long field, unsigned long value)
 611 {
 612         u8 error;
 613
 614         asm volatile (__ex(ASM_VMX_VMWRITE_RAX_RDX) "; setna %0"
 615                        : "=q"(error) : "a"(value), "d"(field) : "cc");
 616         if (unlikely(error))
 617                 vmwrite_error(field, value);
 618 }
 619
 620 static void vmcs_write16(unsigned long field, u16 value)
 621 {
 622         vmcs_writel(field, value);
 623 }
 624
 625 static void vmcs_write32(unsigned long field, u32 value)
 626 {
 627         vmcs_writel(field, value);
 628 }
 629
 630 static void vmcs_write64(unsigned long field, u64 value)
 631 {
 632         vmcs_writel(field, value);
 633 #ifndef CONFIG_X86_64
 634         asm volatile ("");
 635         vmcs_writel(field+1, value >> 32);
 636 #endif
 637 }
 638
 639 static void vmcs_clear_bits(unsigned long field, u32 mask)
 640 {
 641         vmcs_writel(field, vmcs_readl(field) & ~mask);
 642 }
 643
 644 static void vmcs_set_bits(unsigned long field, u32 mask)
 645 {
 646         vmcs_writel(field, vmcs_readl(field) | mask);
 647 }
 648
 649 static void update_exception_bitmap(struct kvm_vcpu *vcpu)
 650 {
 651         u32 eb;
 652
 653         eb = (1u << PF_VECTOR) | (1u << UD_VECTOR) | (1u << MC_VECTOR) |
 654              (1u << NM_VECTOR) | (1u << DB_VECTOR);
 655         if ((vcpu->guest_debug &
 656              (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP)) ==
 657             (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP))
 658                 eb |= 1u << BP_VECTOR;
 659         if (to_vmx(vcpu)->rmode.vm86_active)
 660                 eb = ~0;
 661         if (enable_ept)
 662                 eb &= ~(1u << PF_VECTOR); /* bypass_guest_pf = 0 */
 663         if (vcpu->fpu_active)
 664                 eb &= ~(1u << NM_VECTOR);
 665         vmcs_write32(EXCEPTION_BITMAP, eb);
 666 }
 667
 668 static void clear_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr)
 669 {
 670         unsigned i;
 671         struct msr_autoload *m = &vmx->msr_autoload;
 672
 673         if (msr == MSR_EFER && cpu_has_load_ia32_efer) {
 674                 vmcs_clear_bits(VM_ENTRY_CONTROLS, VM_ENTRY_LOAD_IA32_EFER);
 675                 vmcs_clear_bits(VM_EXIT_CONTROLS, VM_EXIT_LOAD_IA32_EFER);
 676                 return;
 677         }
 678
 679         for (i = 0; i < m->nr; ++i)
 680                 if (m->guest[i].index == msr)
 681                         break;
 682
 683         if (i == m->nr)
 684                 return;
 685         --m->nr;
 686         m->guest[i] = m->guest[m->nr];
 687         m->host[i] = m->host[m->nr];
 688         vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->nr);
 689         vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->nr);
 690 }
 691
 692 static void add_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr,
 693                                   u64 guest_val, u64 host_val)
 694 {
 695         unsigned i;
 696         struct msr_autoload *m = &vmx->msr_autoload;
 697
 698         if (msr == MSR_EFER && cpu_has_load_ia32_efer) {
 699                 vmcs_write64(GUEST_IA32_EFER, guest_val);
 700                 vmcs_write64(HOST_IA32_EFER, host_val);
 701                 vmcs_set_bits(VM_ENTRY_CONTROLS, VM_ENTRY_LOAD_IA32_EFER);
 702                 vmcs_set_bits(VM_EXIT_CONTROLS, VM_EXIT_LOAD_IA32_EFER);
 703                 return;
 704         }
 705
 706         for (i = 0; i < m->nr; ++i)
 707                 if (m->guest[i].index == msr)
 708                         break;
 709
 710         if (i == m->nr) {
 711                 ++m->nr;
 712                 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->nr);
 713                 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->nr);
 714         }
 715
 716         m->guest[i].index = msr;
 717         m->guest[i].value = guest_val;
 718         m->host[i].index = msr;
 719         m->host[i].value = host_val;
 720 }
 721
 722 static void reload_tss(void)
 723 {
 724         /*
 725          * VT restores TR but not its size.  Useless.
 726          */
 727         struct desc_ptr *gdt = &__get_cpu_var(host_gdt);
 728         struct desc_struct *descs;
 729
 730         descs = (void *)gdt->address;
 731         descs[GDT_ENTRY_TSS].type = 9; /* available TSS */
 732         load_TR_desc();
 733 }
 734
 735 static bool update_transition_efer(struct vcpu_vmx *vmx, int efer_offset)
 736 {
 737         u64 guest_efer;
 738         u64 ignore_bits;
 739
 740         guest_efer = vmx->vcpu.arch.efer;
 741
 742         /*
 743          * NX is emulated; LMA and LME handled by hardware; SCE meaninless
 744          * outside long mode
 745          */
 746         ignore_bits = EFER_NX | EFER_SCE;
 747 #ifdef CONFIG_X86_64
 748         ignore_bits |= EFER_LMA | EFER_LME;
 749         /* SCE is meaningful only in long mode on Intel */
 750         if (guest_efer & EFER_LMA)
 751                 ignore_bits &= ~(u64)EFER_SCE;
 752 #endif
 753         guest_efer &= ~ignore_bits;
 754         guest_efer |= host_efer & ignore_bits;
 755         vmx->guest_msrs[efer_offset].data = guest_efer;
 756         vmx->guest_msrs[efer_offset].mask = ~ignore_bits;
 757
 758         clear_atomic_switch_msr(vmx, MSR_EFER);
 759         /* On ept, can't emulate nx, and must switch nx atomically */
 760         if (enable_ept && ((vmx->vcpu.arch.efer ^ host_efer) & EFER_NX)) {
 761                 guest_efer = vmx->vcpu.arch.efer;
 762                 if (!(guest_efer & EFER_LMA))
 763                         guest_efer &= ~EFER_LME;
 764                 add_atomic_switch_msr(vmx, MSR_EFER, guest_efer, host_efer);
 765                 return false;
 766         }
 767
 768         return true;
 769 }
 770
 771 static unsigned long segment_base(u16 selector)
 772 {
 773         struct desc_ptr *gdt = &__get_cpu_var(host_gdt);
 774         struct desc_struct *d;
 775         unsigned long table_base;
 776         unsigned long v;
 777
 778         if (!(selector & ~3))
 779                 return 0;
 780
 781         table_base = gdt->address;
 782
 783         if (selector & 4) {           /* from ldt */
 784                 u16 ldt_selector = kvm_read_ldt();
 785
 786                 if (!(ldt_selector & ~3))
 787                         return 0;
 788
 789                 table_base = segment_base(ldt_selector);
 790         }
 791         d = (struct desc_struct *)(table_base + (selector & ~7));
 792         v = get_desc_base(d);
 793 #ifdef CONFIG_X86_64
 794        if (d->s == 0 && (d->type == 2 || d->type == 9 || d->type == 11))
 795                v |= ((unsigned long)((struct ldttss_desc64 *)d)->base3) << 32;
 796 #endif
 797         return v;
 798 }
 799
 800 static inline unsigned long kvm_read_tr_base(void)
 801 {
 802         u16 tr;
 803         asm("str %0" : "=g"(tr));
 804         return segment_base(tr);
 805 }
 806
 807 static void vmx_save_host_state(struct kvm_vcpu *vcpu)
 808 {
 809         struct vcpu_vmx *vmx = to_vmx(vcpu);
 810         int i;
 811
 812         if (vmx->host_state.loaded)
 813                 return;
 814
 815         vmx->host_state.loaded = 1;
 816         /*
 817          * Set host fs and gs selectors.  Unfortunately, 22.2.3 does not
 818          * allow segment selectors with cpl > 0 or ti == 1.
 819          */
 820         vmx->host_state.ldt_sel = kvm_read_ldt();
 821         vmx->host_state.gs_ldt_reload_needed = vmx->host_state.ldt_sel;
 822         savesegment(fs, vmx->host_state.fs_sel);
 823         if (!(vmx->host_state.fs_sel & 7)) {
 824                 vmcs_write16(HOST_FS_SELECTOR, vmx->host_state.fs_sel);
 825                 vmx->host_state.fs_reload_needed = 0;
 826         } else {
 827                 vmcs_write16(HOST_FS_SELECTOR, 0);
 828                 vmx->host_state.fs_reload_needed = 1;
 829         }
 830         savesegment(gs, vmx->host_state.gs_sel);
 831         if (!(vmx->host_state.gs_sel & 7))
 832                 vmcs_write16(HOST_GS_SELECTOR, vmx->host_state.gs_sel);
 833         else {
 834                 vmcs_write16(HOST_GS_SELECTOR, 0);
 835                 vmx->host_state.gs_ldt_reload_needed = 1;
 836         }
 837
 838 #ifdef CONFIG_X86_64
 839         vmcs_writel(HOST_FS_BASE, read_msr(MSR_FS_BASE));
 840         vmcs_writel(HOST_GS_BASE, read_msr(MSR_GS_BASE));
 841 #else
 842         vmcs_writel(HOST_FS_BASE, segment_base(vmx->host_state.fs_sel));
 843         vmcs_writel(HOST_GS_BASE, segment_base(vmx->host_state.gs_sel));
 844 #endif
 845
 846 #ifdef CONFIG_X86_64
 847         rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
 848         if (is_long_mode(&vmx->vcpu))
 849                 wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
 850 #endif
 851         for (i = 0; i < vmx->save_nmsrs; ++i)
 852                 kvm_set_shared_msr(vmx->guest_msrs[i].index,
 853                                    vmx->guest_msrs[i].data,
 854                                    vmx->guest_msrs[i].mask);
 855 }
 856
 857 static void __vmx_load_host_state(struct vcpu_vmx *vmx)
 858 {
 859         if (!vmx->host_state.loaded)
 860                 return;
 861
 862         ++vmx->vcpu.stat.host_state_reload;
 863         vmx->host_state.loaded = 0;
 864 #ifdef CONFIG_X86_64
 865         if (is_long_mode(&vmx->vcpu))
 866                 rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
 867 #endif
 868         if (vmx->host_state.gs_ldt_reload_needed) {
 869                 kvm_load_ldt(vmx->host_state.ldt_sel);
 870 #ifdef CONFIG_X86_64
 871                 load_gs_index(vmx->host_state.gs_sel);
 872 #else
 873                 loadsegment(gs, vmx->host_state.gs_sel);
 874 #endif
 875         }
 876         if (vmx->host_state.fs_reload_needed)
 877                 loadsegment(fs, vmx->host_state.fs_sel);
 878         reload_tss();
 879 #ifdef CONFIG_X86_64
 880         wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
 881 #endif
 882         if (current_thread_info()->status & TS_USEDFPU)
 883                 clts();
 884         load_gdt(&__get_cpu_var(host_gdt));
 885 }
 886
 887 static void vmx_load_host_state(struct vcpu_vmx *vmx)
 888 {
 889         preempt_disable();
 890         __vmx_load_host_state(vmx);
 891         preempt_enable();
 892 }
 893
 894 /*
 895  * Switches to specified vcpu, until a matching vcpu_put(), but assumes
 896  * vcpu mutex is already taken.
 897  */
 898 static void vmx_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
 899 {
 900         struct vcpu_vmx *vmx = to_vmx(vcpu);
 901         u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
 902
 903         if (!vmm_exclusive)
 904                 kvm_cpu_vmxon(phys_addr);
 905         else if (vcpu->cpu != cpu)
 906                 vcpu_clear(vmx);
 907
 908         if (per_cpu(current_vmcs, cpu) != vmx->vmcs) {
 909                 per_cpu(current_vmcs, cpu) = vmx->vmcs;
 910                 vmcs_load(vmx->vmcs);
 911         }
 912
 913         if (vcpu->cpu != cpu) {
 914                 struct desc_ptr *gdt = &__get_cpu_var(host_gdt);
 915                 unsigned long sysenter_esp;
 916
 917                 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
 918                 local_irq_disable();
 919                 list_add(&vmx->local_vcpus_link,
 920                          &per_cpu(vcpus_on_cpu, cpu));
 921                 local_irq_enable();
 922
 923                 /*
 924                  * Linux uses per-cpu TSS and GDT, so set these when switching
 925                  * processors.
 926                  */
 927                 vmcs_writel(HOST_TR_BASE, kvm_read_tr_base()); /* 22.2.4 */
 928                 vmcs_writel(HOST_GDTR_BASE, gdt->address);   /* 22.2.4 */
 929
 930                 rdmsrl(MSR_IA32_SYSENTER_ESP, sysenter_esp);
 931                 vmcs_writel(HOST_IA32_SYSENTER_ESP, sysenter_esp); /* 22.2.3 */
 932         }
 933 }
 934
 935 static void vmx_vcpu_put(struct kvm_vcpu *vcpu)
 936 {
 937         __vmx_load_host_state(to_vmx(vcpu));
 938         if (!vmm_exclusive) {
 939                 __vcpu_clear(to_vmx(vcpu));
 940                 kvm_cpu_vmxoff();
 941         }
 942 }
 943
 944 static void vmx_fpu_activate(struct kvm_vcpu *vcpu)
 945 {
 946         ulong cr0;
 947
 948         if (vcpu->fpu_active)
 949                 return;
 950         vcpu->fpu_active = 1;
 951         cr0 = vmcs_readl(GUEST_CR0);
 952         cr0 &= ~(X86_CR0_TS | X86_CR0_MP);
 953         cr0 |= kvm_read_cr0_bits(vcpu, X86_CR0_TS | X86_CR0_MP);
 954         vmcs_writel(GUEST_CR0, cr0);
 955         update_exception_bitmap(vcpu);
 956         vcpu->arch.cr0_guest_owned_bits = X86_CR0_TS;
 957         vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
 958 }
 959
 960 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu);
 961
 962 static void vmx_fpu_deactivate(struct kvm_vcpu *vcpu)
 963 {
 964         vmx_decache_cr0_guest_bits(vcpu);
 965         vmcs_set_bits(GUEST_CR0, X86_CR0_TS | X86_CR0_MP);
 966         update_exception_bitmap(vcpu);
 967         vcpu->arch.cr0_guest_owned_bits = 0;
 968         vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
 969         vmcs_writel(CR0_READ_SHADOW, vcpu->arch.cr0);
 970 }
 971
 972 static unsigned long vmx_get_rflags(struct kvm_vcpu *vcpu)
 973 {
 974         unsigned long rflags, save_rflags;
 975
 976         if (!test_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail)) {
 977                 __set_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail);
 978                 rflags = vmcs_readl(GUEST_RFLAGS);
 979                 if (to_vmx(vcpu)->rmode.vm86_active) {
 980                         rflags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
 981                         save_rflags = to_vmx(vcpu)->rmode.save_rflags;
 982                         rflags |= save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
 983                 }
 984                 to_vmx(vcpu)->rflags = rflags;
 985         }
 986         return to_vmx(vcpu)->rflags;
 987 }
 988
 989 static void vmx_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
 990 {
 991         __set_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail);
 992         __clear_bit(VCPU_EXREG_CPL, (ulong *)&vcpu->arch.regs_avail);
 993         to_vmx(vcpu)->rflags = rflags;
 994         if (to_vmx(vcpu)->rmode.vm86_active) {
 995                 to_vmx(vcpu)->rmode.save_rflags = rflags;
 996                 rflags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
 997         }
 998         vmcs_writel(GUEST_RFLAGS, rflags);
 999 }
1000
1001 static u32 vmx_get_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
1002 {
1003         u32 interruptibility = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
1004         int ret = 0;
1005
1006         if (interruptibility & GUEST_INTR_STATE_STI)
1007                 ret |= KVM_X86_SHADOW_INT_STI;
1008         if (interruptibility & GUEST_INTR_STATE_MOV_SS)
1009                 ret |= KVM_X86_SHADOW_INT_MOV_SS;
1010
1011         return ret & mask;
1012 }
1013
1014 static void vmx_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
1015 {
1016         u32 interruptibility_old = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
1017         u32 interruptibility = interruptibility_old;
1018
1019         interruptibility &= ~(GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS);
1020
1021         if (mask & KVM_X86_SHADOW_INT_MOV_SS)
1022                 interruptibility |= GUEST_INTR_STATE_MOV_SS;
1023         else if (mask & KVM_X86_SHADOW_INT_STI)
1024                 interruptibility |= GUEST_INTR_STATE_STI;
1025
1026         if ((interruptibility != interruptibility_old))
1027                 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, interruptibility);
1028 }
1029
1030 static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
1031 {
1032         unsigned long rip;
1033
1034         rip = kvm_rip_read(vcpu);
1035         rip += vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
1036         kvm_rip_write(vcpu, rip);
1037
1038         /* skipping an emulated instruction also counts */
1039         vmx_set_interrupt_shadow(vcpu, 0);
1040 }
1041
1042 static void vmx_clear_hlt(struct kvm_vcpu *vcpu)
1043 {
1044         /* Ensure that we clear the HLT state in the VMCS.  We don't need to
1045          * explicitly skip the instruction because if the HLT state is set, then
1046          * the instruction is already executing and RIP has already been
1047          * advanced. */
1048         if (!yield_on_hlt &&
1049             vmcs_read32(GUEST_ACTIVITY_STATE) == GUEST_ACTIVITY_HLT)
1050                 vmcs_write32(GUEST_ACTIVITY_STATE, GUEST_ACTIVITY_ACTIVE);
1051 }
1052
1053 static void vmx_queue_exception(struct kvm_vcpu *vcpu, unsigned nr,
1054                                 bool has_error_code, u32 error_code,
1055                                 bool reinject)
1056 {
1057         struct vcpu_vmx *vmx = to_vmx(vcpu);
1058         u32 intr_info = nr | INTR_INFO_VALID_MASK;
1059
1060         if (has_error_code) {
1061                 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE, error_code);
1062                 intr_info |= INTR_INFO_DELIVER_CODE_MASK;
1063         }
1064
1065         if (vmx->rmode.vm86_active) {
1066                 int inc_eip = 0;
1067                 if (kvm_exception_is_soft(nr))
1068                         inc_eip = vcpu->arch.event_exit_inst_len;
1069                 if (kvm_inject_realmode_interrupt(vcpu, nr, inc_eip) != EMULATE_DONE)
1070                         kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
1071                 return;
1072         }
1073
1074         if (kvm_exception_is_soft(nr)) {
1075                 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
1076                              vmx->vcpu.arch.event_exit_inst_len);
1077                 intr_info |= INTR_TYPE_SOFT_EXCEPTION;
1078         } else
1079                 intr_info |= INTR_TYPE_HARD_EXCEPTION;
1080
1081         vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr_info);
1082         vmx_clear_hlt(vcpu);
1083 }
1084
1085 static bool vmx_rdtscp_supported(void)
1086 {
1087         return cpu_has_vmx_rdtscp();
1088 }
1089
1090 /*
1091  * Swap MSR entry in host/guest MSR entry array.
1092  */
1093 static void move_msr_up(struct vcpu_vmx *vmx, int from, int to)
1094 {
1095         struct shared_msr_entry tmp;
1096
1097         tmp = vmx->guest_msrs[to];
1098         vmx->guest_msrs[to] = vmx->guest_msrs[from];
1099         vmx->guest_msrs[from] = tmp;
1100 }
1101
1102 /*
1103  * Set up the vmcs to automatically save and restore system
1104  * msrs.  Don't touch the 64-bit msrs if the guest is in legacy
1105  * mode, as fiddling with msrs is very expensive.
1106  */
1107 static void setup_msrs(struct vcpu_vmx *vmx)
1108 {
1109         int save_nmsrs, index;
1110         unsigned long *msr_bitmap;
1111
1112         vmx_load_host_state(vmx);
1113         save_nmsrs = 0;
1114 #ifdef CONFIG_X86_64
1115         if (is_long_mode(&vmx->vcpu)) {
1116                 index = __find_msr_index(vmx, MSR_SYSCALL_MASK);
1117                 if (index >= 0)
1118                         move_msr_up(vmx, index, save_nmsrs++);
1119                 index = __find_msr_index(vmx, MSR_LSTAR);
1120                 if (index >= 0)
1121                         move_msr_up(vmx, index, save_nmsrs++);
1122                 index = __find_msr_index(vmx, MSR_CSTAR);
1123                 if (index >= 0)
1124                         move_msr_up(vmx, index, save_nmsrs++);
1125                 index = __find_msr_index(vmx, MSR_TSC_AUX);
1126                 if (index >= 0 && vmx->rdtscp_enabled)
1127                         move_msr_up(vmx, index, save_nmsrs++);
1128                 /*
1129                  * MSR_STAR is only needed on long mode guests, and only
1130                  * if efer.sce is enabled.
1131                  */
1132                 index = __find_msr_index(vmx, MSR_STAR);
1133                 if ((index >= 0) && (vmx->vcpu.arch.efer & EFER_SCE))
1134                         move_msr_up(vmx, index, save_nmsrs++);
1135         }
1136 #endif
1137         index = __find_msr_index(vmx, MSR_EFER);
1138         if (index >= 0 && update_transition_efer(vmx, index))
1139                 move_msr_up(vmx, index, save_nmsrs++);
1140
1141         vmx->save_nmsrs = save_nmsrs;
1142
1143         if (cpu_has_vmx_msr_bitmap()) {
1144                 if (is_long_mode(&vmx->vcpu))
1145                         msr_bitmap = vmx_msr_bitmap_longmode;
1146                 else
1147                         msr_bitmap = vmx_msr_bitmap_legacy;
1148
1149                 vmcs_write64(MSR_BITMAP, __pa(msr_bitmap));
1150         }
1151 }
1152
1153 /*
1154  * reads and returns guest's timestamp counter "register"
1155  * guest_tsc = host_tsc + tsc_offset    -- 21.3
1156  */
1157 static u64 guest_read_tsc(void)
1158 {
1159         u64 host_tsc, tsc_offset;
1160
1161         rdtscll(host_tsc);
1162         tsc_offset = vmcs_read64(TSC_OFFSET);
1163         return host_tsc + tsc_offset;
1164 }
1165
1166 /*
1167  * Empty call-back. Needs to be implemented when VMX enables the SET_TSC_KHZ
1168  * ioctl. In this case the call-back should update internal vmx state to make
1169  * the changes effective.
1170  */
1171 static void vmx_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz)
1172 {
1173         /* Nothing to do here */
1174 }
1175
1176 /*
1177  * writes 'offset' into guest's timestamp counter offset register
1178  */
1179 static void vmx_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
1180 {
1181         vmcs_write64(TSC_OFFSET, offset);
1182 }
1183
1184 static void vmx_adjust_tsc_offset(struct kvm_vcpu *vcpu, s64 adjustment)
1185 {
1186         u64 offset = vmcs_read64(TSC_OFFSET);
1187         vmcs_write64(TSC_OFFSET, offset + adjustment);
1188 }
1189
1190 static u64 vmx_compute_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
1191 {
1192         return target_tsc - native_read_tsc();
1193 }
1194
1195 /*
1196  * Reads an msr value (of 'msr_index') into 'pdata'.
1197  * Returns 0 on success, non-0 otherwise.
1198  * Assumes vcpu_load() was already called.
1199  */
1200 static int vmx_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1201 {
1202         u64 data;
1203         struct shared_msr_entry *msr;
1204
1205         if (!pdata) {
1206                 printk(KERN_ERR "BUG: get_msr called with NULL pdata\n");
1207                 return -EINVAL;
1208         }
1209
1210         switch (msr_index) {
1211 #ifdef CONFIG_X86_64
1212         case MSR_FS_BASE:
1213                 data = vmcs_readl(GUEST_FS_BASE);
1214                 break;
1215         case MSR_GS_BASE:
1216                 data = vmcs_readl(GUEST_GS_BASE);
1217                 break;
1218         case MSR_KERNEL_GS_BASE:
1219                 vmx_load_host_state(to_vmx(vcpu));
1220                 data = to_vmx(vcpu)->msr_guest_kernel_gs_base;
1221                 break;
1222 #endif
1223         case MSR_EFER:
1224                 return kvm_get_msr_common(vcpu, msr_index, pdata);
1225         case MSR_IA32_TSC:
1226                 data = guest_read_tsc();
1227                 break;
1228         case MSR_IA32_SYSENTER_CS:
1229                 data = vmcs_read32(GUEST_SYSENTER_CS);
1230                 break;
1231         case MSR_IA32_SYSENTER_EIP:
1232                 data = vmcs_readl(GUEST_SYSENTER_EIP);
1233                 break;
1234         case MSR_IA32_SYSENTER_ESP:
1235                 data = vmcs_readl(GUEST_SYSENTER_ESP);
1236                 break;
1237         case MSR_TSC_AUX:
1238                 if (!to_vmx(vcpu)->rdtscp_enabled)
1239                         return 1;
1240                 /* Otherwise falls through */
1241         default:
1242                 vmx_load_host_state(to_vmx(vcpu));
1243                 msr = find_msr_entry(to_vmx(vcpu), msr_index);
1244                 if (msr) {
1245                         vmx_load_host_state(to_vmx(vcpu));
1246                         data = msr->data;
1247                         break;
1248                 }
1249                 return kvm_get_msr_common(vcpu, msr_index, pdata);
1250         }
1251
1252         *pdata = data;
1253         return 0;
1254 }
1255
1256 /*
1257  * Writes msr value into into the appropriate "register".
1258  * Returns 0 on success, non-0 otherwise.
1259  * Assumes vcpu_load() was already called.
1260  */
1261 static int vmx_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
1262 {
1263         struct vcpu_vmx *vmx = to_vmx(vcpu);
1264         struct shared_msr_entry *msr;
1265         int ret = 0;
1266
1267         switch (msr_index) {
1268         case MSR_EFER:
1269                 vmx_load_host_state(vmx);
1270                 ret = kvm_set_msr_common(vcpu, msr_index, data);
1271                 break;
1272 #ifdef CONFIG_X86_64
1273         case MSR_FS_BASE:
1274                 vmcs_writel(GUEST_FS_BASE, data);
1275                 break;
1276         case MSR_GS_BASE:
1277                 vmcs_writel(GUEST_GS_BASE, data);
1278                 break;
1279         case MSR_KERNEL_GS_BASE:
1280                 vmx_load_host_state(vmx);
1281                 vmx->msr_guest_kernel_gs_base = data;
1282                 break;
1283 #endif
1284         case MSR_IA32_SYSENTER_CS:
1285                 vmcs_write32(GUEST_SYSENTER_CS, data);
1286                 break;
1287         case MSR_IA32_SYSENTER_EIP:
1288                 vmcs_writel(GUEST_SYSENTER_EIP, data);
1289                 break;
1290         case MSR_IA32_SYSENTER_ESP:
1291                 vmcs_writel(GUEST_SYSENTER_ESP, data);
1292                 break;
1293         case MSR_IA32_TSC:
1294                 kvm_write_tsc(vcpu, data);
1295                 break;
1296         case MSR_IA32_CR_PAT:
1297                 if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
1298                         vmcs_write64(GUEST_IA32_PAT, data);
1299                         vcpu->arch.pat = data;
1300                         break;
1301                 }
1302                 ret = kvm_set_msr_common(vcpu, msr_index, data);
1303                 break;
1304         case MSR_TSC_AUX:
1305                 if (!vmx->rdtscp_enabled)
1306                         return 1;
1307                 /* Check reserved bit, higher 32 bits should be zero */
1308                 if ((data >> 32) != 0)
1309                         return 1;
1310                 /* Otherwise falls through */
1311         default:
1312                 msr = find_msr_entry(vmx, msr_index);
1313                 if (msr) {
1314                         vmx_load_host_state(vmx);
1315                         msr->data = data;
1316                         break;
1317                 }
1318                 ret = kvm_set_msr_common(vcpu, msr_index, data);
1319         }
1320
1321         return ret;
1322 }
1323
1324 static void vmx_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
1325 {
1326         __set_bit(reg, (unsigned long *)&vcpu->arch.regs_avail);
1327         switch (reg) {
1328         case VCPU_REGS_RSP:
1329                 vcpu->arch.regs[VCPU_REGS_RSP] = vmcs_readl(GUEST_RSP);
1330                 break;
1331         case VCPU_REGS_RIP:
1332                 vcpu->arch.regs[VCPU_REGS_RIP] = vmcs_readl(GUEST_RIP);
1333                 break;
1334         case VCPU_EXREG_PDPTR:
1335                 if (enable_ept)
1336                         ept_save_pdptrs(vcpu);
1337                 break;
1338         default:
1339                 break;
1340         }
1341 }
1342
1343 static void set_guest_debug(struct kvm_vcpu *vcpu, struct kvm_guest_debug *dbg)
1344 {
1345         if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1346                 vmcs_writel(GUEST_DR7, dbg->arch.debugreg[7]);
1347         else
1348                 vmcs_writel(GUEST_DR7, vcpu->arch.dr7);
1349
1350         update_exception_bitmap(vcpu);
1351 }
1352
1353 static __init int cpu_has_kvm_support(void)
1354 {
1355         return cpu_has_vmx();
1356 }
1357
1358 static __init int vmx_disabled_by_bios(void)
1359 {
1360         u64 msr;
1361
1362         rdmsrl(MSR_IA32_FEATURE_CONTROL, msr);
1363         if (msr & FEATURE_CONTROL_LOCKED) {
1364                 /* launched w/ TXT and VMX disabled */
1365                 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX)
1366                         && tboot_enabled())
1367                         return 1;
1368                 /* launched w/o TXT and VMX only enabled w/ TXT */
1369                 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX)
1370                         && (msr & FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX)
1371                         && !tboot_enabled()) {
1372                         printk(KERN_WARNING "kvm: disable TXT in the BIOS or "
1373                                 "activate TXT before enabling KVM\n");
1374                         return 1;
1375                 }
1376                 /* launched w/o TXT and VMX disabled */
1377                 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX)
1378                         && !tboot_enabled())
1379                         return 1;
1380         }
1381
1382         return 0;
1383 }
1384
1385 static void kvm_cpu_vmxon(u64 addr)
1386 {
1387         asm volatile (ASM_VMX_VMXON_RAX
1388                         : : "a"(&addr), "m"(addr)
1389                         : "memory", "cc");
1390 }
1391
1392 static int hardware_enable(void *garbage)
1393 {
1394         int cpu = raw_smp_processor_id();
1395         u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
1396         u64 old, test_bits;
1397
1398         if (read_cr4() & X86_CR4_VMXE)
1399                 return -EBUSY;
1400
1401         INIT_LIST_HEAD(&per_cpu(vcpus_on_cpu, cpu));
1402         rdmsrl(MSR_IA32_FEATURE_CONTROL, old);
1403
1404         test_bits = FEATURE_CONTROL_LOCKED;
1405         test_bits |= FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
1406         if (tboot_enabled())
1407                 test_bits |= FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX;
1408
1409         if ((old & test_bits) != test_bits) {
1410                 /* enable and lock */
1411                 wrmsrl(MSR_IA32_FEATURE_CONTROL, old | test_bits);
1412         }
1413         write_cr4(read_cr4() | X86_CR4_VMXE); /* FIXME: not cpu hotplug safe */
1414
1415         if (vmm_exclusive) {
1416                 kvm_cpu_vmxon(phys_addr);
1417                 ept_sync_global();
1418         }
1419
1420         store_gdt(&__get_cpu_var(host_gdt));
1421
1422         return 0;
1423 }
1424
1425 static void vmclear_local_vcpus(void)
1426 {
1427         int cpu = raw_smp_processor_id();
1428         struct vcpu_vmx *vmx, *n;
1429
1430         list_for_each_entry_safe(vmx, n, &per_cpu(vcpus_on_cpu, cpu),
1431                                  local_vcpus_link)
1432                 __vcpu_clear(vmx);
1433 }
1434
1435
1436 /* Just like cpu_vmxoff(), but with the __kvm_handle_fault_on_reboot()
1437  * tricks.
1438  */
1439 static void kvm_cpu_vmxoff(void)
1440 {
1441         asm volatile (__ex(ASM_VMX_VMXOFF) : : : "cc");
1442 }
1443
1444 static void hardware_disable(void *garbage)
1445 {
1446         if (vmm_exclusive) {
1447                 vmclear_local_vcpus();
1448                 kvm_cpu_vmxoff();
1449         }
1450         write_cr4(read_cr4() & ~X86_CR4_VMXE);
1451 }
1452
1453 static __init int adjust_vmx_controls(u32 ctl_min, u32 ctl_opt,
1454                                       u32 msr, u32 *result)
1455 {
1456         u32 vmx_msr_low, vmx_msr_high;
1457         u32 ctl = ctl_min | ctl_opt;
1458
1459         rdmsr(msr, vmx_msr_low, vmx_msr_high);
1460
1461         ctl &= vmx_msr_high; /* bit == 0 in high word ==> must be zero */
1462         ctl |= vmx_msr_low;  /* bit == 1 in low word  ==> must be one  */
1463
1464         /* Ensure minimum (required) set of control bits are supported. */
1465         if (ctl_min & ~ctl)
1466                 return -EIO;
1467
1468         *result = ctl;
1469         return 0;
1470 }
1471
1472 static __init bool allow_1_setting(u32 msr, u32 ctl)
1473 {
1474         u32 vmx_msr_low, vmx_msr_high;
1475
1476         rdmsr(msr, vmx_msr_low, vmx_msr_high);
1477         return vmx_msr_high & ctl;
1478 }
1479
1480 static __init int setup_vmcs_config(struct vmcs_config *vmcs_conf)
1481 {
1482         u32 vmx_msr_low, vmx_msr_high;
1483         u32 min, opt, min2, opt2;
1484         u32 _pin_based_exec_control = 0;
1485         u32 _cpu_based_exec_control = 0;
1486         u32 _cpu_based_2nd_exec_control = 0;
1487         u32 _vmexit_control = 0;
1488         u32 _vmentry_control = 0;
1489
1490         min = PIN_BASED_EXT_INTR_MASK | PIN_BASED_NMI_EXITING;
1491         opt = PIN_BASED_VIRTUAL_NMIS;
1492         if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PINBASED_CTLS,
1493                                 &_pin_based_exec_control) < 0)
1494                 return -EIO;
1495
1496         min =
1497 #ifdef CONFIG_X86_64
1498               CPU_BASED_CR8_LOAD_EXITING |
1499               CPU_BASED_CR8_STORE_EXITING |
1500 #endif
1501               CPU_BASED_CR3_LOAD_EXITING |
1502               CPU_BASED_CR3_STORE_EXITING |
1503               CPU_BASED_USE_IO_BITMAPS |
1504               CPU_BASED_MOV_DR_EXITING |
1505               CPU_BASED_USE_TSC_OFFSETING |
1506               CPU_BASED_MWAIT_EXITING |
1507               CPU_BASED_MONITOR_EXITING |
1508               CPU_BASED_INVLPG_EXITING;
1509
1510         if (yield_on_hlt)
1511                 min |= CPU_BASED_HLT_EXITING;
1512
1513         opt = CPU_BASED_TPR_SHADOW |
1514               CPU_BASED_USE_MSR_BITMAPS |
1515               CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
1516         if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PROCBASED_CTLS,
1517                                 &_cpu_based_exec_control) < 0)
1518                 return -EIO;
1519 #ifdef CONFIG_X86_64
1520         if ((_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
1521                 _cpu_based_exec_control &= ~CPU_BASED_CR8_LOAD_EXITING &
1522                                            ~CPU_BASED_CR8_STORE_EXITING;
1523 #endif
1524         if (_cpu_based_exec_control & CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) {
1525                 min2 = 0;
1526                 opt2 = SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
1527                         SECONDARY_EXEC_WBINVD_EXITING |
1528                         SECONDARY_EXEC_ENABLE_VPID |
1529                         SECONDARY_EXEC_ENABLE_EPT |
1530                         SECONDARY_EXEC_UNRESTRICTED_GUEST |
1531                         SECONDARY_EXEC_PAUSE_LOOP_EXITING |
1532                         SECONDARY_EXEC_RDTSCP;
1533                 if (adjust_vmx_controls(min2, opt2,
1534                                         MSR_IA32_VMX_PROCBASED_CTLS2,
1535                                         &_cpu_based_2nd_exec_control) < 0)
1536                         return -EIO;
1537         }
1538 #ifndef CONFIG_X86_64
1539         if (!(_cpu_based_2nd_exec_control &
1540                                 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
1541                 _cpu_based_exec_control &= ~CPU_BASED_TPR_SHADOW;
1542 #endif
1543         if (_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_EPT) {
1544                 /* CR3 accesses and invlpg don't need to cause VM Exits when EPT
1545                    enabled */
1546                 _cpu_based_exec_control &= ~(CPU_BASED_CR3_LOAD_EXITING |
1547                                              CPU_BASED_CR3_STORE_EXITING |
1548                                              CPU_BASED_INVLPG_EXITING);
1549                 rdmsr(MSR_IA32_VMX_EPT_VPID_CAP,
1550                       vmx_capability.ept, vmx_capability.vpid);
1551         }
1552
1553         min = 0;
1554 #ifdef CONFIG_X86_64
1555         min |= VM_EXIT_HOST_ADDR_SPACE_SIZE;
1556 #endif
1557         opt = VM_EXIT_SAVE_IA32_PAT | VM_EXIT_LOAD_IA32_PAT;
1558         if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_EXIT_CTLS,
1559                                 &_vmexit_control) < 0)
1560                 return -EIO;
1561
1562         min = 0;
1563         opt = VM_ENTRY_LOAD_IA32_PAT;
1564         if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_ENTRY_CTLS,
1565                                 &_vmentry_control) < 0)
1566                 return -EIO;
1567
1568         rdmsr(MSR_IA32_VMX_BASIC, vmx_msr_low, vmx_msr_high);
1569
1570         /* IA-32 SDM Vol 3B: VMCS size is never greater than 4kB. */
1571         if ((vmx_msr_high & 0x1fff) > PAGE_SIZE)
1572                 return -EIO;
1573
1574 #ifdef CONFIG_X86_64
1575         /* IA-32 SDM Vol 3B: 64-bit CPUs always have VMX_BASIC_MSR[48]==0. */
1576         if (vmx_msr_high & (1u<<16))
1577                 return -EIO;
1578 #endif
1579
1580         /* Require Write-Back (WB) memory type for VMCS accesses. */
1581         if (((vmx_msr_high >> 18) & 15) != 6)
1582                 return -EIO;
1583
1584         vmcs_conf->size = vmx_msr_high & 0x1fff;
1585         vmcs_conf->order = get_order(vmcs_config.size);
1586         vmcs_conf->revision_id = vmx_msr_low;
1587
1588         vmcs_conf->pin_based_exec_ctrl = _pin_based_exec_control;
1589         vmcs_conf->cpu_based_exec_ctrl = _cpu_based_exec_control;
1590         vmcs_conf->cpu_based_2nd_exec_ctrl = _cpu_based_2nd_exec_control;
1591         vmcs_conf->vmexit_ctrl         = _vmexit_control;
1592         vmcs_conf->vmentry_ctrl        = _vmentry_control;
1593
1594         cpu_has_load_ia32_efer =
1595                 allow_1_setting(MSR_IA32_VMX_ENTRY_CTLS,
1596                                 VM_ENTRY_LOAD_IA32_EFER)
1597                 && allow_1_setting(MSR_IA32_VMX_EXIT_CTLS,
1598                                    VM_EXIT_LOAD_IA32_EFER);
1599
1600         return 0;
1601 }
1602
1603 static struct vmcs *alloc_vmcs_cpu(int cpu)
1604 {
1605         int node = cpu_to_node(cpu);
1606         struct page *pages;
1607         struct vmcs *vmcs;
1608
1609         pages = alloc_pages_exact_node(node, GFP_KERNEL, vmcs_config.order);
1610         if (!pages)
1611                 return NULL;
1612         vmcs = page_address(pages);
1613         memset(vmcs, 0, vmcs_config.size);
1614         vmcs->revision_id = vmcs_config.revision_id; /* vmcs revision id */
1615         return vmcs;
1616 }
1617
1618 static struct vmcs *alloc_vmcs(void)
1619 {
1620         return alloc_vmcs_cpu(raw_smp_processor_id());
1621 }
1622
1623 static void free_vmcs(struct vmcs *vmcs)
1624 {
1625         free_pages((unsigned long)vmcs, vmcs_config.order);
1626 }
1627
1628 static void free_kvm_area(void)
1629 {
1630         int cpu;
1631
1632         for_each_possible_cpu(cpu) {
1633                 free_vmcs(per_cpu(vmxarea, cpu));
1634                 per_cpu(vmxarea, cpu) = NULL;
1635         }
1636 }
1637
1638 static __init int alloc_kvm_area(void)
1639 {
1640         int cpu;
1641
1642         for_each_possible_cpu(cpu) {
1643                 struct vmcs *vmcs;
1644
1645                 vmcs = alloc_vmcs_cpu(cpu);
1646                 if (!vmcs) {
1647                         free_kvm_area();
1648                         return -ENOMEM;
1649                 }
1650
1651                 per_cpu(vmxarea, cpu) = vmcs;
1652         }
1653         return 0;
1654 }
1655
1656 static __init int hardware_setup(void)
1657 {
1658         if (setup_vmcs_config(&vmcs_config) < 0)
1659                 return -EIO;
1660
1661         if (boot_cpu_has(X86_FEATURE_NX))
1662                 kvm_enable_efer_bits(EFER_NX);
1663
1664         if (!cpu_has_vmx_vpid())
1665                 enable_vpid = 0;
1666
1667         if (!cpu_has_vmx_ept() ||
1668             !cpu_has_vmx_ept_4levels()) {
1669                 enable_ept = 0;
1670                 enable_unrestricted_guest = 0;
1671         }
1672
1673         if (!cpu_has_vmx_unrestricted_guest())
1674                 enable_unrestricted_guest = 0;
1675
1676         if (!cpu_has_vmx_flexpriority())
1677                 flexpriority_enabled = 0;
1678
1679         if (!cpu_has_vmx_tpr_shadow())
1680                 kvm_x86_ops->update_cr8_intercept = NULL;
1681
1682         if (enable_ept && !cpu_has_vmx_ept_2m_page())
1683                 kvm_disable_largepages();
1684
1685         if (!cpu_has_vmx_ple())
1686                 ple_gap = 0;
1687
1688         return alloc_kvm_area();
1689 }
1690
1691 static __exit void hardware_unsetup(void)
1692 {
1693         free_kvm_area();
1694 }
1695
1696 static void fix_pmode_dataseg(int seg, struct kvm_save_segment *save)
1697 {
1698         struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1699
1700         if (vmcs_readl(sf->base) == save->base && (save->base & AR_S_MASK)) {
1701                 vmcs_write16(sf->selector, save->selector);
1702                 vmcs_writel(sf->base, save->base);
1703                 vmcs_write32(sf->limit, save->limit);
1704                 vmcs_write32(sf->ar_bytes, save->ar);
1705         } else {
1706                 u32 dpl = (vmcs_read16(sf->selector) & SELECTOR_RPL_MASK)
1707                         << AR_DPL_SHIFT;
1708                 vmcs_write32(sf->ar_bytes, 0x93 | dpl);
1709         }
1710 }
1711
1712 static void enter_pmode(struct kvm_vcpu *vcpu)
1713 {
1714         unsigned long flags;
1715         struct vcpu_vmx *vmx = to_vmx(vcpu);
1716
1717         vmx->emulation_required = 1;
1718         vmx->rmode.vm86_active = 0;
1719
1720         vmcs_write16(GUEST_TR_SELECTOR, vmx->rmode.tr.selector);
1721         vmcs_writel(GUEST_TR_BASE, vmx->rmode.tr.base);
1722         vmcs_write32(GUEST_TR_LIMIT, vmx->rmode.tr.limit);
1723         vmcs_write32(GUEST_TR_AR_BYTES, vmx->rmode.tr.ar);
1724
1725         flags = vmcs_readl(GUEST_RFLAGS);
1726         flags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
1727         flags |= vmx->rmode.save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
1728         vmcs_writel(GUEST_RFLAGS, flags);
1729
1730         vmcs_writel(GUEST_CR4, (vmcs_readl(GUEST_CR4) & ~X86_CR4_VME) |
1731                         (vmcs_readl(CR4_READ_SHADOW) & X86_CR4_VME));
1732
1733         update_exception_bitmap(vcpu);
1734
1735         if (emulate_invalid_guest_state)
1736                 return;
1737
1738         fix_pmode_dataseg(VCPU_SREG_ES, &vmx->rmode.es);
1739         fix_pmode_dataseg(VCPU_SREG_DS, &vmx->rmode.ds);
1740         fix_pmode_dataseg(VCPU_SREG_GS, &vmx->rmode.gs);
1741         fix_pmode_dataseg(VCPU_SREG_FS, &vmx->rmode.fs);
1742
1743         vmcs_write16(GUEST_SS_SELECTOR, 0);
1744         vmcs_write32(GUEST_SS_AR_BYTES, 0x93);
1745
1746         vmcs_write16(GUEST_CS_SELECTOR,
1747                      vmcs_read16(GUEST_CS_SELECTOR) & ~SELECTOR_RPL_MASK);
1748         vmcs_write32(GUEST_CS_AR_BYTES, 0x9b);
1749 }
1750
1751 static gva_t rmode_tss_base(struct kvm *kvm)
1752 {
1753         if (!kvm->arch.tss_addr) {
1754                 struct kvm_memslots *slots;
1755                 gfn_t base_gfn;
1756
1757                 slots = kvm_memslots(kvm);
1758                 base_gfn = slots->memslots[0].base_gfn +
1759                                  kvm->memslots->memslots[0].npages - 3;
1760                 return base_gfn << PAGE_SHIFT;
1761         }
1762         return kvm->arch.tss_addr;
1763 }
1764
1765 static void fix_rmode_seg(int seg, struct kvm_save_segment *save)
1766 {
1767         struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1768
1769         save->selector = vmcs_read16(sf->selector);
1770         save->base = vmcs_readl(sf->base);
1771         save->limit = vmcs_read32(sf->limit);
1772         save->ar = vmcs_read32(sf->ar_bytes);
1773         vmcs_write16(sf->selector, save->base >> 4);
1774         vmcs_write32(sf->base, save->base & 0xffff0);
1775         vmcs_write32(sf->limit, 0xffff);
1776         vmcs_write32(sf->ar_bytes, 0xf3);
1777         if (save->base & 0xf)
1778                 printk_once(KERN_WARNING "kvm: segment base is not paragraph"
1779                             " aligned when entering protected mode (seg=%d)",
1780                             seg);
1781 }
1782
1783 static void enter_rmode(struct kvm_vcpu *vcpu)
1784 {
1785         unsigned long flags;
1786         struct vcpu_vmx *vmx = to_vmx(vcpu);
1787
1788         if (enable_unrestricted_guest)
1789                 return;
1790
1791         vmx->emulation_required = 1;
1792         vmx->rmode.vm86_active = 1;
1793
1794         /*
1795          * Very old userspace does not call KVM_SET_TSS_ADDR before entering
1796          * vcpu. Call it here with phys address pointing 16M below 4G.
1797          */
1798         if (!vcpu->kvm->arch.tss_addr) {
1799                 printk_once(KERN_WARNING "kvm: KVM_SET_TSS_ADDR need to be "
1800                              "called before entering vcpu\n");
1801                 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
1802                 vmx_set_tss_addr(vcpu->kvm, 0xfeffd000);
1803                 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1804         }
1805
1806         vmx->rmode.tr.selector = vmcs_read16(GUEST_TR_SELECTOR);
1807         vmx->rmode.tr.base = vmcs_readl(GUEST_TR_BASE);
1808         vmcs_writel(GUEST_TR_BASE, rmode_tss_base(vcpu->kvm));
1809
1810         vmx->rmode.tr.limit = vmcs_read32(GUEST_TR_LIMIT);
1811         vmcs_write32(GUEST_TR_LIMIT, RMODE_TSS_SIZE - 1);
1812
1813         vmx->rmode.tr.ar = vmcs_read32(GUEST_TR_AR_BYTES);
1814         vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
1815
1816         flags = vmcs_readl(GUEST_RFLAGS);
1817         vmx->rmode.save_rflags = flags;
1818
1819         flags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
1820
1821         vmcs_writel(GUEST_RFLAGS, flags);
1822         vmcs_writel(GUEST_CR4, vmcs_readl(GUEST_CR4) | X86_CR4_VME);
1823         update_exception_bitmap(vcpu);
1824
1825         if (emulate_invalid_guest_state)
1826                 goto continue_rmode;
1827
1828         vmcs_write16(GUEST_SS_SELECTOR, vmcs_readl(GUEST_SS_BASE) >> 4);
1829         vmcs_write32(GUEST_SS_LIMIT, 0xffff);
1830         vmcs_write32(GUEST_SS_AR_BYTES, 0xf3);
1831
1832         vmcs_write32(GUEST_CS_AR_BYTES, 0xf3);
1833         vmcs_write32(GUEST_CS_LIMIT, 0xffff);
1834         if (vmcs_readl(GUEST_CS_BASE) == 0xffff0000)
1835                 vmcs_writel(GUEST_CS_BASE, 0xf0000);
1836         vmcs_write16(GUEST_CS_SELECTOR, vmcs_readl(GUEST_CS_BASE) >> 4);
1837
1838         fix_rmode_seg(VCPU_SREG_ES, &vmx->rmode.es);
1839         fix_rmode_seg(VCPU_SREG_DS, &vmx->rmode.ds);
1840         fix_rmode_seg(VCPU_SREG_GS, &vmx->rmode.gs);
1841         fix_rmode_seg(VCPU_SREG_FS, &vmx->rmode.fs);
1842
1843 continue_rmode:
1844         kvm_mmu_reset_context(vcpu);
1845 }
1846
1847 static void vmx_set_efer(struct kvm_vcpu *vcpu, u64 efer)
1848 {
1849         struct vcpu_vmx *vmx = to_vmx(vcpu);
1850         struct shared_msr_entry *msr = find_msr_entry(vmx, MSR_EFER);
1851
1852         if (!msr)
1853                 return;
1854
1855         /*
1856          * Force kernel_gs_base reloading before EFER changes, as control
1857          * of this msr depends on is_long_mode().
1858          */
1859         vmx_load_host_state(to_vmx(vcpu));
1860         vcpu->arch.efer = efer;
1861         if (efer & EFER_LMA) {
1862                 vmcs_write32(VM_ENTRY_CONTROLS,
1863                              vmcs_read32(VM_ENTRY_CONTROLS) |
1864                              VM_ENTRY_IA32E_MODE);
1865                 msr->data = efer;
1866         } else {
1867                 vmcs_write32(VM_ENTRY_CONTROLS,
1868                              vmcs_read32(VM_ENTRY_CONTROLS) &
1869                              ~VM_ENTRY_IA32E_MODE);
1870
1871                 msr->data = efer & ~EFER_LME;
1872         }
1873         setup_msrs(vmx);
1874 }
1875
1876 #ifdef CONFIG_X86_64
1877
1878 static void enter_lmode(struct kvm_vcpu *vcpu)
1879 {
1880         u32 guest_tr_ar;
1881
1882         guest_tr_ar = vmcs_read32(GUEST_TR_AR_BYTES);
1883         if ((guest_tr_ar & AR_TYPE_MASK) != AR_TYPE_BUSY_64_TSS) {
1884                 printk(KERN_DEBUG "%s: tss fixup for long mode. \n",
1885                        __func__);
1886                 vmcs_write32(GUEST_TR_AR_BYTES,
1887                              (guest_tr_ar & ~AR_TYPE_MASK)
1888                              | AR_TYPE_BUSY_64_TSS);
1889         }
1890         vmx_set_efer(vcpu, vcpu->arch.efer | EFER_LMA);
1891 }
1892
1893 static void exit_lmode(struct kvm_vcpu *vcpu)
1894 {
1895         vmcs_write32(VM_ENTRY_CONTROLS,
1896                      vmcs_read32(VM_ENTRY_CONTROLS)
1897                      & ~VM_ENTRY_IA32E_MODE);
1898         vmx_set_efer(vcpu, vcpu->arch.efer & ~EFER_LMA);
1899 }
1900
1901 #endif
1902
1903 static void vmx_flush_tlb(struct kvm_vcpu *vcpu)
1904 {
1905         vpid_sync_context(to_vmx(vcpu));
1906         if (enable_ept) {
1907                 if (!VALID_PAGE(vcpu->arch.mmu.root_hpa))
1908                         return;
1909                 ept_sync_context(construct_eptp(vcpu->arch.mmu.root_hpa));
1910         }
1911 }
1912
1913 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu)
1914 {
1915         ulong cr0_guest_owned_bits = vcpu->arch.cr0_guest_owned_bits;
1916
1917         vcpu->arch.cr0 &= ~cr0_guest_owned_bits;
1918         vcpu->arch.cr0 |= vmcs_readl(GUEST_CR0) & cr0_guest_owned_bits;
1919 }
1920
1921 static void vmx_decache_cr3(struct kvm_vcpu *vcpu)
1922 {
1923         if (enable_ept && is_paging(vcpu))
1924                 vcpu->arch.cr3 = vmcs_readl(GUEST_CR3);
1925         __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
1926 }
1927
1928 static void vmx_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
1929 {
1930         ulong cr4_guest_owned_bits = vcpu->arch.cr4_guest_owned_bits;
1931
1932         vcpu->arch.cr4 &= ~cr4_guest_owned_bits;
1933         vcpu->arch.cr4 |= vmcs_readl(GUEST_CR4) & cr4_guest_owned_bits;
1934 }
1935
1936 static void ept_load_pdptrs(struct kvm_vcpu *vcpu)
1937 {
1938         if (!test_bit(VCPU_EXREG_PDPTR,
1939                       (unsigned long *)&vcpu->arch.regs_dirty))
1940                 return;
1941
1942         if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
1943                 vmcs_write64(GUEST_PDPTR0, vcpu->arch.mmu.pdptrs[0]);
1944                 vmcs_write64(GUEST_PDPTR1, vcpu->arch.mmu.pdptrs[1]);
1945                 vmcs_write64(GUEST_PDPTR2, vcpu->arch.mmu.pdptrs[2]);
1946                 vmcs_write64(GUEST_PDPTR3, vcpu->arch.mmu.pdptrs[3]);
1947         }
1948 }
1949
1950 static void ept_save_pdptrs(struct kvm_vcpu *vcpu)
1951 {
1952         if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
1953                 vcpu->arch.mmu.pdptrs[0] = vmcs_read64(GUEST_PDPTR0);
1954                 vcpu->arch.mmu.pdptrs[1] = vmcs_read64(GUEST_PDPTR1);
1955                 vcpu->arch.mmu.pdptrs[2] = vmcs_read64(GUEST_PDPTR2);
1956                 vcpu->arch.mmu.pdptrs[3] = vmcs_read64(GUEST_PDPTR3);
1957         }
1958
1959         __set_bit(VCPU_EXREG_PDPTR,
1960                   (unsigned long *)&vcpu->arch.regs_avail);
1961         __set_bit(VCPU_EXREG_PDPTR,
1962                   (unsigned long *)&vcpu->arch.regs_dirty);
1963 }
1964
1965 static void vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4);
1966
1967 static void ept_update_paging_mode_cr0(unsigned long *hw_cr0,
1968                                         unsigned long cr0,
1969                                         struct kvm_vcpu *vcpu)
1970 {
1971         vmx_decache_cr3(vcpu);
1972         if (!(cr0 & X86_CR0_PG)) {
1973                 /* From paging/starting to nonpaging */
1974                 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
1975                              vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) |
1976                              (CPU_BASED_CR3_LOAD_EXITING |
1977                               CPU_BASED_CR3_STORE_EXITING));
1978                 vcpu->arch.cr0 = cr0;
1979                 vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
1980         } else if (!is_paging(vcpu)) {
1981                 /* From nonpaging to paging */
1982                 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
1983                              vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) &
1984                              ~(CPU_BASED_CR3_LOAD_EXITING |
1985                                CPU_BASED_CR3_STORE_EXITING));
1986                 vcpu->arch.cr0 = cr0;
1987                 vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
1988         }
1989
1990         if (!(cr0 & X86_CR0_WP))
1991                 *hw_cr0 &= ~X86_CR0_WP;
1992 }
1993
1994 static void vmx_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
1995 {
1996         struct vcpu_vmx *vmx = to_vmx(vcpu);
1997         unsigned long hw_cr0;
1998
1999         if (enable_unrestricted_guest)
2000                 hw_cr0 = (cr0 & ~KVM_GUEST_CR0_MASK_UNRESTRICTED_GUEST)
2001                         | KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST;
2002         else
2003                 hw_cr0 = (cr0 & ~KVM_GUEST_CR0_MASK) | KVM_VM_CR0_ALWAYS_ON;
2004
2005         if (vmx->rmode.vm86_active && (cr0 & X86_CR0_PE))
2006                 enter_pmode(vcpu);
2007
2008         if (!vmx->rmode.vm86_active && !(cr0 & X86_CR0_PE))
2009                 enter_rmode(vcpu);
2010
2011 #ifdef CONFIG_X86_64
2012         if (vcpu->arch.efer & EFER_LME) {
2013                 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG))
2014                         enter_lmode(vcpu);
2015                 if (is_paging(vcpu) && !(cr0 & X86_CR0_PG))
2016                         exit_lmode(vcpu);
2017         }
2018 #endif
2019
2020         if (enable_ept)
2021                 ept_update_paging_mode_cr0(&hw_cr0, cr0, vcpu);
2022
2023         if (!vcpu->fpu_active)
2024                 hw_cr0 |= X86_CR0_TS | X86_CR0_MP;
2025
2026         vmcs_writel(CR0_READ_SHADOW, cr0);
2027         vmcs_writel(GUEST_CR0, hw_cr0);
2028         vcpu->arch.cr0 = cr0;
2029         __clear_bit(VCPU_EXREG_CPL, (ulong *)&vcpu->arch.regs_avail);
2030 }
2031
2032 static u64 construct_eptp(unsigned long root_hpa)
2033 {
2034         u64 eptp;
2035
2036         /* TODO write the value reading from MSR */
2037         eptp = VMX_EPT_DEFAULT_MT |
2038                 VMX_EPT_DEFAULT_GAW << VMX_EPT_GAW_EPTP_SHIFT;
2039         eptp |= (root_hpa & PAGE_MASK);
2040
2041         return eptp;
2042 }
2043
2044 static void vmx_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
2045 {
2046         unsigned long guest_cr3;
2047         u64 eptp;
2048
2049         guest_cr3 = cr3;
2050         if (enable_ept) {
2051                 eptp = construct_eptp(cr3);
2052                 vmcs_write64(EPT_POINTER, eptp);
2053                 guest_cr3 = is_paging(vcpu) ? kvm_read_cr3(vcpu) :
2054                         vcpu->kvm->arch.ept_identity_map_addr;
2055                 ept_load_pdptrs(vcpu);
2056         }
2057
2058         vmx_flush_tlb(vcpu);
2059         vmcs_writel(GUEST_CR3, guest_cr3);
2060 }
2061
2062 static void vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
2063 {
2064         unsigned long hw_cr4 = cr4 | (to_vmx(vcpu)->rmode.vm86_active ?
2065                     KVM_RMODE_VM_CR4_ALWAYS_ON : KVM_PMODE_VM_CR4_ALWAYS_ON);
2066
2067         vcpu->arch.cr4 = cr4;
2068         if (enable_ept) {
2069                 if (!is_paging(vcpu)) {
2070                         hw_cr4 &= ~X86_CR4_PAE;
2071                         hw_cr4 |= X86_CR4_PSE;
2072                 } else if (!(cr4 & X86_CR4_PAE)) {
2073                         hw_cr4 &= ~X86_CR4_PAE;
2074                 }
2075         }
2076
2077         vmcs_writel(CR4_READ_SHADOW, cr4);
2078         vmcs_writel(GUEST_CR4, hw_cr4);
2079 }
2080
2081 static void vmx_get_segment(struct kvm_vcpu *vcpu,
2082                             struct kvm_segment *var, int seg)
2083 {
2084         struct vcpu_vmx *vmx = to_vmx(vcpu);
2085         struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
2086         struct kvm_save_segment *save;
2087         u32 ar;
2088
2089         if (vmx->rmode.vm86_active
2090             && (seg == VCPU_SREG_TR || seg == VCPU_SREG_ES
2091                 || seg == VCPU_SREG_DS || seg == VCPU_SREG_FS
2092                 || seg == VCPU_SREG_GS)
2093             && !emulate_invalid_guest_state) {
2094                 switch (seg) {
2095                 case VCPU_SREG_TR: save = &vmx->rmode.tr; break;
2096                 case VCPU_SREG_ES: save = &vmx->rmode.es; break;
2097                 case VCPU_SREG_DS: save = &vmx->rmode.ds; break;
2098                 case VCPU_SREG_FS: save = &vmx->rmode.fs; break;
2099                 case VCPU_SREG_GS: save = &vmx->rmode.gs; break;
2100                 default: BUG();
2101                 }
2102                 var->selector = save->selector;
2103                 var->base = save->base;
2104                 var->limit = save->limit;
2105                 ar = save->ar;
2106                 if (seg == VCPU_SREG_TR
2107                     || var->selector == vmcs_read16(sf->selector))
2108                         goto use_saved_rmode_seg;
2109         }
2110         var->base = vmcs_readl(sf->base);
2111         var->limit = vmcs_read32(sf->limit);
2112         var->selector = vmcs_read16(sf->selector);
2113         ar = vmcs_read32(sf->ar_bytes);
2114 use_saved_rmode_seg:
2115         if ((ar & AR_UNUSABLE_MASK) && !emulate_invalid_guest_state)
2116                 ar = 0;
2117         var->type = ar & 15;
2118         var->s = (ar >> 4) & 1;
2119         var->dpl = (ar >> 5) & 3;
2120         var->present = (ar >> 7) & 1;
2121         var->avl = (ar >> 12) & 1;
2122         var->l = (ar >> 13) & 1;
2123         var->db = (ar >> 14) & 1;
2124         var->g = (ar >> 15) & 1;
2125         var->unusable = (ar >> 16) & 1;
2126 }
2127
2128 static u64 vmx_get_segment_base(struct kvm_vcpu *vcpu, int seg)
2129 {
2130         struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
2131         struct kvm_segment s;
2132
2133         if (to_vmx(vcpu)->rmode.vm86_active) {
2134                 vmx_get_segment(vcpu, &s, seg);
2135                 return s.base;
2136         }
2137         return vmcs_readl(sf->base);
2138 }
2139
2140 static int __vmx_get_cpl(struct kvm_vcpu *vcpu)
2141 {
2142         if (!is_protmode(vcpu))
2143                 return 0;
2144
2145         if (!is_long_mode(vcpu)
2146             && (kvm_get_rflags(vcpu) & X86_EFLAGS_VM)) /* if virtual 8086 */
2147                 return 3;
2148
2149         return vmcs_read16(GUEST_CS_SELECTOR) & 3;
2150 }
2151
2152 static int vmx_get_cpl(struct kvm_vcpu *vcpu)
2153 {
2154         if (!test_bit(VCPU_EXREG_CPL, (ulong *)&vcpu->arch.regs_avail)) {
2155                 __set_bit(VCPU_EXREG_CPL, (ulong *)&vcpu->arch.regs_avail);
2156                 to_vmx(vcpu)->cpl = __vmx_get_cpl(vcpu);
2157         }
2158         return to_vmx(vcpu)->cpl;
2159 }
2160
2161
2162 static u32 vmx_segment_access_rights(struct kvm_segment *var)
2163 {
2164         u32 ar;
2165
2166         if (var->unusable)
2167                 ar = 1 << 16;
2168         else {
2169                 ar = var->type & 15;
2170                 ar |= (var->s & 1) << 4;
2171                 ar |= (var->dpl & 3) << 5;
2172                 ar |= (var->present & 1) << 7;
2173                 ar |= (var->avl & 1) << 12;
2174                 ar |= (var->l & 1) << 13;
2175                 ar |= (var->db & 1) << 14;
2176                 ar |= (var->g & 1) << 15;
2177         }
2178         if (ar == 0) /* a 0 value means unusable */
2179                 ar = AR_UNUSABLE_MASK;
2180
2181         return ar;
2182 }
2183
2184 static void vmx_set_segment(struct kvm_vcpu *vcpu,
2185                             struct kvm_segment *var, int seg)
2186 {
2187         struct vcpu_vmx *vmx = to_vmx(vcpu);
2188         struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
2189         u32 ar;
2190
2191         if (vmx->rmode.vm86_active && seg == VCPU_SREG_TR) {
2192                 vmcs_write16(sf->selector, var->selector);
2193                 vmx->rmode.tr.selector = var->selector;
2194                 vmx->rmode.tr.base = var->base;
2195                 vmx->rmode.tr.limit = var->limit;
2196                 vmx->rmode.tr.ar = vmx_segment_access_rights(var);
2197                 return;
2198         }
2199         vmcs_writel(sf->base, var->base);
2200         vmcs_write32(sf->limit, var->limit);
2201         vmcs_write16(sf->selector, var->selector);
2202         if (vmx->rmode.vm86_active && var->s) {
2203                 /*
2204                  * Hack real-mode segments into vm86 compatibility.
2205                  */
2206                 if (var->base == 0xffff0000 && var->selector == 0xf000)
2207                         vmcs_writel(sf->base, 0xf0000);
2208                 ar = 0xf3;
2209         } else
2210                 ar = vmx_segment_access_rights(var);
2211
2212         /*
2213          *   Fix the "Accessed" bit in AR field of segment registers for older
2214          * qemu binaries.
2215          *   IA32 arch specifies that at the time of processor reset the
2216          * "Accessed" bit in the AR field of segment registers is 1. And qemu
2217          * is setting it to 0 in the usedland code. This causes invalid guest
2218          * state vmexit when "unrestricted guest" mode is turned on.
2219          *    Fix for this setup issue in cpu_reset is being pushed in the qemu
2220          * tree. Newer qemu binaries with that qemu fix would not need this
2221          * kvm hack.
2222          */
2223         if (enable_unrestricted_guest && (seg != VCPU_SREG_LDTR))
2224                 ar |= 0x1; /* Accessed */
2225
2226         vmcs_write32(sf->ar_bytes, ar);
2227         __clear_bit(VCPU_EXREG_CPL, (ulong *)&vcpu->arch.regs_avail);
2228 }
2229
2230 static void vmx_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
2231 {
2232         u32 ar = vmcs_read32(GUEST_CS_AR_BYTES);
2233
2234         *db = (ar >> 14) & 1;
2235         *l = (ar >> 13) & 1;
2236 }
2237
2238 static void vmx_get_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
2239 {
2240         dt->size = vmcs_read32(GUEST_IDTR_LIMIT);
2241         dt->address = vmcs_readl(GUEST_IDTR_BASE);
2242 }
2243
2244 static void vmx_set_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
2245 {
2246         vmcs_write32(GUEST_IDTR_LIMIT, dt->size);
2247         vmcs_writel(GUEST_IDTR_BASE, dt->address);
2248 }
2249
2250 static void vmx_get_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
2251 {
2252         dt->size = vmcs_read32(GUEST_GDTR_LIMIT);
2253         dt->address = vmcs_readl(GUEST_GDTR_BASE);
2254 }
2255
2256 static void vmx_set_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
2257 {
2258         vmcs_write32(GUEST_GDTR_LIMIT, dt->size);
2259         vmcs_writel(GUEST_GDTR_BASE, dt->address);
2260 }
2261
2262 static bool rmode_segment_valid(struct kvm_vcpu *vcpu, int seg)
2263 {
2264         struct kvm_segment var;
2265         u32 ar;
2266
2267         vmx_get_segment(vcpu, &var, seg);
2268         ar = vmx_segment_access_rights(&var);
2269
2270         if (var.base != (var.selector << 4))
2271                 return false;
2272         if (var.limit != 0xffff)
2273                 return false;
2274         if (ar != 0xf3)
2275                 return false;
2276
2277         return true;
2278 }
2279
2280 static bool code_segment_valid(struct kvm_vcpu *vcpu)
2281 {
2282         struct kvm_segment cs;
2283         unsigned int cs_rpl;
2284
2285         vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
2286         cs_rpl = cs.selector & SELECTOR_RPL_MASK;
2287
2288         if (cs.unusable)
2289                 return false;
2290         if (~cs.type & (AR_TYPE_CODE_MASK|AR_TYPE_ACCESSES_MASK))
2291                 return false;
2292         if (!cs.s)
2293                 return false;
2294         if (cs.type & AR_TYPE_WRITEABLE_MASK) {
2295                 if (cs.dpl > cs_rpl)
2296                         return false;
2297         } else {
2298                 if (cs.dpl != cs_rpl)
2299                         return false;
2300         }
2301         if (!cs.present)
2302                 return false;
2303
2304         /* TODO: Add Reserved field check, this'll require a new member in the kvm_segment_field structure */
2305         return true;
2306 }
2307
2308 static bool stack_segment_valid(struct kvm_vcpu *vcpu)
2309 {
2310         struct kvm_segment ss;
2311         unsigned int ss_rpl;
2312
2313         vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
2314         ss_rpl = ss.selector & SELECTOR_RPL_MASK;
2315
2316         if (ss.unusable)
2317                 return true;
2318         if (ss.type != 3 && ss.type != 7)
2319                 return false;
2320         if (!ss.s)
2321                 return false;
2322         if (ss.dpl != ss_rpl) /* DPL != RPL */
2323                 return false;
2324         if (!ss.present)
2325                 return false;
2326
2327         return true;
2328 }
2329
2330 static bool data_segment_valid(struct kvm_vcpu *vcpu, int seg)
2331 {
2332         struct kvm_segment var;
2333         unsigned int rpl;
2334
2335         vmx_get_segment(vcpu, &var, seg);
2336         rpl = var.selector & SELECTOR_RPL_MASK;
2337
2338         if (var.unusable)
2339                 return true;
2340         if (!var.s)
2341                 return false;
2342         if (!var.present)
2343                 return false;
2344         if (~var.type & (AR_TYPE_CODE_MASK|AR_TYPE_WRITEABLE_MASK)) {
2345                 if (var.dpl < rpl) /* DPL < RPL */
2346                         return false;
2347         }
2348
2349         /* TODO: Add other members to kvm_segment_field to allow checking for other access
2350          * rights flags
2351          */
2352         return true;
2353 }
2354
2355 static bool tr_valid(struct kvm_vcpu *vcpu)
2356 {
2357         struct kvm_segment tr;
2358
2359         vmx_get_segment(vcpu, &tr, VCPU_SREG_TR);
2360
2361         if (tr.unusable)
2362                 return false;
2363         if (tr.selector & SELECTOR_TI_MASK)     /* TI = 1 */
2364                 return false;
2365         if (tr.type != 3 && tr.type != 11) /* TODO: Check if guest is in IA32e mode */
2366                 return false;
2367         if (!tr.present)
2368                 return false;
2369
2370         return true;
2371 }
2372
2373 static bool ldtr_valid(struct kvm_vcpu *vcpu)
2374 {
2375         struct kvm_segment ldtr;
2376
2377         vmx_get_segment(vcpu, &ldtr, VCPU_SREG_LDTR);
2378
2379         if (ldtr.unusable)
2380                 return true;
2381         if (ldtr.selector & SELECTOR_TI_MASK)   /* TI = 1 */
2382                 return false;
2383         if (ldtr.type != 2)
2384                 return false;
2385         if (!ldtr.present)
2386                 return false;
2387
2388         return true;
2389 }
2390
2391 static bool cs_ss_rpl_check(struct kvm_vcpu *vcpu)
2392 {
2393         struct kvm_segment cs, ss;
2394
2395         vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
2396         vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
2397
2398         return ((cs.selector & SELECTOR_RPL_MASK) ==
2399                  (ss.selector & SELECTOR_RPL_MASK));
2400 }
2401
2402 /*
2403  * Check if guest state is valid. Returns true if valid, false if
2404  * not.
2405  * We assume that registers are always usable
2406  */
2407 static bool guest_state_valid(struct kvm_vcpu *vcpu)
2408 {
2409         /* real mode guest state checks */
2410         if (!is_protmode(vcpu)) {
2411                 if (!rmode_segment_valid(vcpu, VCPU_SREG_CS))
2412                         return false;
2413                 if (!rmode_segment_valid(vcpu, VCPU_SREG_SS))
2414                         return false;
2415                 if (!rmode_segment_valid(vcpu, VCPU_SREG_DS))
2416                         return false;
2417                 if (!rmode_segment_valid(vcpu, VCPU_SREG_ES))
2418                         return false;
2419                 if (!rmode_segment_valid(vcpu, VCPU_SREG_FS))
2420                         return false;
2421                 if (!rmode_segment_valid(vcpu, VCPU_SREG_GS))
2422                         return false;
2423         } else {
2424         /* protected mode guest state checks */
2425                 if (!cs_ss_rpl_check(vcpu))
2426                         return false;
2427                 if (!code_segment_valid(vcpu))
2428                         return false;
2429                 if (!stack_segment_valid(vcpu))
2430                         return false;
2431                 if (!data_segment_valid(vcpu, VCPU_SREG_DS))
2432                         return false;
2433                 if (!data_segment_valid(vcpu, VCPU_SREG_ES))
2434                         return false;
2435                 if (!data_segment_valid(vcpu, VCPU_SREG_FS))
2436                         return false;
2437                 if (!data_segment_valid(vcpu, VCPU_SREG_GS))
2438                         return false;
2439                 if (!tr_valid(vcpu))
2440                         return false;
2441                 if (!ldtr_valid(vcpu))
2442                         return false;
2443         }
2444         /* TODO:
2445          * - Add checks on RIP
2446          * - Add checks on RFLAGS
2447          */
2448
2449         return true;
2450 }
2451
2452 static int init_rmode_tss(struct kvm *kvm)
2453 {
2454         gfn_t fn;
2455         u16 data = 0;
2456         int r, idx, ret = 0;
2457
2458         idx = srcu_read_lock(&kvm->srcu);
2459         fn = rmode_tss_base(kvm) >> PAGE_SHIFT;
2460         r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
2461         if (r < 0)
2462                 goto out;
2463         data = TSS_BASE_SIZE + TSS_REDIRECTION_SIZE;
2464         r = kvm_write_guest_page(kvm, fn++, &data,
2465                         TSS_IOPB_BASE_OFFSET, sizeof(u16));
2466         if (r < 0)
2467                 goto out;
2468         r = kvm_clear_guest_page(kvm, fn++, 0, PAGE_SIZE);
2469         if (r < 0)
2470                 goto out;
2471         r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
2472         if (r < 0)
2473                 goto out;
2474         data = ~0;
2475         r = kvm_write_guest_page(kvm, fn, &data,
2476                                  RMODE_TSS_SIZE - 2 * PAGE_SIZE - 1,
2477                                  sizeof(u8));
2478         if (r < 0)
2479                 goto out;
2480
2481         ret = 1;
2482 out:
2483         srcu_read_unlock(&kvm->srcu, idx);
2484         return ret;
2485 }
2486
2487 static int init_rmode_identity_map(struct kvm *kvm)
2488 {
2489         int i, idx, r, ret;
2490         pfn_t identity_map_pfn;
2491         u32 tmp;
2492
2493         if (!enable_ept)
2494                 return 1;
2495         if (unlikely(!kvm->arch.ept_identity_pagetable)) {
2496                 printk(KERN_ERR "EPT: identity-mapping pagetable "
2497                         "haven't been allocated!\n");
2498                 return 0;
2499         }
2500         if (likely(kvm->arch.ept_identity_pagetable_done))
2501                 return 1;
2502         ret = 0;
2503         identity_map_pfn = kvm->arch.ept_identity_map_addr >> PAGE_SHIFT;
2504         idx = srcu_read_lock(&kvm->srcu);
2505         r = kvm_clear_guest_page(kvm, identity_map_pfn, 0, PAGE_SIZE);
2506         if (r < 0)
2507                 goto out;
2508         /* Set up identity-mapping pagetable for EPT in real mode */
2509         for (i = 0; i < PT32_ENT_PER_PAGE; i++) {
2510                 tmp = (i << 22) + (_PAGE_PRESENT | _PAGE_RW | _PAGE_USER |
2511                         _PAGE_ACCESSED | _PAGE_DIRTY | _PAGE_PSE);
2512                 r = kvm_write_guest_page(kvm, identity_map_pfn,
2513                                 &tmp, i * sizeof(tmp), sizeof(tmp));
2514                 if (r < 0)
2515                         goto out;
2516         }
2517         kvm->arch.ept_identity_pagetable_done = true;
2518         ret = 1;
2519 out:
2520         srcu_read_unlock(&kvm->srcu, idx);
2521         return ret;
2522 }
2523
2524 static void seg_setup(int seg)
2525 {
2526         struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
2527         unsigned int ar;
2528
2529         vmcs_write16(sf->selector, 0);
2530         vmcs_writel(sf->base, 0);
2531         vmcs_write32(sf->limit, 0xffff);
2532         if (enable_unrestricted_guest) {
2533                 ar = 0x93;
2534                 if (seg == VCPU_SREG_CS)
2535                         ar |= 0x08; /* code segment */
2536         } else
2537                 ar = 0xf3;
2538
2539         vmcs_write32(sf->ar_bytes, ar);
2540 }
2541
2542 static int alloc_apic_access_page(struct kvm *kvm)
2543 {
2544         struct kvm_userspace_memory_region kvm_userspace_mem;
2545         int r = 0;
2546
2547         mutex_lock(&kvm->slots_lock);
2548         if (kvm->arch.apic_access_page)
2549                 goto out;
2550         kvm_userspace_mem.slot = APIC_ACCESS_PAGE_PRIVATE_MEMSLOT;
2551         kvm_userspace_mem.flags = 0;
2552         kvm_userspace_mem.guest_phys_addr = 0xfee00000ULL;
2553         kvm_userspace_mem.memory_size = PAGE_SIZE;
2554         r = __kvm_set_memory_region(kvm, &kvm_userspace_mem, 0);
2555         if (r)
2556                 goto out;
2557
2558         kvm->arch.apic_access_page = gfn_to_page(kvm, 0xfee00);
2559 out:
2560         mutex_unlock(&kvm->slots_lock);
2561         return r;
2562 }
2563
2564 static int alloc_identity_pagetable(struct kvm *kvm)
2565 {
2566         struct kvm_userspace_memory_region kvm_userspace_mem;
2567         int r = 0;
2568
2569         mutex_lock(&kvm->slots_lock);
2570         if (kvm->arch.ept_identity_pagetable)
2571                 goto out;
2572         kvm_userspace_mem.slot = IDENTITY_PAGETABLE_PRIVATE_MEMSLOT;
2573         kvm_userspace_mem.flags = 0;
2574         kvm_userspace_mem.guest_phys_addr =
2575                 kvm->arch.ept_identity_map_addr;
2576         kvm_userspace_mem.memory_size = PAGE_SIZE;
2577         r = __kvm_set_memory_region(kvm, &kvm_userspace_mem, 0);
2578         if (r)
2579                 goto out;
2580
2581         kvm->arch.ept_identity_pagetable = gfn_to_page(kvm,
2582                         kvm->arch.ept_identity_map_addr >> PAGE_SHIFT);
2583 out:
2584         mutex_unlock(&kvm->slots_lock);
2585         return r;
2586 }
2587
2588 static void allocate_vpid(struct vcpu_vmx *vmx)
2589 {
2590         int vpid;
2591
2592         vmx->vpid = 0;
2593         if (!enable_vpid)
2594                 return;
2595         spin_lock(&vmx_vpid_lock);
2596         vpid = find_first_zero_bit(vmx_vpid_bitmap, VMX_NR_VPIDS);
2597         if (vpid < VMX_NR_VPIDS) {
2598                 vmx->vpid = vpid;
2599                 __set_bit(vpid, vmx_vpid_bitmap);
2600         }
2601         spin_unlock(&vmx_vpid_lock);
2602 }
2603
2604 static void free_vpid(struct vcpu_vmx *vmx)
2605 {
2606         if (!enable_vpid)
2607                 return;
2608         spin_lock(&vmx_vpid_lock);
2609         if (vmx->vpid != 0)
2610                 __clear_bit(vmx->vpid, vmx_vpid_bitmap);
2611         spin_unlock(&vmx_vpid_lock);
2612 }
2613
2614 static void __vmx_disable_intercept_for_msr(unsigned long *msr_bitmap, u32 msr)
2615 {
2616         int f = sizeof(unsigned long);
2617
2618         if (!cpu_has_vmx_msr_bitmap())
2619                 return;
2620
2621         /*
2622          * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
2623          * have the write-low and read-high bitmap offsets the wrong way round.
2624          * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
2625          */
2626         if (msr <= 0x1fff) {
2627                 __clear_bit(msr, msr_bitmap + 0x000 / f); /* read-low */
2628                 __clear_bit(msr, msr_bitmap + 0x800 / f); /* write-low */
2629         } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
2630                 msr &= 0x1fff;
2631                 __clear_bit(msr, msr_bitmap + 0x400 / f); /* read-high */
2632                 __clear_bit(msr, msr_bitmap + 0xc00 / f); /* write-high */
2633         }
2634 }
2635
2636 static void vmx_disable_intercept_for_msr(u32 msr, bool longmode_only)
2637 {
2638         if (!longmode_only)
2639                 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_legacy, msr);
2640         __vmx_disable_intercept_for_msr(vmx_msr_bitmap_longmode, msr);
2641 }
2642
2643 /*
2644  * Sets up the vmcs for emulated real mode.
2645  */
2646 static int vmx_vcpu_setup(struct vcpu_vmx *vmx)
2647 {
2648         u32 host_sysenter_cs, msr_low, msr_high;
2649         u32 junk;
2650         u64 host_pat;
2651         unsigned long a;
2652         struct desc_ptr dt;
2653         int i;
2654         unsigned long kvm_vmx_return;
2655         u32 exec_control;
2656
2657         /* I/O */
2658         vmcs_write64(IO_BITMAP_A, __pa(vmx_io_bitmap_a));
2659         vmcs_write64(IO_BITMAP_B, __pa(vmx_io_bitmap_b));
2660
2661         if (cpu_has_vmx_msr_bitmap())
2662                 vmcs_write64(MSR_BITMAP, __pa(vmx_msr_bitmap_legacy));
2663
2664         vmcs_write64(VMCS_LINK_POINTER, -1ull); /* 22.3.1.5 */
2665
2666         /* Control */
2667         vmcs_write32(PIN_BASED_VM_EXEC_CONTROL,
2668                 vmcs_config.pin_based_exec_ctrl);
2669
2670         exec_control = vmcs_config.cpu_based_exec_ctrl;
2671         if (!vm_need_tpr_shadow(vmx->vcpu.kvm)) {
2672                 exec_control &= ~CPU_BASED_TPR_SHADOW;
2673 #ifdef CONFIG_X86_64
2674                 exec_control |= CPU_BASED_CR8_STORE_EXITING |
2675                                 CPU_BASED_CR8_LOAD_EXITING;
2676 #endif
2677         }
2678         if (!enable_ept)
2679                 exec_control |= CPU_BASED_CR3_STORE_EXITING |
2680                                 CPU_BASED_CR3_LOAD_EXITING  |
2681                                 CPU_BASED_INVLPG_EXITING;
2682         vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, exec_control);
2683
2684         if (cpu_has_secondary_exec_ctrls()) {
2685                 exec_control = vmcs_config.cpu_based_2nd_exec_ctrl;
2686                 if (!vm_need_virtualize_apic_accesses(vmx->vcpu.kvm))
2687                         exec_control &=
2688                                 ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
2689                 if (vmx->vpid == 0)
2690                         exec_control &= ~SECONDARY_EXEC_ENABLE_VPID;
2691                 if (!enable_ept) {
2692                         exec_control &= ~SECONDARY_EXEC_ENABLE_EPT;
2693                         enable_unrestricted_guest = 0;
2694                 }
2695                 if (!enable_unrestricted_guest)
2696                         exec_control &= ~SECONDARY_EXEC_UNRESTRICTED_GUEST;
2697                 if (!ple_gap)
2698                         exec_control &= ~SECONDARY_EXEC_PAUSE_LOOP_EXITING;
2699                 vmcs_write32(SECONDARY_VM_EXEC_CONTROL, exec_control);
2700         }
2701
2702         if (ple_gap) {
2703                 vmcs_write32(PLE_GAP, ple_gap);
2704                 vmcs_write32(PLE_WINDOW, ple_window);
2705         }
2706
2707         vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, !!bypass_guest_pf);
2708         vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, !!bypass_guest_pf);
2709         vmcs_write32(CR3_TARGET_COUNT, 0);           /* 22.2.1 */
2710
2711         vmcs_writel(HOST_CR0, read_cr0() | X86_CR0_TS);  /* 22.2.3 */
2712         vmcs_writel(HOST_CR4, read_cr4());  /* 22.2.3, 22.2.5 */
2713         vmcs_writel(HOST_CR3, read_cr3());  /* 22.2.3  FIXME: shadow tables */
2714
2715         vmcs_write16(HOST_CS_SELECTOR, __KERNEL_CS);  /* 22.2.4 */
2716         vmcs_write16(HOST_DS_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
2717         vmcs_write16(HOST_ES_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
2718         vmcs_write16(HOST_FS_SELECTOR, 0);            /* 22.2.4 */
2719         vmcs_write16(HOST_GS_SELECTOR, 0);            /* 22.2.4 */
2720         vmcs_write16(HOST_SS_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
2721 #ifdef CONFIG_X86_64
2722         rdmsrl(MSR_FS_BASE, a);
2723         vmcs_writel(HOST_FS_BASE, a); /* 22.2.4 */
2724         rdmsrl(MSR_GS_BASE, a);
2725         vmcs_writel(HOST_GS_BASE, a); /* 22.2.4 */
2726 #else
2727         vmcs_writel(HOST_FS_BASE, 0); /* 22.2.4 */
2728         vmcs_writel(HOST_GS_BASE, 0); /* 22.2.4 */
2729 #endif
2730
2731         vmcs_write16(HOST_TR_SELECTOR, GDT_ENTRY_TSS*8);  /* 22.2.4 */
2732
2733         native_store_idt(&dt);
2734         vmcs_writel(HOST_IDTR_BASE, dt.address);   /* 22.2.4 */
2735
2736         asm("mov $.Lkvm_vmx_return, %0" : "=r"(kvm_vmx_return));
2737         vmcs_writel(HOST_RIP, kvm_vmx_return); /* 22.2.5 */
2738         vmcs_write32(VM_EXIT_MSR_STORE_COUNT, 0);
2739         vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, 0);
2740         vmcs_write64(VM_EXIT_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.host));
2741         vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, 0);
2742         vmcs_write64(VM_ENTRY_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.guest));
2743
2744         rdmsr(MSR_IA32_SYSENTER_CS, host_sysenter_cs, junk);
2745         vmcs_write32(HOST_IA32_SYSENTER_CS, host_sysenter_cs);
2746         rdmsrl(MSR_IA32_SYSENTER_ESP, a);
2747         vmcs_writel(HOST_IA32_SYSENTER_ESP, a);   /* 22.2.3 */
2748         rdmsrl(MSR_IA32_SYSENTER_EIP, a);
2749         vmcs_writel(HOST_IA32_SYSENTER_EIP, a);   /* 22.2.3 */
2750
2751         if (vmcs_config.vmexit_ctrl & VM_EXIT_LOAD_IA32_PAT) {
2752                 rdmsr(MSR_IA32_CR_PAT, msr_low, msr_high);
2753                 host_pat = msr_low | ((u64) msr_high << 32);
2754                 vmcs_write64(HOST_IA32_PAT, host_pat);
2755         }
2756         if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
2757                 rdmsr(MSR_IA32_CR_PAT, msr_low, msr_high);
2758                 host_pat = msr_low | ((u64) msr_high << 32);
2759                 /* Write the default value follow host pat */
2760                 vmcs_write64(GUEST_IA32_PAT, host_pat);
2761                 /* Keep arch.pat sync with GUEST_IA32_PAT */
2762                 vmx->vcpu.arch.pat = host_pat;
2763         }
2764
2765         for (i = 0; i < NR_VMX_MSR; ++i) {
2766                 u32 index = vmx_msr_index[i];
2767                 u32 data_low, data_high;
2768                 int j = vmx->nmsrs;
2769
2770                 if (rdmsr_safe(index, &data_low, &data_high) < 0)
2771                         continue;
2772                 if (wrmsr_safe(index, data_low, data_high) < 0)
2773                         continue;
2774                 vmx->guest_msrs[j].index = i;
2775                 vmx->guest_msrs[j].data = 0;
2776                 vmx->guest_msrs[j].mask = -1ull;
2777                 ++vmx->nmsrs;
2778         }
2779
2780         vmcs_write32(VM_EXIT_CONTROLS, vmcs_config.vmexit_ctrl);
2781
2782         /* 22.2.1, 20.8.1 */
2783         vmcs_write32(VM_ENTRY_CONTROLS, vmcs_config.vmentry_ctrl);
2784
2785         vmcs_writel(CR0_GUEST_HOST_MASK, ~0UL);
2786         vmx->vcpu.arch.cr4_guest_owned_bits = KVM_CR4_GUEST_OWNED_BITS;
2787         if (enable_ept)
2788                 vmx->vcpu.arch.cr4_guest_owned_bits |= X86_CR4_PGE;
2789         vmcs_writel(CR4_GUEST_HOST_MASK, ~vmx->vcpu.arch.cr4_guest_owned_bits);
2790
2791         kvm_write_tsc(&vmx->vcpu, 0);
2792
2793         return 0;
2794 }
2795
2796 static int vmx_vcpu_reset(struct kvm_vcpu *vcpu)
2797 {
2798         struct vcpu_vmx *vmx = to_vmx(vcpu);
2799         u64 msr;
2800         int ret;
2801
2802         vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP));
2803
2804         vmx->rmode.vm86_active = 0;
2805
2806         vmx->soft_vnmi_blocked = 0;
2807
2808         vmx->vcpu.arch.regs[VCPU_REGS_RDX] = get_rdx_init_val();
2809         kvm_set_cr8(&vmx->vcpu, 0);
2810         msr = 0xfee00000 | MSR_IA32_APICBASE_ENABLE;
2811         if (kvm_vcpu_is_bsp(&vmx->vcpu))
2812                 msr |= MSR_IA32_APICBASE_BSP;
2813         kvm_set_apic_base(&vmx->vcpu, msr);
2814
2815         ret = fx_init(&vmx->vcpu);
2816         if (ret != 0)
2817                 goto out;
2818
2819         seg_setup(VCPU_SREG_CS);
2820         /*
2821          * GUEST_CS_BASE should really be 0xffff0000, but VT vm86 mode
2822          * insists on having GUEST_CS_BASE == GUEST_CS_SELECTOR << 4.  Sigh.
2823          */
2824         if (kvm_vcpu_is_bsp(&vmx->vcpu)) {
2825                 vmcs_write16(GUEST_CS_SELECTOR, 0xf000);
2826                 vmcs_writel(GUEST_CS_BASE, 0x000f0000);
2827         } else {
2828                 vmcs_write16(GUEST_CS_SELECTOR, vmx->vcpu.arch.sipi_vector << 8);
2829                 vmcs_writel(GUEST_CS_BASE, vmx->vcpu.arch.sipi_vector << 12);
2830         }
2831
2832         seg_setup(VCPU_SREG_DS);
2833         seg_setup(VCPU_SREG_ES);
2834         seg_setup(VCPU_SREG_FS);
2835         seg_setup(VCPU_SREG_GS);
2836         seg_setup(VCPU_SREG_SS);
2837
2838         vmcs_write16(GUEST_TR_SELECTOR, 0);
2839         vmcs_writel(GUEST_TR_BASE, 0);
2840         vmcs_write32(GUEST_TR_LIMIT, 0xffff);
2841         vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
2842
2843         vmcs_write16(GUEST_LDTR_SELECTOR, 0);
2844         vmcs_writel(GUEST_LDTR_BASE, 0);
2845         vmcs_write32(GUEST_LDTR_LIMIT, 0xffff);
2846         vmcs_write32(GUEST_LDTR_AR_BYTES, 0x00082);
2847
2848         vmcs_write32(GUEST_SYSENTER_CS, 0);
2849         vmcs_writel(GUEST_SYSENTER_ESP, 0);
2850         vmcs_writel(GUEST_SYSENTER_EIP, 0);
2851
2852         vmcs_writel(GUEST_RFLAGS, 0x02);
2853         if (kvm_vcpu_is_bsp(&vmx->vcpu))
2854                 kvm_rip_write(vcpu, 0xfff0);
2855         else
2856                 kvm_rip_write(vcpu, 0);
2857         kvm_register_write(vcpu, VCPU_REGS_RSP, 0);
2858
2859         vmcs_writel(GUEST_DR7, 0x400);
2860
2861         vmcs_writel(GUEST_GDTR_BASE, 0);
2862         vmcs_write32(GUEST_GDTR_LIMIT, 0xffff);
2863
2864         vmcs_writel(GUEST_IDTR_BASE, 0);
2865         vmcs_write32(GUEST_IDTR_LIMIT, 0xffff);
2866
2867         vmcs_write32(GUEST_ACTIVITY_STATE, GUEST_ACTIVITY_ACTIVE);
2868         vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, 0);
2869         vmcs_write32(GUEST_PENDING_DBG_EXCEPTIONS, 0);
2870
2871         /* Special registers */
2872         vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
2873
2874         setup_msrs(vmx);
2875
2876         vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);  /* 22.2.1 */
2877
2878         if (cpu_has_vmx_tpr_shadow()) {
2879                 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, 0);
2880                 if (vm_need_tpr_shadow(vmx->vcpu.kvm))
2881                         vmcs_write64(VIRTUAL_APIC_PAGE_ADDR,
2882                                      __pa(vmx->vcpu.arch.apic->regs));
2883                 vmcs_write32(TPR_THRESHOLD, 0);
2884         }
2885
2886         if (vm_need_virtualize_apic_accesses(vmx->vcpu.kvm))
2887                 vmcs_write64(APIC_ACCESS_ADDR,
2888                              page_to_phys(vmx->vcpu.kvm->arch.apic_access_page));
2889
2890         if (vmx->vpid != 0)
2891                 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
2892
2893         vmx->vcpu.arch.cr0 = X86_CR0_NW | X86_CR0_CD | X86_CR0_ET;
2894         vmx_set_cr0(&vmx->vcpu, kvm_read_cr0(vcpu)); /* enter rmode */
2895         vmx_set_cr4(&vmx->vcpu, 0);
2896         vmx_set_efer(&vmx->vcpu, 0);
2897         vmx_fpu_activate(&vmx->vcpu);
2898         update_exception_bitmap(&vmx->vcpu);
2899
2900         vpid_sync_context(vmx);
2901
2902         ret = 0;
2903
2904         /* HACK: Don't enable emulation on guest boot/reset */
2905         vmx->emulation_required = 0;
2906
2907 out:
2908         return ret;
2909 }
2910
2911 static void enable_irq_window(struct kvm_vcpu *vcpu)
2912 {
2913         u32 cpu_based_vm_exec_control;
2914
2915         cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
2916         cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_INTR_PENDING;
2917         vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
2918 }
2919
2920 static void enable_nmi_window(struct kvm_vcpu *vcpu)
2921 {
2922         u32 cpu_based_vm_exec_control;
2923
2924         if (!cpu_has_virtual_nmis()) {
2925                 enable_irq_window(vcpu);
2926                 return;
2927         }
2928
2929         if (vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_STI) {
2930                 enable_irq_window(vcpu);
2931                 return;
2932         }
2933         cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
2934         cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_NMI_PENDING;
2935         vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
2936 }
2937
2938 static void vmx_inject_irq(struct kvm_vcpu *vcpu)
2939 {
2940         struct vcpu_vmx *vmx = to_vmx(vcpu);
2941         uint32_t intr;
2942         int irq = vcpu->arch.interrupt.nr;
2943
2944         trace_kvm_inj_virq(irq);
2945
2946         ++vcpu->stat.irq_injections;
2947         if (vmx->rmode.vm86_active) {
2948                 int inc_eip = 0;
2949                 if (vcpu->arch.interrupt.soft)
2950                         inc_eip = vcpu->arch.event_exit_inst_len;
2951                 if (kvm_inject_realmode_interrupt(vcpu, irq, inc_eip) != EMULATE_DONE)
2952                         kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
2953                 return;
2954         }
2955         intr = irq | INTR_INFO_VALID_MASK;
2956         if (vcpu->arch.interrupt.soft) {
2957                 intr |= INTR_TYPE_SOFT_INTR;
2958                 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
2959                              vmx->vcpu.arch.event_exit_inst_len);
2960         } else
2961                 intr |= INTR_TYPE_EXT_INTR;
2962         vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr);
2963         vmx_clear_hlt(vcpu);
2964 }
2965
2966 static void vmx_inject_nmi(struct kvm_vcpu *vcpu)
2967 {
2968         struct vcpu_vmx *vmx = to_vmx(vcpu);
2969
2970         if (!cpu_has_virtual_nmis()) {
2971                 /*
2972                  * Tracking the NMI-blocked state in software is built upon
2973                  * finding the next open IRQ window. This, in turn, depends on
2974                  * well-behaving guests: They have to keep IRQs disabled at
2975                  * least as long as the NMI handler runs. Otherwise we may
2976                  * cause NMI nesting, maybe breaking the guest. But as this is
2977                  * highly unlikely, we can live with the residual risk.
2978                  */
2979                 vmx->soft_vnmi_blocked = 1;
2980                 vmx->vnmi_blocked_time = 0;
2981         }
2982
2983         ++vcpu->stat.nmi_injections;
2984         vmx->nmi_known_unmasked = false;
2985         if (vmx->rmode.vm86_active) {
2986                 if (kvm_inject_realmode_interrupt(vcpu, NMI_VECTOR, 0) != EMULATE_DONE)
2987                         kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
2988                 return;
2989         }
2990         vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
2991                         INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR);
2992         vmx_clear_hlt(vcpu);
2993 }
2994
2995 static int vmx_nmi_allowed(struct kvm_vcpu *vcpu)
2996 {
2997         if (!cpu_has_virtual_nmis() && to_vmx(vcpu)->soft_vnmi_blocked)
2998                 return 0;
2999
3000         return  !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
3001                   (GUEST_INTR_STATE_MOV_SS | GUEST_INTR_STATE_STI
3002                    | GUEST_INTR_STATE_NMI));
3003 }
3004
3005 static bool vmx_get_nmi_mask(struct kvm_vcpu *vcpu)
3006 {
3007         if (!cpu_has_virtual_nmis())
3008                 return to_vmx(vcpu)->soft_vnmi_blocked;
3009         if (to_vmx(vcpu)->nmi_known_unmasked)
3010                 return false;
3011         return vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_NMI;
3012 }
3013
3014 static void vmx_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked)
3015 {
3016         struct vcpu_vmx *vmx = to_vmx(vcpu);
3017
3018         if (!cpu_has_virtual_nmis()) {
3019                 if (vmx->soft_vnmi_blocked != masked) {
3020                         vmx->soft_vnmi_blocked = masked;
3021                         vmx->vnmi_blocked_time = 0;
3022                 }
3023         } else {
3024                 vmx->nmi_known_unmasked = !masked;
3025                 if (masked)
3026                         vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
3027                                       GUEST_INTR_STATE_NMI);
3028                 else
3029                         vmcs_clear_bits(GUEST_INTERRUPTIBILITY_INFO,
3030                                         GUEST_INTR_STATE_NMI);
3031         }
3032 }
3033
3034 static int vmx_interrupt_allowed(struct kvm_vcpu *vcpu)
3035 {
3036         return (vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_IF) &&
3037                 !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
3038                         (GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS));
3039 }
3040
3041 static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr)
3042 {
3043         int ret;
3044         struct kvm_userspace_memory_region tss_mem = {
3045                 .slot = TSS_PRIVATE_MEMSLOT,
3046                 .guest_phys_addr = addr,
3047                 .memory_size = PAGE_SIZE * 3,
3048                 .flags = 0,
3049         };
3050
3051         ret = kvm_set_memory_region(kvm, &tss_mem, 0);
3052         if (ret)
3053                 return ret;
3054         kvm->arch.tss_addr = addr;
3055         if (!init_rmode_tss(kvm))
3056                 return  -ENOMEM;
3057
3058         return 0;
3059 }
3060
3061 static int handle_rmode_exception(struct kvm_vcpu *vcpu,
3062                                   int vec, u32 err_code)
3063 {
3064         /*
3065          * Instruction with address size override prefix opcode 0x67
3066          * Cause the #SS fault with 0 error code in VM86 mode.
3067          */
3068         if (((vec == GP_VECTOR) || (vec == SS_VECTOR)) && err_code == 0)
3069                 if (emulate_instruction(vcpu, 0) == EMULATE_DONE)
3070                         return 1;
3071         /*
3072          * Forward all other exceptions that are valid in real mode.
3073          * FIXME: Breaks guest debugging in real mode, needs to be fixed with
3074          *        the required debugging infrastructure rework.
3075          */
3076         switch (vec) {
3077         case DB_VECTOR:
3078                 if (vcpu->guest_debug &
3079                     (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
3080                         return 0;
3081                 kvm_queue_exception(vcpu, vec);
3082                 return 1;
3083         case BP_VECTOR:
3084                 /*
3085                  * Update instruction length as we may reinject the exception
3086                  * from user space while in guest debugging mode.
3087                  */
3088                 to_vmx(vcpu)->vcpu.arch.event_exit_inst_len =
3089                         vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
3090                 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
3091                         return 0;
3092                 /* fall through */
3093         case DE_VECTOR:
3094         case OF_VECTOR:
3095         case BR_VECTOR:
3096         case UD_VECTOR:
3097         case DF_VECTOR:
3098         case SS_VECTOR:
3099         case GP_VECTOR:
3100         case MF_VECTOR:
3101                 kvm_queue_exception(vcpu, vec);
3102                 return 1;
3103         }
3104         return 0;
3105 }
3106
3107 /*
3108  * Trigger machine check on the host. We assume all the MSRs are already set up
3109  * by the CPU and that we still run on the same CPU as the MCE occurred on.
3110  * We pass a fake environment to the machine check handler because we want
3111  * the guest to be always treated like user space, no matter what context
3112  * it used internally.
3113  */
3114 static void kvm_machine_check(void)
3115 {
3116 #if defined(CONFIG_X86_MCE) && defined(CONFIG_X86_64)
3117         struct pt_regs regs = {
3118                 .cs = 3, /* Fake ring 3 no matter what the guest ran on */
3119                 .flags = X86_EFLAGS_IF,
3120         };
3121
3122         do_machine_check(&regs, 0);
3123 #endif
3124 }
3125
3126 static int handle_machine_check(struct kvm_vcpu *vcpu)
3127 {
3128         /* already handled by vcpu_run */
3129         return 1;
3130 }
3131
3132 static int handle_exception(struct kvm_vcpu *vcpu)
3133 {
3134         struct vcpu_vmx *vmx = to_vmx(vcpu);
3135         struct kvm_run *kvm_run = vcpu->run;
3136         u32 intr_info, ex_no, error_code;
3137         unsigned long cr2, rip, dr6;
3138         u32 vect_info;
3139         enum emulation_result er;
3140
3141         vect_info = vmx->idt_vectoring_info;
3142         intr_info = vmx->exit_intr_info;
3143
3144         if (is_machine_check(intr_info))
3145                 return handle_machine_check(vcpu);
3146
3147         if ((vect_info & VECTORING_INFO_VALID_MASK) &&
3148             !is_page_fault(intr_info)) {
3149                 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
3150                 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_SIMUL_EX;
3151                 vcpu->run->internal.ndata = 2;
3152                 vcpu->run->internal.data[0] = vect_info;
3153                 vcpu->run->internal.data[1] = intr_info;
3154                 return 0;
3155         }
3156
3157         if ((intr_info & INTR_INFO_INTR_TYPE_MASK) == INTR_TYPE_NMI_INTR)
3158                 return 1;  /* already handled by vmx_vcpu_run() */
3159
3160         if (is_no_device(intr_info)) {
3161                 vmx_fpu_activate(vcpu);
3162                 return 1;
3163         }
3164
3165         if (is_invalid_opcode(intr_info)) {
3166                 er = emulate_instruction(vcpu, EMULTYPE_TRAP_UD);
3167                 if (er != EMULATE_DONE)
3168                         kvm_queue_exception(vcpu, UD_VECTOR);
3169                 return 1;
3170         }
3171
3172         error_code = 0;
3173         rip = kvm_rip_read(vcpu);
3174         if (intr_info & INTR_INFO_DELIVER_CODE_MASK)
3175                 error_code = vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
3176         if (is_page_fault(intr_info)) {
3177                 /* EPT won't cause page fault directly */
3178                 if (enable_ept)
3179                         BUG();
3180                 cr2 = vmcs_readl(EXIT_QUALIFICATION);
3181                 trace_kvm_page_fault(cr2, error_code);
3182
3183                 if (kvm_event_needs_reinjection(vcpu))
3184                         kvm_mmu_unprotect_page_virt(vcpu, cr2);
3185                 return kvm_mmu_page_fault(vcpu, cr2, error_code, NULL, 0);
3186         }
3187
3188         if (vmx->rmode.vm86_active &&
3189             handle_rmode_exception(vcpu, intr_info & INTR_INFO_VECTOR_MASK,
3190                                                                 error_code)) {
3191                 if (vcpu->arch.halt_request) {
3192                         vcpu->arch.halt_request = 0;
3193                         return kvm_emulate_halt(vcpu);
3194                 }
3195                 return 1;
3196         }
3197
3198         ex_no = intr_info & INTR_INFO_VECTOR_MASK;
3199         switch (ex_no) {
3200         case DB_VECTOR:
3201                 dr6 = vmcs_readl(EXIT_QUALIFICATION);
3202                 if (!(vcpu->guest_debug &
3203                       (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))) {
3204                         vcpu->arch.dr6 = dr6 | DR6_FIXED_1;
3205                         kvm_queue_exception(vcpu, DB_VECTOR);
3206                         return 1;
3207                 }
3208                 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1;
3209                 kvm_run->debug.arch.dr7 = vmcs_readl(GUEST_DR7);
3210                 /* fall through */
3211         case BP_VECTOR:
3212                 /*
3213                  * Update instruction length as we may reinject #BP from
3214                  * user space while in guest debugging mode. Reading it for
3215                  * #DB as well causes no harm, it is not used in that case.
3216                  */
3217                 vmx->vcpu.arch.event_exit_inst_len =
3218                         vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
3219                 kvm_run->exit_reason = KVM_EXIT_DEBUG;
3220                 kvm_run->debug.arch.pc = vmcs_readl(GUEST_CS_BASE) + rip;
3221                 kvm_run->debug.arch.exception = ex_no;
3222                 break;
3223         default:
3224                 kvm_run->exit_reason = KVM_EXIT_EXCEPTION;
3225                 kvm_run->ex.exception = ex_no;
3226                 kvm_run->ex.error_code = error_code;
3227                 break;
3228         }
3229         return 0;
3230 }
3231
3232 static int handle_external_interrupt(struct kvm_vcpu *vcpu)
3233 {
3234         ++vcpu->stat.irq_exits;
3235         return 1;
3236 }
3237
3238 static int handle_triple_fault(struct kvm_vcpu *vcpu)
3239 {
3240         vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
3241         return 0;
3242 }
3243
3244 static int handle_io(struct kvm_vcpu *vcpu)
3245 {
3246         unsigned long exit_qualification;
3247         int size, in, string;
3248         unsigned port;
3249
3250         exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
3251         string = (exit_qualification & 16) != 0;
3252         in = (exit_qualification & 8) != 0;
3253
3254         ++vcpu->stat.io_exits;
3255
3256         if (string || in)
3257                 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
3258
3259         port = exit_qualification >> 16;
3260         size = (exit_qualification & 7) + 1;
3261         skip_emulated_instruction(vcpu);
3262
3263         return kvm_fast_pio_out(vcpu, size, port);
3264 }
3265
3266 static void
3267 vmx_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
3268 {
3269         /*
3270          * Patch in the VMCALL instruction:
3271          */
3272         hypercall[0] = 0x0f;
3273         hypercall[1] = 0x01;
3274         hypercall[2] = 0xc1;
3275 }
3276
3277 static int handle_cr(struct kvm_vcpu *vcpu)
3278 {
3279         unsigned long exit_qualification, val;
3280         int cr;
3281         int reg;
3282         int err;
3283
3284         exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
3285         cr = exit_qualification & 15;
3286         reg = (exit_qualification >> 8) & 15;
3287         switch ((exit_qualification >> 4) & 3) {
3288         case 0: /* mov to cr */
3289                 val = kvm_register_read(vcpu, reg);
3290                 trace_kvm_cr_write(cr, val);
3291                 switch (cr) {
3292                 case 0:
3293                         err = kvm_set_cr0(vcpu, val);
3294                         kvm_complete_insn_gp(vcpu, err);
3295                         return 1;
3296                 case 3:
3297                         err = kvm_set_cr3(vcpu, val);
3298                         kvm_complete_insn_gp(vcpu, err);
3299                         return 1;
3300                 case 4:
3301                         err = kvm_set_cr4(vcpu, val);
3302                         kvm_complete_insn_gp(vcpu, err);
3303                         return 1;
3304                 case 8: {
3305                                 u8 cr8_prev = kvm_get_cr8(vcpu);
3306                                 u8 cr8 = kvm_register_read(vcpu, reg);
3307                                 err = kvm_set_cr8(vcpu, cr8);
3308                                 kvm_complete_insn_gp(vcpu, err);
3309                                 if (irqchip_in_kernel(vcpu->kvm))
3310                                         return 1;
3311                                 if (cr8_prev <= cr8)
3312                                         return 1;
3313                                 vcpu->run->exit_reason = KVM_EXIT_SET_TPR;
3314                                 return 0;
3315                         }
3316                 };
3317                 break;
3318         case 2: /* clts */
3319                 vmx_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~X86_CR0_TS));
3320                 trace_kvm_cr_write(0, kvm_read_cr0(vcpu));
3321                 skip_emulated_instruction(vcpu);
3322                 vmx_fpu_activate(vcpu);
3323                 return 1;
3324         case 1: /*mov from cr*/
3325                 switch (cr) {
3326                 case 3:
3327                         val = kvm_read_cr3(vcpu);
3328                         kvm_register_write(vcpu, reg, val);
3329                         trace_kvm_cr_read(cr, val);
3330                         skip_emulated_instruction(vcpu);
3331                         return 1;
3332                 case 8:
3333                         val = kvm_get_cr8(vcpu);
3334                         kvm_register_write(vcpu, reg, val);
3335                         trace_kvm_cr_read(cr, val);
3336                         skip_emulated_instruction(vcpu);
3337                         return 1;
3338                 }
3339                 break;
3340         case 3: /* lmsw */
3341                 val = (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f;
3342                 trace_kvm_cr_write(0, (kvm_read_cr0(vcpu) & ~0xful) | val);
3343                 kvm_lmsw(vcpu, val);
3344
3345                 skip_emulated_instruction(vcpu);
3346                 return 1;
3347         default:
3348                 break;
3349         }
3350         vcpu->run->exit_reason = 0;
3351         pr_unimpl(vcpu, "unhandled control register: op %d cr %d\n",
3352                (int)(exit_qualification >> 4) & 3, cr);
3353         return 0;
3354 }
3355
3356 static int handle_dr(struct kvm_vcpu *vcpu)
3357 {
3358         unsigned long exit_qualification;
3359         int dr, reg;
3360
3361         /* Do not handle if the CPL > 0, will trigger GP on re-entry */
3362         if (!kvm_require_cpl(vcpu, 0))
3363                 return 1;
3364         dr = vmcs_readl(GUEST_DR7);
3365         if (dr & DR7_GD) {
3366                 /*
3367                  * As the vm-exit takes precedence over the debug trap, we
3368                  * need to emulate the latter, either for the host or the
3369                  * guest debugging itself.
3370                  */
3371                 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
3372                         vcpu->run->debug.arch.dr6 = vcpu->arch.dr6;
3373                         vcpu->run->debug.arch.dr7 = dr;
3374                         vcpu->run->debug.arch.pc =
3375                                 vmcs_readl(GUEST_CS_BASE) +
3376                                 vmcs_readl(GUEST_RIP);
3377                         vcpu->run->debug.arch.exception = DB_VECTOR;
3378                         vcpu->run->exit_reason = KVM_EXIT_DEBUG;
3379                         return 0;
3380                 } else {
3381                         vcpu->arch.dr7 &= ~DR7_GD;
3382                         vcpu->arch.dr6 |= DR6_BD;
3383                         vmcs_writel(GUEST_DR7, vcpu->arch.dr7);
3384                         kvm_queue_exception(vcpu, DB_VECTOR);
3385                         return 1;
3386                 }
3387         }
3388
3389         exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
3390         dr = exit_qualification & DEBUG_REG_ACCESS_NUM;
3391         reg = DEBUG_REG_ACCESS_REG(exit_qualification);
3392         if (exit_qualification & TYPE_MOV_FROM_DR) {
3393                 unsigned long val;
3394                 if (!kvm_get_dr(vcpu, dr, &val))
3395                         kvm_register_write(vcpu, reg, val);
3396         } else
3397                 kvm_set_dr(vcpu, dr, vcpu->arch.regs[reg]);
3398         skip_emulated_instruction(vcpu);
3399         return 1;
3400 }
3401
3402 static void vmx_set_dr7(struct kvm_vcpu *vcpu, unsigned long val)
3403 {
3404         vmcs_writel(GUEST_DR7, val);
3405 }
3406
3407 static int handle_cpuid(struct kvm_vcpu *vcpu)
3408 {
3409         kvm_emulate_cpuid(vcpu);
3410         return 1;
3411 }
3412
3413 static int handle_rdmsr(struct kvm_vcpu *vcpu)
3414 {
3415         u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
3416         u64 data;
3417
3418         if (vmx_get_msr(vcpu, ecx, &data)) {
3419                 trace_kvm_msr_read_ex(ecx);
3420                 kvm_inject_gp(vcpu, 0);
3421                 return 1;
3422         }
3423
3424         trace_kvm_msr_read(ecx, data);
3425
3426         /* FIXME: handling of bits 32:63 of rax, rdx */
3427         vcpu->arch.regs[VCPU_REGS_RAX] = data & -1u;
3428         vcpu->arch.regs[VCPU_REGS_RDX] = (data >> 32) & -1u;
3429         skip_emulated_instruction(vcpu);
3430         return 1;
3431 }
3432
3433 static int handle_wrmsr(struct kvm_vcpu *vcpu)
3434 {
3435         u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
3436         u64 data = (vcpu->arch.regs[VCPU_REGS_RAX] & -1u)
3437                 | ((u64)(vcpu->arch.regs[VCPU_REGS_RDX] & -1u) << 32);
3438
3439         if (vmx_set_msr(vcpu, ecx, data) != 0) {
3440                 trace_kvm_msr_write_ex(ecx, data);
3441                 kvm_inject_gp(vcpu, 0);
3442                 return 1;
3443         }
3444
3445         trace_kvm_msr_write(ecx, data);
3446         skip_emulated_instruction(vcpu);
3447         return 1;
3448 }
3449
3450 static int handle_tpr_below_threshold(struct kvm_vcpu *vcpu)
3451 {
3452         kvm_make_request(KVM_REQ_EVENT, vcpu);
3453         return 1;
3454 }
3455
3456 static int handle_interrupt_window(struct kvm_vcpu *vcpu)
3457 {
3458         u32 cpu_based_vm_exec_control;
3459
3460         /* clear pending irq */
3461         cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
3462         cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_INTR_PENDING;
3463         vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
3464
3465         kvm_make_request(KVM_REQ_EVENT, vcpu);
3466
3467         ++vcpu->stat.irq_window_exits;
3468
3469         /*
3470          * If the user space waits to inject interrupts, exit as soon as
3471          * possible
3472          */
3473         if (!irqchip_in_kernel(vcpu->kvm) &&
3474             vcpu->run->request_interrupt_window &&
3475             !kvm_cpu_has_interrupt(vcpu)) {
3476                 vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
3477                 return 0;
3478         }
3479         return 1;
3480 }
3481
3482 static int handle_halt(struct kvm_vcpu *vcpu)
3483 {
3484         skip_emulated_instruction(vcpu);
3485         return kvm_emulate_halt(vcpu);
3486 }
3487
3488 static int handle_vmcall(struct kvm_vcpu *vcpu)
3489 {
3490         skip_emulated_instruction(vcpu);
3491         kvm_emulate_hypercall(vcpu);
3492         return 1;
3493 }
3494
3495 static int handle_vmx_insn(struct kvm_vcpu *vcpu)
3496 {
3497         kvm_queue_exception(vcpu, UD_VECTOR);
3498         return 1;
3499 }
3500
3501 static int handle_invd(struct kvm_vcpu *vcpu)
3502 {
3503         return emulate_instruction(vcpu, 0) == EMULATE_DONE;
3504 }
3505
3506 static int handle_invlpg(struct kvm_vcpu *vcpu)
3507 {
3508         unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
3509
3510         kvm_mmu_invlpg(vcpu, exit_qualification);
3511         skip_emulated_instruction(vcpu);
3512         return 1;
3513 }
3514
3515 static int handle_wbinvd(struct kvm_vcpu *vcpu)
3516 {
3517         skip_emulated_instruction(vcpu);
3518         kvm_emulate_wbinvd(vcpu);
3519         return 1;
3520 }
3521
3522 static int handle_xsetbv(struct kvm_vcpu *vcpu)
3523 {
3524         u64 new_bv = kvm_read_edx_eax(vcpu);
3525         u32 index = kvm_register_read(vcpu, VCPU_REGS_RCX);
3526
3527         if (kvm_set_xcr(vcpu, index, new_bv) == 0)
3528                 skip_emulated_instruction(vcpu);
3529         return 1;
3530 }
3531
3532 static int handle_apic_access(struct kvm_vcpu *vcpu)
3533 {
3534         return emulate_instruction(vcpu, 0) == EMULATE_DONE;
3535 }
3536
3537 static int handle_task_switch(struct kvm_vcpu *vcpu)
3538 {
3539         struct vcpu_vmx *vmx = to_vmx(vcpu);
3540         unsigned long exit_qualification;
3541         bool has_error_code = false;
3542         u32 error_code = 0;
3543         u16 tss_selector;
3544         int reason, type, idt_v;
3545
3546         idt_v = (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK);
3547         type = (vmx->idt_vectoring_info & VECTORING_INFO_TYPE_MASK);
3548
3549         exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
3550
3551         reason = (u32)exit_qualification >> 30;
3552         if (reason == TASK_SWITCH_GATE && idt_v) {
3553                 switch (type) {
3554                 case INTR_TYPE_NMI_INTR:
3555                         vcpu->arch.nmi_injected = false;
3556                         vmx_set_nmi_mask(vcpu, true);
3557                         break;
3558                 case INTR_TYPE_EXT_INTR:
3559                 case INTR_TYPE_SOFT_INTR:
3560                         kvm_clear_interrupt_queue(vcpu);
3561                         break;
3562                 case INTR_TYPE_HARD_EXCEPTION:
3563                         if (vmx->idt_vectoring_info &
3564                             VECTORING_INFO_DELIVER_CODE_MASK) {
3565                                 has_error_code = true;
3566                                 error_code =
3567                                         vmcs_read32(IDT_VECTORING_ERROR_CODE);
3568                         }
3569                         /* fall through */
3570                 case INTR_TYPE_SOFT_EXCEPTION:
3571                         kvm_clear_exception_queue(vcpu);
3572                         break;
3573                 default:
3574                         break;
3575                 }
3576         }
3577         tss_selector = exit_qualification;
3578
3579         if (!idt_v || (type != INTR_TYPE_HARD_EXCEPTION &&
3580                        type != INTR_TYPE_EXT_INTR &&
3581                        type != INTR_TYPE_NMI_INTR))
3582                 skip_emulated_instruction(vcpu);
3583
3584         if (kvm_task_switch(vcpu, tss_selector, reason,
3585                                 has_error_code, error_code) == EMULATE_FAIL) {
3586                 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
3587                 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
3588                 vcpu->run->internal.ndata = 0;
3589                 return 0;
3590         }
3591
3592         /* clear all local breakpoint enable flags */
3593         vmcs_writel(GUEST_DR7, vmcs_readl(GUEST_DR7) & ~55);
3594
3595         /*
3596          * TODO: What about debug traps on tss switch?
3597          *       Are we supposed to inject them and update dr6?
3598          */
3599
3600         return 1;
3601 }
3602
3603 static int handle_ept_violation(struct kvm_vcpu *vcpu)
3604 {
3605         unsigned long exit_qualification;
3606         gpa_t gpa;
3607         int gla_validity;
3608
3609         exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
3610
3611         if (exit_qualification & (1 << 6)) {
3612                 printk(KERN_ERR "EPT: GPA exceeds GAW!\n");
3613                 return -EINVAL;
3614         }
3615
3616         gla_validity = (exit_qualification >> 7) & 0x3;
3617         if (gla_validity != 0x3 && gla_validity != 0x1 && gla_validity != 0) {
3618                 printk(KERN_ERR "EPT: Handling EPT violation failed!\n");
3619                 printk(KERN_ERR "EPT: GPA: 0x%lx, GVA: 0x%lx\n",
3620                         (long unsigned int)vmcs_read64(GUEST_PHYSICAL_ADDRESS),
3621                         vmcs_readl(GUEST_LINEAR_ADDRESS));
3622                 printk(KERN_ERR "EPT: Exit qualification is 0x%lx\n",
3623                         (long unsigned int)exit_qualification);
3624                 vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
3625                 vcpu->run->hw.hardware_exit_reason = EXIT_REASON_EPT_VIOLATION;
3626                 return 0;
3627         }
3628
3629         gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
3630         trace_kvm_page_fault(gpa, exit_qualification);
3631         return kvm_mmu_page_fault(vcpu, gpa, exit_qualification & 0x3, NULL, 0);
3632 }
3633
3634 static u64 ept_rsvd_mask(u64 spte, int level)
3635 {
3636         int i;
3637         u64 mask = 0;
3638
3639         for (i = 51; i > boot_cpu_data.x86_phys_bits; i--)
3640                 mask |= (1ULL << i);
3641
3642         if (level > 2)
3643                 /* bits 7:3 reserved */
3644                 mask |= 0xf8;
3645         else if (level == 2) {
3646                 if (spte & (1ULL << 7))
3647                         /* 2MB ref, bits 20:12 reserved */
3648                         mask |= 0x1ff000;
3649                 else
3650                         /* bits 6:3 reserved */
3651                         mask |= 0x78;
3652         }
3653
3654         return mask;
3655 }
3656
3657 static void ept_misconfig_inspect_spte(struct kvm_vcpu *vcpu, u64 spte,
3658                                        int level)
3659 {
3660         printk(KERN_ERR "%s: spte 0x%llx level %d\n", __func__, spte, level);
3661
3662         /* 010b (write-only) */
3663         WARN_ON((spte & 0x7) == 0x2);
3664
3665         /* 110b (write/execute) */
3666         WARN_ON((spte & 0x7) == 0x6);
3667
3668         /* 100b (execute-only) and value not supported by logical processor */
3669         if (!cpu_has_vmx_ept_execute_only())
3670                 WARN_ON((spte & 0x7) == 0x4);
3671
3672         /* not 000b */
3673         if ((spte & 0x7)) {
3674                 u64 rsvd_bits = spte & ept_rsvd_mask(spte, level);
3675
3676                 if (rsvd_bits != 0) {
3677                         printk(KERN_ERR "%s: rsvd_bits = 0x%llx\n",
3678                                          __func__, rsvd_bits);
3679                         WARN_ON(1);
3680                 }
3681
3682                 if (level == 1 || (level == 2 && (spte & (1ULL << 7)))) {
3683                         u64 ept_mem_type = (spte & 0x38) >> 3;
3684
3685                         if (ept_mem_type == 2 || ept_mem_type == 3 ||
3686                             ept_mem_type == 7) {
3687                                 printk(KERN_ERR "%s: ept_mem_type=0x%llx\n",
3688                                                 __func__, ept_mem_type);
3689                                 WARN_ON(1);
3690                         }
3691                 }
3692         }
3693 }
3694
3695 static int handle_ept_misconfig(struct kvm_vcpu *vcpu)
3696 {
3697         u64 sptes[4];
3698         int nr_sptes, i;
3699         gpa_t gpa;
3700
3701         gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
3702
3703         printk(KERN_ERR "EPT: Misconfiguration.\n");
3704         printk(KERN_ERR "EPT: GPA: 0x%llx\n", gpa);
3705
3706         nr_sptes = kvm_mmu_get_spte_hierarchy(vcpu, gpa, sptes);
3707
3708         for (i = PT64_ROOT_LEVEL; i > PT64_ROOT_LEVEL - nr_sptes; --i)
3709                 ept_misconfig_inspect_spte(vcpu, sptes[i-1], i);
3710
3711         vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
3712         vcpu->run->hw.hardware_exit_reason = EXIT_REASON_EPT_MISCONFIG;
3713
3714         return 0;
3715 }
3716
3717 static int handle_nmi_window(struct kvm_vcpu *vcpu)
3718 {
3719         u32 cpu_based_vm_exec_control;
3720
3721         /* clear pending NMI */
3722         cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
3723         cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_NMI_PENDING;
3724         vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
3725         ++vcpu->stat.nmi_window_exits;
3726         kvm_make_request(KVM_REQ_EVENT, vcpu);
3727
3728         return 1;
3729 }
3730
3731 static int handle_invalid_guest_state(struct kvm_vcpu *vcpu)
3732 {
3733         struct vcpu_vmx *vmx = to_vmx(vcpu);
3734         enum emulation_result err = EMULATE_DONE;
3735         int ret = 1;
3736         u32 cpu_exec_ctrl;
3737         bool intr_window_requested;
3738
3739         cpu_exec_ctrl = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
3740         intr_window_requested = cpu_exec_ctrl & CPU_BASED_VIRTUAL_INTR_PENDING;
3741
3742         while (!guest_state_valid(vcpu)) {
3743                 if (intr_window_requested
3744                     && (kvm_get_rflags(&vmx->vcpu) & X86_EFLAGS_IF))
3745                         return handle_interrupt_window(&vmx->vcpu);
3746
3747                 err = emulate_instruction(vcpu, 0);
3748
3749                 if (err == EMULATE_DO_MMIO) {
3750                         ret = 0;
3751                         goto out;
3752                 }
3753
3754                 if (err != EMULATE_DONE)
3755                         return 0;
3756
3757                 if (signal_pending(current))
3758                         goto out;
3759                 if (need_resched())
3760                         schedule();
3761         }
3762
3763         vmx->emulation_required = 0;
3764 out:
3765         return ret;
3766 }
3767
3768 /*
3769  * Indicate a busy-waiting vcpu in spinlock. We do not enable the PAUSE
3770  * exiting, so only get here on cpu with PAUSE-Loop-Exiting.
3771  */
3772 static int handle_pause(struct kvm_vcpu *vcpu)
3773 {
3774         skip_emulated_instruction(vcpu);
3775         kvm_vcpu_on_spin(vcpu);
3776
3777         return 1;
3778 }
3779
3780 static int handle_invalid_op(struct kvm_vcpu *vcpu)
3781 {
3782         kvm_queue_exception(vcpu, UD_VECTOR);
3783         return 1;
3784 }
3785
3786 /*
3787  * The exit handlers return 1 if the exit was handled fully and guest execution
3788  * may resume.  Otherwise they set the kvm_run parameter to indicate what needs
3789  * to be done to userspace and return 0.
3790  */
3791 static int (*kvm_vmx_exit_handlers[])(struct kvm_vcpu *vcpu) = {
3792         [EXIT_REASON_EXCEPTION_NMI]           = handle_exception,
3793         [EXIT_REASON_EXTERNAL_INTERRUPT]      = handle_external_interrupt,
3794         [EXIT_REASON_TRIPLE_FAULT]            = handle_triple_fault,
3795         [EXIT_REASON_NMI_WINDOW]              = handle_nmi_window,
3796         [EXIT_REASON_IO_INSTRUCTION]          = handle_io,
3797         [EXIT_REASON_CR_ACCESS]               = handle_cr,
3798         [EXIT_REASON_DR_ACCESS]               = handle_dr,
3799         [EXIT_REASON_CPUID]                   = handle_cpuid,
3800         [EXIT_REASON_MSR_READ]                = handle_rdmsr,
3801         [EXIT_REASON_MSR_WRITE]               = handle_wrmsr,
3802         [EXIT_REASON_PENDING_INTERRUPT]       = handle_interrupt_window,
3803         [EXIT_REASON_HLT]                     = handle_halt,
3804         [EXIT_REASON_INVD]                    = handle_invd,
3805         [EXIT_REASON_INVLPG]                  = handle_invlpg,
3806         [EXIT_REASON_VMCALL]                  = handle_vmcall,
3807         [EXIT_REASON_VMCLEAR]                 = handle_vmx_insn,
3808         [EXIT_REASON_VMLAUNCH]                = handle_vmx_insn,
3809         [EXIT_REASON_VMPTRLD]                 = handle_vmx_insn,
3810         [EXIT_REASON_VMPTRST]                 = handle_vmx_insn,
3811         [EXIT_REASON_VMREAD]                  = handle_vmx_insn,
3812         [EXIT_REASON_VMRESUME]                = handle_vmx_insn,
3813         [EXIT_REASON_VMWRITE]                 = handle_vmx_insn,
3814         [EXIT_REASON_VMOFF]                   = handle_vmx_insn,
3815         [EXIT_REASON_VMON]                    = handle_vmx_insn,
3816         [EXIT_REASON_TPR_BELOW_THRESHOLD]     = handle_tpr_below_threshold,
3817         [EXIT_REASON_APIC_ACCESS]             = handle_apic_access,
3818         [EXIT_REASON_WBINVD]                  = handle_wbinvd,
3819         [EXIT_REASON_XSETBV]                  = handle_xsetbv,
3820         [EXIT_REASON_TASK_SWITCH]             = handle_task_switch,
3821         [EXIT_REASON_MCE_DURING_VMENTRY]      = handle_machine_check,
3822         [EXIT_REASON_EPT_VIOLATION]           = handle_ept_violation,
3823         [EXIT_REASON_EPT_MISCONFIG]           = handle_ept_misconfig,
3824         [EXIT_REASON_PAUSE_INSTRUCTION]       = handle_pause,
3825         [EXIT_REASON_MWAIT_INSTRUCTION]       = handle_invalid_op,
3826         [EXIT_REASON_MONITOR_INSTRUCTION]     = handle_invalid_op,
3827 };
3828
3829 static const int kvm_vmx_max_exit_handlers =
3830         ARRAY_SIZE(kvm_vmx_exit_handlers);
3831
3832 static void vmx_get_exit_info(struct kvm_vcpu *vcpu, u64 *info1, u64 *info2)
3833 {
3834         *info1 = vmcs_readl(EXIT_QUALIFICATION);
3835         *info2 = vmcs_read32(VM_EXIT_INTR_INFO);
3836 }
3837
3838 /*
3839  * The guest has exited.  See if we can fix it or if we need userspace
3840  * assistance.
3841  */
3842 static int vmx_handle_exit(struct kvm_vcpu *vcpu)
3843 {
3844         struct vcpu_vmx *vmx = to_vmx(vcpu);
3845         u32 exit_reason = vmx->exit_reason;
3846         u32 vectoring_info = vmx->idt_vectoring_info;
3847
3848         trace_kvm_exit(exit_reason, vcpu, KVM_ISA_VMX);
3849
3850         /* If guest state is invalid, start emulating */
3851         if (vmx->emulation_required && emulate_invalid_guest_state)
3852                 return handle_invalid_guest_state(vcpu);
3853
3854         if (exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY) {
3855                 vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
3856                 vcpu->run->fail_entry.hardware_entry_failure_reason
3857                         = exit_reason;
3858                 return 0;
3859         }
3860
3861         if (unlikely(vmx->fail)) {
3862                 vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
3863                 vcpu->run->fail_entry.hardware_entry_failure_reason
3864                         = vmcs_read32(VM_INSTRUCTION_ERROR);
3865                 return 0;
3866         }
3867
3868         if ((vectoring_info & VECTORING_INFO_VALID_MASK) &&
3869                         (exit_reason != EXIT_REASON_EXCEPTION_NMI &&
3870                         exit_reason != EXIT_REASON_EPT_VIOLATION &&
3871                         exit_reason != EXIT_REASON_TASK_SWITCH))
3872                 printk(KERN_WARNING "%s: unexpected, valid vectoring info "
3873                        "(0x%x) and exit reason is 0x%x\n",
3874                        __func__, vectoring_info, exit_reason);
3875
3876         if (unlikely(!cpu_has_virtual_nmis() && vmx->soft_vnmi_blocked)) {
3877                 if (vmx_interrupt_allowed(vcpu)) {
3878                         vmx->soft_vnmi_blocked = 0;
3879                 } else if (vmx->vnmi_blocked_time > 1000000000LL &&
3880                            vcpu->arch.nmi_pending) {
3881                         /*
3882                          * This CPU don't support us in finding the end of an
3883                          * NMI-blocked window if the guest runs with IRQs
3884                          * disabled. So we pull the trigger after 1 s of
3885                          * futile waiting, but inform the user about this.
3886                          */
3887                         printk(KERN_WARNING "%s: Breaking out of NMI-blocked "
3888                                "state on VCPU %d after 1 s timeout\n",
3889                                __func__, vcpu->vcpu_id);
3890                         vmx->soft_vnmi_blocked = 0;
3891                 }
3892         }
3893
3894         if (exit_reason < kvm_vmx_max_exit_handlers
3895             && kvm_vmx_exit_handlers[exit_reason])
3896                 return kvm_vmx_exit_handlers[exit_reason](vcpu);
3897         else {
3898                 vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
3899                 vcpu->run->hw.hardware_exit_reason = exit_reason;
3900         }
3901         return 0;
3902 }
3903
3904 static void update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
3905 {
3906         if (irr == -1 || tpr < irr) {
3907                 vmcs_write32(TPR_THRESHOLD, 0);
3908                 return;
3909         }
3910
3911         vmcs_write32(TPR_THRESHOLD, irr);
3912 }
3913
3914 static void vmx_complete_atomic_exit(struct vcpu_vmx *vmx)
3915 {
3916         u32 exit_intr_info;
3917
3918         if (!(vmx->exit_reason == EXIT_REASON_MCE_DURING_VMENTRY
3919               || vmx->exit_reason == EXIT_REASON_EXCEPTION_NMI))
3920                 return;
3921
3922         vmx->exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
3923         exit_intr_info = vmx->exit_intr_info;
3924
3925         /* Handle machine checks before interrupts are enabled */
3926         if (is_machine_check(exit_intr_info))
3927                 kvm_machine_check();
3928
3929         /* We need to handle NMIs before interrupts are enabled */
3930         if ((exit_intr_info & INTR_INFO_INTR_TYPE_MASK) == INTR_TYPE_NMI_INTR &&
3931             (exit_intr_info & INTR_INFO_VALID_MASK)) {
3932                 kvm_before_handle_nmi(&vmx->vcpu);
3933                 asm("int $2");
3934                 kvm_after_handle_nmi(&vmx->vcpu);
3935         }
3936 }
3937
3938 static void vmx_recover_nmi_blocking(struct vcpu_vmx *vmx)
3939 {
3940         u32 exit_intr_info;
3941         bool unblock_nmi;
3942         u8 vector;
3943         bool idtv_info_valid;
3944
3945         idtv_info_valid = vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK;
3946
3947         if (cpu_has_virtual_nmis()) {
3948                 if (vmx->nmi_known_unmasked)
3949                         return;
3950                 /*
3951                  * Can't use vmx->exit_intr_info since we're not sure what
3952                  * the exit reason is.
3953                  */
3954                 exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
3955                 unblock_nmi = (exit_intr_info & INTR_INFO_UNBLOCK_NMI) != 0;
3956                 vector = exit_intr_info & INTR_INFO_VECTOR_MASK;
3957                 /*
3958                  * SDM 3: 27.7.1.2 (September 2008)
3959                  * Re-set bit "block by NMI" before VM entry if vmexit caused by
3960                  * a guest IRET fault.
3961                  * SDM 3: 23.2.2 (September 2008)
3962                  * Bit 12 is undefined in any of the following cases:
3963                  *  If the VM exit sets the valid bit in the IDT-vectoring
3964                  *   information field.
3965                  *  If the VM exit is due to a double fault.
3966                  */
3967                 if ((exit_intr_info & INTR_INFO_VALID_MASK) && unblock_nmi &&
3968                     vector != DF_VECTOR && !idtv_info_valid)
3969                         vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
3970                                       GUEST_INTR_STATE_NMI);
3971                 else
3972                         vmx->nmi_known_unmasked =
3973                                 !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO)
3974                                   & GUEST_INTR_STATE_NMI);
3975         } else if (unlikely(vmx->soft_vnmi_blocked))
3976                 vmx->vnmi_blocked_time +=
3977                         ktime_to_ns(ktime_sub(ktime_get(), vmx->entry_time));
3978 }
3979
3980 static void __vmx_complete_interrupts(struct vcpu_vmx *vmx,
3981                                       u32 idt_vectoring_info,
3982                                       int instr_len_field,
3983                                       int error_code_field)
3984 {
3985         u8 vector;
3986         int type;
3987         bool idtv_info_valid;
3988
3989         idtv_info_valid = idt_vectoring_info & VECTORING_INFO_VALID_MASK;
3990
3991         vmx->vcpu.arch.nmi_injected = false;
3992         kvm_clear_exception_queue(&vmx->vcpu);
3993         kvm_clear_interrupt_queue(&vmx->vcpu);
3994
3995         if (!idtv_info_valid)
3996                 return;
3997
3998         kvm_make_request(KVM_REQ_EVENT, &vmx->vcpu);
3999
4000         vector = idt_vectoring_info & VECTORING_INFO_VECTOR_MASK;
4001         type = idt_vectoring_info & VECTORING_INFO_TYPE_MASK;
4002
4003         switch (type) {
4004         case INTR_TYPE_NMI_INTR:
4005                 vmx->vcpu.arch.nmi_injected = true;
4006                 /*
4007                  * SDM 3: 27.7.1.2 (September 2008)
4008                  * Clear bit "block by NMI" before VM entry if a NMI
4009                  * delivery faulted.
4010                  */
4011                 vmx_set_nmi_mask(&vmx->vcpu, false);
4012                 break;
4013         case INTR_TYPE_SOFT_EXCEPTION:
4014                 vmx->vcpu.arch.event_exit_inst_len =
4015                         vmcs_read32(instr_len_field);
4016                 /* fall through */
4017         case INTR_TYPE_HARD_EXCEPTION:
4018                 if (idt_vectoring_info & VECTORING_INFO_DELIVER_CODE_MASK) {
4019                         u32 err = vmcs_read32(error_code_field);
4020                         kvm_queue_exception_e(&vmx->vcpu, vector, err);
4021                 } else
4022                         kvm_queue_exception(&vmx->vcpu, vector);
4023                 break;
4024         case INTR_TYPE_SOFT_INTR:
4025                 vmx->vcpu.arch.event_exit_inst_len =
4026                         vmcs_read32(instr_len_field);
4027                 /* fall through */
4028         case INTR_TYPE_EXT_INTR:
4029                 kvm_queue_interrupt(&vmx->vcpu, vector,
4030                         type == INTR_TYPE_SOFT_INTR);
4031                 break;
4032         default:
4033                 break;
4034         }
4035 }
4036
4037 static void vmx_complete_interrupts(struct vcpu_vmx *vmx)
4038 {
4039         __vmx_complete_interrupts(vmx, vmx->idt_vectoring_info,
4040                                   VM_EXIT_INSTRUCTION_LEN,
4041                                   IDT_VECTORING_ERROR_CODE);
4042 }
4043
4044 static void vmx_cancel_injection(struct kvm_vcpu *vcpu)
4045 {
4046         __vmx_complete_interrupts(to_vmx(vcpu),
4047                                   vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
4048                                   VM_ENTRY_INSTRUCTION_LEN,
4049                                   VM_ENTRY_EXCEPTION_ERROR_CODE);
4050
4051         vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);
4052 }
4053
4054 #ifdef CONFIG_X86_64
4055 #define R "r"
4056 #define Q "q"
4057 #else
4058 #define R "e"
4059 #define Q "l"
4060 #endif
4061
4062 static void __noclone vmx_vcpu_run(struct kvm_vcpu *vcpu)
4063 {
4064         struct vcpu_vmx *vmx = to_vmx(vcpu);
4065
4066         /* Record the guest's net vcpu time for enforced NMI injections. */
4067         if (unlikely(!cpu_has_virtual_nmis() && vmx->soft_vnmi_blocked))
4068                 vmx->entry_time = ktime_get();
4069
4070         /* Don't enter VMX if guest state is invalid, let the exit handler
4071            start emulation until we arrive back to a valid state */
4072         if (vmx->emulation_required && emulate_invalid_guest_state)
4073                 return;
4074
4075         if (test_bit(VCPU_REGS_RSP, (unsigned long *)&vcpu->arch.regs_dirty))
4076                 vmcs_writel(GUEST_RSP, vcpu->arch.regs[VCPU_REGS_RSP]);
4077         if (test_bit(VCPU_REGS_RIP, (unsigned long *)&vcpu->arch.regs_dirty))
4078                 vmcs_writel(GUEST_RIP, vcpu->arch.regs[VCPU_REGS_RIP]);
4079
4080         /* When single-stepping over STI and MOV SS, we must clear the
4081          * corresponding interruptibility bits in the guest state. Otherwise
4082          * vmentry fails as it then expects bit 14 (BS) in pending debug
4083          * exceptions being set, but that's not correct for the guest debugging
4084          * case. */
4085         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
4086                 vmx_set_interrupt_shadow(vcpu, 0);
4087
4088         asm(
4089                 /* Store host registers */
4090                 "push %%"R"dx; push %%"R"bp;"
4091                 "push %%"R"cx \n\t" /* placeholder for guest rcx */
4092                 "push %%"R"cx \n\t"
4093                 "cmp %%"R"sp, %c[host_rsp](%0) \n\t"
4094                 "je 1f \n\t"
4095                 "mov %%"R"sp, %c[host_rsp](%0) \n\t"
4096                 __ex(ASM_VMX_VMWRITE_RSP_RDX) "\n\t"
4097                 "1: \n\t"
4098                 /* Reload cr2 if changed */
4099                 "mov %c[cr2](%0), %%"R"ax \n\t"
4100                 "mov %%cr2, %%"R"dx \n\t"
4101                 "cmp %%"R"ax, %%"R"dx \n\t"
4102                 "je 2f \n\t"
4103                 "mov %%"R"ax, %%cr2 \n\t"
4104                 "2: \n\t"
4105                 /* Check if vmlaunch of vmresume is needed */
4106                 "cmpl $0, %c[launched](%0) \n\t"
4107                 /* Load guest registers.  Don't clobber flags. */
4108                 "mov %c[rax](%0), %%"R"ax \n\t"
4109                 "mov %c[rbx](%0), %%"R"bx \n\t"
4110                 "mov %c[rdx](%0), %%"R"dx \n\t"
4111                 "mov %c[rsi](%0), %%"R"si \n\t"
4112                 "mov %c[rdi](%0), %%"R"di \n\t"
4113                 "mov %c[rbp](%0), %%"R"bp \n\t"
4114 #ifdef CONFIG_X86_64
4115                 "mov %c[r8](%0),  %%r8  \n\t"
4116                 "mov %c[r9](%0),  %%r9  \n\t"
4117                 "mov %c[r10](%0), %%r10 \n\t"
4118                 "mov %c[r11](%0), %%r11 \n\t"
4119                 "mov %c[r12](%0), %%r12 \n\t"
4120                 "mov %c[r13](%0), %%r13 \n\t"
4121                 "mov %c[r14](%0), %%r14 \n\t"
4122                 "mov %c[r15](%0), %%r15 \n\t"
4123 #endif
4124                 "mov %c[rcx](%0), %%"R"cx \n\t" /* kills %0 (ecx) */
4125
4126                 /* Enter guest mode */
4127                 "jne .Llaunched \n\t"
4128                 __ex(ASM_VMX_VMLAUNCH) "\n\t"
4129                 "jmp .Lkvm_vmx_return \n\t"
4130                 ".Llaunched: " __ex(ASM_VMX_VMRESUME) "\n\t"
4131                 ".Lkvm_vmx_return: "
4132                 /* Save guest registers, load host registers, keep flags */
4133                 "mov %0, %c[wordsize](%%"R"sp) \n\t"
4134                 "pop %0 \n\t"
4135                 "mov %%"R"ax, %c[rax](%0) \n\t"
4136                 "mov %%"R"bx, %c[rbx](%0) \n\t"
4137                 "pop"Q" %c[rcx](%0) \n\t"
4138                 "mov %%"R"dx, %c[rdx](%0) \n\t"
4139                 "mov %%"R"si, %c[rsi](%0) \n\t"
4140                 "mov %%"R"di, %c[rdi](%0) \n\t"
4141                 "mov %%"R"bp, %c[rbp](%0) \n\t"
4142 #ifdef CONFIG_X86_64
4143                 "mov %%r8,  %c[r8](%0) \n\t"
4144                 "mov %%r9,  %c[r9](%0) \n\t"
4145                 "mov %%r10, %c[r10](%0) \n\t"
4146                 "mov %%r11, %c[r11](%0) \n\t"
4147                 "mov %%r12, %c[r12](%0) \n\t"
4148                 "mov %%r13, %c[r13](%0) \n\t"
4149                 "mov %%r14, %c[r14](%0) \n\t"
4150                 "mov %%r15, %c[r15](%0) \n\t"
4151 #endif
4152                 "mov %%cr2, %%"R"ax   \n\t"
4153                 "mov %%"R"ax, %c[cr2](%0) \n\t"
4154
4155                 "pop  %%"R"bp; pop  %%"R"dx \n\t"
4156                 "setbe %c[fail](%0) \n\t"
4157               : : "c"(vmx), "d"((unsigned long)HOST_RSP),
4158                 [launched]"i"(offsetof(struct vcpu_vmx, launched)),
4159                 [fail]"i"(offsetof(struct vcpu_vmx, fail)),
4160                 [host_rsp]"i"(offsetof(struct vcpu_vmx, host_rsp)),
4161                 [rax]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RAX])),
4162                 [rbx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBX])),
4163                 [rcx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RCX])),
4164                 [rdx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDX])),
4165                 [rsi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RSI])),
4166                 [rdi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDI])),
4167                 [rbp]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBP])),
4168 #ifdef CONFIG_X86_64
4169                 [r8]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R8])),
4170                 [r9]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R9])),
4171                 [r10]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R10])),
4172                 [r11]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R11])),
4173                 [r12]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R12])),
4174                 [r13]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R13])),
4175                 [r14]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R14])),
4176                 [r15]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R15])),
4177 #endif
4178                 [cr2]"i"(offsetof(struct vcpu_vmx, vcpu.arch.cr2)),
4179                 [wordsize]"i"(sizeof(ulong))
4180               : "cc", "memory"
4181                 , R"ax", R"bx", R"di", R"si"
4182 #ifdef CONFIG_X86_64
4183                 , "r8", "r9", "r10", "r11", "r12", "r13", "r14", "r15"
4184 #endif
4185               );
4186
4187         vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP)
4188                                   | (1 << VCPU_EXREG_RFLAGS)
4189                                   | (1 << VCPU_EXREG_CPL)
4190                                   | (1 << VCPU_EXREG_PDPTR)
4191                                   | (1 << VCPU_EXREG_CR3));
4192         vcpu->arch.regs_dirty = 0;
4193
4194         vmx->idt_vectoring_info = vmcs_read32(IDT_VECTORING_INFO_FIELD);
4195
4196         asm("mov %0, %%ds; mov %0, %%es" : : "r"(__USER_DS));
4197         vmx->launched = 1;
4198
4199         vmx->exit_reason = vmcs_read32(VM_EXIT_REASON);
4200
4201         vmx_complete_atomic_exit(vmx);
4202         vmx_recover_nmi_blocking(vmx);
4203         vmx_complete_interrupts(vmx);
4204 }
4205
4206 #undef R
4207 #undef Q
4208
4209 static void vmx_free_vmcs(struct kvm_vcpu *vcpu)
4210 {
4211         struct vcpu_vmx *vmx = to_vmx(vcpu);
4212
4213         if (vmx->vmcs) {
4214                 vcpu_clear(vmx);
4215                 free_vmcs(vmx->vmcs);
4216                 vmx->vmcs = NULL;
4217         }
4218 }
4219
4220 static void vmx_free_vcpu(struct kvm_vcpu *vcpu)
4221 {
4222         struct vcpu_vmx *vmx = to_vmx(vcpu);
4223
4224         free_vpid(vmx);
4225         vmx_free_vmcs(vcpu);
4226         kfree(vmx->guest_msrs);
4227         kvm_vcpu_uninit(vcpu);
4228         kmem_cache_free(kvm_vcpu_cache, vmx);
4229 }
4230
4231 static inline void vmcs_init(struct vmcs *vmcs)
4232 {
4233         u64 phys_addr = __pa(per_cpu(vmxarea, raw_smp_processor_id()));
4234
4235         if (!vmm_exclusive)
4236                 kvm_cpu_vmxon(phys_addr);
4237
4238         vmcs_clear(vmcs);
4239
4240         if (!vmm_exclusive)
4241                 kvm_cpu_vmxoff();
4242 }
4243
4244 static struct kvm_vcpu *vmx_create_vcpu(struct kvm *kvm, unsigned int id)
4245 {
4246         int err;
4247         struct vcpu_vmx *vmx = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
4248         int cpu;
4249
4250         if (!vmx)
4251                 return ERR_PTR(-ENOMEM);
4252
4253         allocate_vpid(vmx);
4254
4255         err = kvm_vcpu_init(&vmx->vcpu, kvm, id);
4256         if (err)
4257                 goto free_vcpu;
4258
4259         vmx->guest_msrs = kmalloc(PAGE_SIZE, GFP_KERNEL);
4260         err = -ENOMEM;
4261         if (!vmx->guest_msrs) {
4262                 goto uninit_vcpu;
4263         }
4264
4265         vmx->vmcs = alloc_vmcs();
4266         if (!vmx->vmcs)
4267                 goto free_msrs;
4268
4269         vmcs_init(vmx->vmcs);
4270
4271         cpu = get_cpu();
4272         vmx_vcpu_load(&vmx->vcpu, cpu);
4273         vmx->vcpu.cpu = cpu;
4274         err = vmx_vcpu_setup(vmx);
4275         vmx_vcpu_put(&vmx->vcpu);
4276         put_cpu();
4277         if (err)
4278                 goto free_vmcs;
4279         if (vm_need_virtualize_apic_accesses(kvm))
4280                 err = alloc_apic_access_page(kvm);
4281                 if (err)
4282                         goto free_vmcs;
4283
4284         if (enable_ept) {
4285                 if (!kvm->arch.ept_identity_map_addr)
4286                         kvm->arch.ept_identity_map_addr =
4287                                 VMX_EPT_IDENTITY_PAGETABLE_ADDR;
4288                 err = -ENOMEM;
4289                 if (alloc_identity_pagetable(kvm) != 0)
4290                         goto free_vmcs;
4291                 if (!init_rmode_identity_map(kvm))
4292                         goto free_vmcs;
4293         }
4294
4295         return &vmx->vcpu;
4296
4297 free_vmcs:
4298         free_vmcs(vmx->vmcs);
4299 free_msrs:
4300         kfree(vmx->guest_msrs);
4301 uninit_vcpu:
4302         kvm_vcpu_uninit(&vmx->vcpu);
4303 free_vcpu:
4304         free_vpid(vmx);
4305         kmem_cache_free(kvm_vcpu_cache, vmx);
4306         return ERR_PTR(err);
4307 }
4308
4309 static void __init vmx_check_processor_compat(void *rtn)
4310 {
4311         struct vmcs_config vmcs_conf;
4312
4313         *(int *)rtn = 0;
4314         if (setup_vmcs_config(&vmcs_conf) < 0)
4315                 *(int *)rtn = -EIO;
4316         if (memcmp(&vmcs_config, &vmcs_conf, sizeof(struct vmcs_config)) != 0) {
4317                 printk(KERN_ERR "kvm: CPU %d feature inconsistency!\n",
4318                                 smp_processor_id());
4319                 *(int *)rtn = -EIO;
4320         }
4321 }
4322
4323 static int get_ept_level(void)
4324 {
4325         return VMX_EPT_DEFAULT_GAW + 1;
4326 }
4327
4328 static u64 vmx_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio)
4329 {
4330         u64 ret;
4331
4332         /* For VT-d and EPT combination
4333          * 1. MMIO: always map as UC
4334          * 2. EPT with VT-d:
4335          *   a. VT-d without snooping control feature: can't guarantee the
4336          *      result, try to trust guest.
4337          *   b. VT-d with snooping control feature: snooping control feature of
4338          *      VT-d engine can guarantee the cache correctness. Just set it
4339          *      to WB to keep consistent with host. So the same as item 3.
4340          * 3. EPT without VT-d: always map as WB and set IPAT=1 to keep
4341          *    consistent with host MTRR
4342          */
4343         if (is_mmio)
4344                 ret = MTRR_TYPE_UNCACHABLE << VMX_EPT_MT_EPTE_SHIFT;
4345         else if (vcpu->kvm->arch.iommu_domain &&
4346                 !(vcpu->kvm->arch.iommu_flags & KVM_IOMMU_CACHE_COHERENCY))
4347                 ret = kvm_get_guest_memory_type(vcpu, gfn) <<
4348                       VMX_EPT_MT_EPTE_SHIFT;
4349         else
4350                 ret = (MTRR_TYPE_WRBACK << VMX_EPT_MT_EPTE_SHIFT)
4351                         | VMX_EPT_IPAT_BIT;
4352
4353         return ret;
4354 }
4355
4356 #define _ER(x) { EXIT_REASON_##x, #x }
4357
4358 static const struct trace_print_flags vmx_exit_reasons_str[] = {
4359         _ER(EXCEPTION_NMI),
4360         _ER(EXTERNAL_INTERRUPT),
4361         _ER(TRIPLE_FAULT),
4362         _ER(PENDING_INTERRUPT),
4363         _ER(NMI_WINDOW),
4364         _ER(TASK_SWITCH),
4365         _ER(CPUID),
4366         _ER(HLT),
4367         _ER(INVLPG),
4368         _ER(RDPMC),
4369         _ER(RDTSC),
4370         _ER(VMCALL),
4371         _ER(VMCLEAR),
4372         _ER(VMLAUNCH),
4373         _ER(VMPTRLD),
4374         _ER(VMPTRST),
4375         _ER(VMREAD),
4376         _ER(VMRESUME),
4377         _ER(VMWRITE),
4378         _ER(VMOFF),
4379         _ER(VMON),
4380         _ER(CR_ACCESS),
4381         _ER(DR_ACCESS),
4382         _ER(IO_INSTRUCTION),
4383         _ER(MSR_READ),
4384         _ER(MSR_WRITE),
4385         _ER(MWAIT_INSTRUCTION),
4386         _ER(MONITOR_INSTRUCTION),
4387         _ER(PAUSE_INSTRUCTION),
4388         _ER(MCE_DURING_VMENTRY),
4389         _ER(TPR_BELOW_THRESHOLD),
4390         _ER(APIC_ACCESS),
4391         _ER(EPT_VIOLATION),
4392         _ER(EPT_MISCONFIG),
4393         _ER(WBINVD),
4394         { -1, NULL }
4395 };
4396
4397 #undef _ER
4398
4399 static int vmx_get_lpage_level(void)
4400 {
4401         if (enable_ept && !cpu_has_vmx_ept_1g_page())
4402                 return PT_DIRECTORY_LEVEL;
4403         else
4404                 /* For shadow and EPT supported 1GB page */
4405                 return PT_PDPE_LEVEL;
4406 }
4407
4408 static void vmx_cpuid_update(struct kvm_vcpu *vcpu)
4409 {
4410         struct kvm_cpuid_entry2 *best;
4411         struct vcpu_vmx *vmx = to_vmx(vcpu);
4412         u32 exec_control;
4413
4414         vmx->rdtscp_enabled = false;
4415         if (vmx_rdtscp_supported()) {
4416                 exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
4417                 if (exec_control & SECONDARY_EXEC_RDTSCP) {
4418                         best = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
4419                         if (best && (best->edx & bit(X86_FEATURE_RDTSCP)))
4420                                 vmx->rdtscp_enabled = true;
4421                         else {
4422                                 exec_control &= ~SECONDARY_EXEC_RDTSCP;
4423                                 vmcs_write32(SECONDARY_VM_EXEC_CONTROL,
4424                                                 exec_control);
4425                         }
4426                 }
4427         }
4428 }
4429
4430 static void vmx_set_supported_cpuid(u32 func, struct kvm_cpuid_entry2 *entry)
4431 {
4432 }
4433
4434 static int vmx_check_intercept(struct kvm_vcpu *vcpu,
4435                                struct x86_instruction_info *info,
4436                                enum x86_intercept_stage stage)
4437 {
4438         return X86EMUL_CONTINUE;
4439 }
4440
4441 static struct kvm_x86_ops vmx_x86_ops = {
4442         .cpu_has_kvm_support = cpu_has_kvm_support,
4443         .disabled_by_bios = vmx_disabled_by_bios,
4444         .hardware_setup = hardware_setup,
4445         .hardware_unsetup = hardware_unsetup,
4446         .check_processor_compatibility = vmx_check_processor_compat,
4447         .hardware_enable = hardware_enable,
4448         .hardware_disable = hardware_disable,
4449         .cpu_has_accelerated_tpr = report_flexpriority,
4450
4451         .vcpu_create = vmx_create_vcpu,
4452         .vcpu_free = vmx_free_vcpu,
4453         .vcpu_reset = vmx_vcpu_reset,
4454
4455         .prepare_guest_switch = vmx_save_host_state,
4456         .vcpu_load = vmx_vcpu_load,
4457         .vcpu_put = vmx_vcpu_put,
4458
4459         .set_guest_debug = set_guest_debug,
4460         .get_msr = vmx_get_msr,
4461         .set_msr = vmx_set_msr,
4462         .get_segment_base = vmx_get_segment_base,
4463         .get_segment = vmx_get_segment,
4464         .set_segment = vmx_set_segment,
4465         .get_cpl = vmx_get_cpl,
4466         .get_cs_db_l_bits = vmx_get_cs_db_l_bits,
4467         .decache_cr0_guest_bits = vmx_decache_cr0_guest_bits,
4468         .decache_cr3 = vmx_decache_cr3,
4469         .decache_cr4_guest_bits = vmx_decache_cr4_guest_bits,
4470         .set_cr0 = vmx_set_cr0,
4471         .set_cr3 = vmx_set_cr3,
4472         .set_cr4 = vmx_set_cr4,
4473         .set_efer = vmx_set_efer,
4474         .get_idt = vmx_get_idt,
4475         .set_idt = vmx_set_idt,
4476         .get_gdt = vmx_get_gdt,
4477         .set_gdt = vmx_set_gdt,
4478         .set_dr7 = vmx_set_dr7,
4479         .cache_reg = vmx_cache_reg,
4480         .get_rflags = vmx_get_rflags,
4481         .set_rflags = vmx_set_rflags,
4482         .fpu_activate = vmx_fpu_activate,
4483         .fpu_deactivate = vmx_fpu_deactivate,
4484
4485         .tlb_flush = vmx_flush_tlb,
4486
4487         .run = vmx_vcpu_run,
4488         .handle_exit = vmx_handle_exit,
4489         .skip_emulated_instruction = skip_emulated_instruction,
4490         .set_interrupt_shadow = vmx_set_interrupt_shadow,
4491         .get_interrupt_shadow = vmx_get_interrupt_shadow,
4492         .patch_hypercall = vmx_patch_hypercall,
4493         .set_irq = vmx_inject_irq,
4494         .set_nmi = vmx_inject_nmi,
4495         .queue_exception = vmx_queue_exception,
4496         .cancel_injection = vmx_cancel_injection,
4497         .interrupt_allowed = vmx_interrupt_allowed,
4498         .nmi_allowed = vmx_nmi_allowed,
4499         .get_nmi_mask = vmx_get_nmi_mask,
4500         .set_nmi_mask = vmx_set_nmi_mask,
4501         .enable_nmi_window = enable_nmi_window,
4502         .enable_irq_window = enable_irq_window,
4503         .update_cr8_intercept = update_cr8_intercept,
4504
4505         .set_tss_addr = vmx_set_tss_addr,
4506         .get_tdp_level = get_ept_level,
4507         .get_mt_mask = vmx_get_mt_mask,
4508
4509         .get_exit_info = vmx_get_exit_info,
4510         .exit_reasons_str = vmx_exit_reasons_str,
4511
4512         .get_lpage_level = vmx_get_lpage_level,
4513
4514         .cpuid_update = vmx_cpuid_update,
4515
4516         .rdtscp_supported = vmx_rdtscp_supported,
4517
4518         .set_supported_cpuid = vmx_set_supported_cpuid,
4519
4520         .has_wbinvd_exit = cpu_has_vmx_wbinvd_exit,
4521
4522         .set_tsc_khz = vmx_set_tsc_khz,
4523         .write_tsc_offset = vmx_write_tsc_offset,
4524         .adjust_tsc_offset = vmx_adjust_tsc_offset,
4525         .compute_tsc_offset = vmx_compute_tsc_offset,
4526
4527         .set_tdp_cr3 = vmx_set_cr3,
4528
4529         .check_intercept = vmx_check_intercept,
4530 };
4531
4532 static int __init vmx_init(void)
4533 {
4534         int r, i;
4535
4536         rdmsrl_safe(MSR_EFER, &host_efer);
4537
4538         for (i = 0; i < NR_VMX_MSR; ++i)
4539                 kvm_define_shared_msr(i, vmx_msr_index[i]);
4540
4541         vmx_io_bitmap_a = (unsigned long *)__get_free_page(GFP_KERNEL);
4542         if (!vmx_io_bitmap_a)
4543                 return -ENOMEM;
4544
4545         vmx_io_bitmap_b = (unsigned long *)__get_free_page(GFP_KERNEL);
4546         if (!vmx_io_bitmap_b) {
4547                 r = -ENOMEM;
4548                 goto out;
4549         }
4550
4551         vmx_msr_bitmap_legacy = (unsigned long *)__get_free_page(GFP_KERNEL);
4552         if (!vmx_msr_bitmap_legacy) {
4553                 r = -ENOMEM;
4554                 goto out1;
4555         }
4556
4557         vmx_msr_bitmap_longmode = (unsigned long *)__get_free_page(GFP_KERNEL);
4558         if (!vmx_msr_bitmap_longmode) {
4559                 r = -ENOMEM;
4560                 goto out2;
4561         }
4562
4563         /*
4564          * Allow direct access to the PC debug port (it is often used for I/O
4565          * delays, but the vmexits simply slow things down).
4566          */
4567         memset(vmx_io_bitmap_a, 0xff, PAGE_SIZE);
4568         clear_bit(0x80, vmx_io_bitmap_a);
4569
4570         memset(vmx_io_bitmap_b, 0xff, PAGE_SIZE);
4571
4572         memset(vmx_msr_bitmap_legacy, 0xff, PAGE_SIZE);
4573         memset(vmx_msr_bitmap_longmode, 0xff, PAGE_SIZE);
4574
4575         set_bit(0, vmx_vpid_bitmap); /* 0 is reserved for host */
4576
4577         r = kvm_init(&vmx_x86_ops, sizeof(struct vcpu_vmx),
4578                      __alignof__(struct vcpu_vmx), THIS_MODULE);
4579         if (r)
4580                 goto out3;
4581
4582         vmx_disable_intercept_for_msr(MSR_FS_BASE, false);
4583         vmx_disable_intercept_for_msr(MSR_GS_BASE, false);
4584         vmx_disable_intercept_for_msr(MSR_KERNEL_GS_BASE, true);
4585         vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_CS, false);
4586         vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_ESP, false);
4587         vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_EIP, false);
4588
4589         if (enable_ept) {
4590                 bypass_guest_pf = 0;
4591                 kvm_mmu_set_mask_ptes(0ull, 0ull, 0ull, 0ull,
4592                                 VMX_EPT_EXECUTABLE_MASK);
4593                 kvm_enable_tdp();
4594         } else
4595                 kvm_disable_tdp();
4596
4597         if (bypass_guest_pf)
4598                 kvm_mmu_set_nonpresent_ptes(~0xffeull, 0ull);
4599
4600         return 0;
4601
4602 out3:
4603         free_page((unsigned long)vmx_msr_bitmap_longmode);
4604 out2:
4605         free_page((unsigned long)vmx_msr_bitmap_legacy);
4606 out1:
4607         free_page((unsigned long)vmx_io_bitmap_b);
4608 out:
4609         free_page((unsigned long)vmx_io_bitmap_a);
4610         return r;
4611 }
4612
4613 static void __exit vmx_exit(void)
4614 {
4615         free_page((unsigned long)vmx_msr_bitmap_legacy);
4616         free_page((unsigned long)vmx_msr_bitmap_longmode);
4617         free_page((unsigned long)vmx_io_bitmap_b);
4618         free_page((unsigned long)vmx_io_bitmap_a);
4619
4620         kvm_exit();
4621 }
4622
4623 module_init(vmx_init)
4624 module_exit(vmx_exit)