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