arch/x86/kvm/vmx.c

   1 /*
   2  * Kernel-based Virtual Machine driver for Linux
   3  *
   4  * This module enables machines with Intel VT-x extensions to run virtual
   5  * machines without emulation or binary translation.
   6  *
   7  * Copyright (C) 2006 Qumranet, Inc.
   8  * Copyright 2010 Red Hat, Inc. and/or its affiliates.
   9  *
  10  * Authors:
  11  *   Avi Kivity   <avi@qumranet.com>
  12  *   Yaniv Kamay  <yaniv@qumranet.com>
  13  *
  14  * This work is licensed under the terms of the GNU GPL, version 2.  See
  15  * the COPYING file in the top-level directory.
  16  *
  17  */
  18
  19 #include "irq.h"
  20 #include "mmu.h"
  21
  22 #include <linux/kvm_host.h>
  23 #include <linux/module.h>
  24 #include <linux/kernel.h>
  25 #include <linux/mm.h>
  26 #include <linux/highmem.h>
  27 #include <linux/sched.h>
  28 #include <linux/moduleparam.h>
  29 #include <linux/ftrace_event.h>
  30 #include <linux/slab.h>
  31 #include <linux/tboot.h>
  32 #include "kvm_cache_regs.h"
  33 #include "x86.h"
  34
  35 #include <asm/io.h>
  36 #include <asm/desc.h>
  37 #include <asm/vmx.h>
  38 #include <asm/virtext.h>
  39 #include <asm/mce.h>
  40 #include <asm/i387.h>
  41 #include <asm/xcr.h>
  42
  43 #include "trace.h"
  44
  45 #define __ex(x) __kvm_handle_fault_on_reboot(x)
  46
  47 MODULE_AUTHOR("Qumranet");
  48 MODULE_LICENSE("GPL");
  49
  50 static int __read_mostly bypass_guest_pf = 1;
  51 module_param(bypass_guest_pf, bool, S_IRUGO);
  52
  53 static int __read_mostly enable_vpid = 1;
  54 module_param_named(vpid, enable_vpid, bool, 0444);
  55
  56 static int __read_mostly flexpriority_enabled = 1;
  57 module_param_named(flexpriority, flexpriority_enabled, bool, S_IRUGO);
  58
  59 static int __read_mostly enable_ept = 1;
  60 module_param_named(ept, enable_ept, bool, S_IRUGO);
  61
  62 static int __read_mostly enable_unrestricted_guest = 1;
  63 module_param_named(unrestricted_guest,
  64                         enable_unrestricted_guest, bool, S_IRUGO);
  65
  66 static int __read_mostly emulate_invalid_guest_state = 0;
  67 module_param(emulate_invalid_guest_state, bool, S_IRUGO);
  68
  69 static int __read_mostly vmm_exclusive = 1;
  70 module_param(vmm_exclusive, bool, S_IRUGO);
  71
  72 static int __read_mostly yield_on_hlt = 1;
  73 module_param(yield_on_hlt, bool, S_IRUGO);
  74
  75 #define KVM_GUEST_CR0_MASK_UNRESTRICTED_GUEST                           \
  76         (X86_CR0_WP | X86_CR0_NE | X86_CR0_NW | X86_CR0_CD)
  77 #define KVM_GUEST_CR0_MASK                                              \
  78         (KVM_GUEST_CR0_MASK_UNRESTRICTED_GUEST | X86_CR0_PG | X86_CR0_PE)
  79 #define KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST                         \
  80         (X86_CR0_WP | X86_CR0_NE)
  81 #define KVM_VM_CR0_ALWAYS_ON                                            \
  82         (KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST | X86_CR0_PG | X86_CR0_PE)
  83 #define KVM_CR4_GUEST_OWNED_BITS                                      \
  84         (X86_CR4_PVI | X86_CR4_DE | X86_CR4_PCE | X86_CR4_OSFXSR      \
  85          | X86_CR4_OSXMMEXCPT)
  86
  87 #define KVM_PMODE_VM_CR4_ALWAYS_ON (X86_CR4_PAE | X86_CR4_VMXE)
  88 #define KVM_RMODE_VM_CR4_ALWAYS_ON (X86_CR4_VME | X86_CR4_PAE | X86_CR4_VMXE)
  89
  90 #define RMODE_GUEST_OWNED_EFLAGS_BITS (~(X86_EFLAGS_IOPL | X86_EFLAGS_VM))
  91
  92 /*
  93  * These 2 parameters are used to config the controls for Pause-Loop Exiting:
  94  * ple_gap:    upper bound on the amount of time between two successive
  95  *             executions of PAUSE in a loop. Also indicate if ple enabled.
  96  *             According to test, this time is usually smaller than 128 cycles.
  97  * ple_window: upper bound on the amount of time a guest is allowed to execute
  98  *             in a PAUSE loop. Tests indicate that most spinlocks are held for
  99  *             less than 2^12 cycles
 100  * Time is measured based on a counter that runs at the same rate as the TSC,
 101  * refer SDM volume 3b section 21.6.13 & 22.1.3.
 102  */
 103 #define KVM_VMX_DEFAULT_PLE_GAP    128
 104 #define KVM_VMX_DEFAULT_PLE_WINDOW 4096
 105 static int ple_gap = KVM_VMX_DEFAULT_PLE_GAP;
 106 module_param(ple_gap, int, S_IRUGO);
 107
 108 static int ple_window = KVM_VMX_DEFAULT_PLE_WINDOW;
 109 module_param(ple_window, int, S_IRUGO);
 110
 111 #define NR_AUTOLOAD_MSRS 1
 112
 113 struct vmcs {
 114         u32 revision_id;
 115         u32 abort;
 116         char data[0];
 117 };
 118
 119 struct shared_msr_entry {
 120         unsigned index;
 121         u64 data;
 122         u64 mask;
 123 };
 124
 125 struct vcpu_vmx {
 126         struct kvm_vcpu       vcpu;
 127         struct list_head      local_vcpus_link;
 128         unsigned long         host_rsp;
 129         int                   launched;
 130         u8                    fail;
 131         u8                    cpl;
 132         bool                  nmi_known_unmasked;
 133         u32                   exit_intr_info;
 134         u32                   idt_vectoring_info;
 135         ulong                 rflags;
 136         struct shared_msr_entry *guest_msrs;
 137         int                   nmsrs;
 138         int                   save_nmsrs;
 139 #ifdef CONFIG_X86_64
 140         u64                   msr_host_kernel_gs_base;
 141         u64                   msr_guest_kernel_gs_base;
 142 #endif
 143         struct vmcs          *vmcs;
 144         struct msr_autoload {
 145                 unsigned nr;
 146                 struct vmx_msr_entry guest[NR_AUTOLOAD_MSRS];
 147                 struct vmx_msr_entry host[NR_AUTOLOAD_MSRS];
 148         } msr_autoload;
 149         struct {
 150                 int           loaded;
 151                 u16           fs_sel, gs_sel, ldt_sel;
 152                 int           gs_ldt_reload_needed;
 153                 int           fs_reload_needed;
 154         } host_state;
 155         struct {
 156                 int vm86_active;
 157                 ulong save_rflags;
 158                 struct kvm_save_segment {
 159                         u16 selector;
 160                         unsigned long base;
 161                         u32 limit;
 162                         u32 ar;
 163                 } tr, es, ds, fs, gs;
 164         } rmode;
 165         struct {
 166                 u32 bitmask; /* 4 bits per segment (1 bit per field) */
 167                 struct kvm_save_segment seg[8];
 168         } segment_cache;
 169         int vpid;
 170         bool emulation_required;
 171
 172         /* Support for vnmi-less CPUs */
 173         int soft_vnmi_blocked;
 174         ktime_t entry_time;
 175         s64 vnmi_blocked_time;
 176         u32 exit_reason;
 177
 178         bool rdtscp_enabled;
 179 };
 180
 181 enum segment_cache_field {
 182         SEG_FIELD_SEL = 0,
 183         SEG_FIELD_BASE = 1,
 184         SEG_FIELD_LIMIT = 2,
 185         SEG_FIELD_AR = 3,
 186
 187         SEG_FIELD_NR = 4
 188 };
 189
 190 static inline struct vcpu_vmx *to_vmx(struct kvm_vcpu *vcpu)
 191 {
 192         return container_of(vcpu, struct vcpu_vmx, vcpu);
 193 }
 194
 195 static u64 construct_eptp(unsigned long root_hpa);
 196 static void kvm_cpu_vmxon(u64 addr);
 197 static void kvm_cpu_vmxoff(void);
 198 static void vmx_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3);
 199 static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr);
 200
 201 static DEFINE_PER_CPU(struct vmcs *, vmxarea);
 202 static DEFINE_PER_CPU(struct vmcs *, current_vmcs);
 203 static DEFINE_PER_CPU(struct list_head, vcpus_on_cpu);
 204 static DEFINE_PER_CPU(struct desc_ptr, host_gdt);
 205
 206 static unsigned long *vmx_io_bitmap_a;
 207 static unsigned long *vmx_io_bitmap_b;
 208 static unsigned long *vmx_msr_bitmap_legacy;
 209 static unsigned long *vmx_msr_bitmap_longmode;
 210
 211 static bool cpu_has_load_ia32_efer;
 212
 213 static DECLARE_BITMAP(vmx_vpid_bitmap, VMX_NR_VPIDS);
 214 static DEFINE_SPINLOCK(vmx_vpid_lock);
 215
 216 static struct vmcs_config {
 217         int size;
 218         int order;
 219         u32 revision_id;
 220         u32 pin_based_exec_ctrl;
 221         u32 cpu_based_exec_ctrl;
 222         u32 cpu_based_2nd_exec_ctrl;
 223         u32 vmexit_ctrl;
 224         u32 vmentry_ctrl;
 225 } vmcs_config;
 226
 227 static struct vmx_capability {
 228         u32 ept;
 229         u32 vpid;
 230 } vmx_capability;
 231
 232 #define VMX_SEGMENT_FIELD(seg)                                  \
 233         [VCPU_SREG_##seg] = {                                   \
 234                 .selector = GUEST_##seg##_SELECTOR,             \
 235                 .base = GUEST_##seg##_BASE,                     \
 236                 .limit = GUEST_##seg##_LIMIT,                   \
 237                 .ar_bytes = GUEST_##seg##_AR_BYTES,             \
 238         }
 239
 240 static struct kvm_vmx_segment_field {
 241         unsigned selector;
 242         unsigned base;
 243         unsigned limit;
 244         unsigned ar_bytes;
 245 } kvm_vmx_segment_fields[] = {
 246         VMX_SEGMENT_FIELD(CS),
 247         VMX_SEGMENT_FIELD(DS),
 248         VMX_SEGMENT_FIELD(ES),
 249         VMX_SEGMENT_FIELD(FS),
 250         VMX_SEGMENT_FIELD(GS),
 251         VMX_SEGMENT_FIELD(SS),
 252         VMX_SEGMENT_FIELD(TR),
 253         VMX_SEGMENT_FIELD(LDTR),
 254 };
 255
 256 static u64 host_efer;
 257
 258 static void ept_save_pdptrs(struct kvm_vcpu *vcpu);
 259
 260 /*
 261  * Keep MSR_STAR at the end, as setup_msrs() will try to optimize it
 262  * away by decrementing the array size.
 263  */
 264 static const u32 vmx_msr_index[] = {
 265 #ifdef CONFIG_X86_64
 266         MSR_SYSCALL_MASK, MSR_LSTAR, MSR_CSTAR,
 267 #endif
 268         MSR_EFER, MSR_TSC_AUX, MSR_STAR,
 269 };
 270 #define NR_VMX_MSR ARRAY_SIZE(vmx_msr_index)
 271
 272 static inline bool is_page_fault(u32 intr_info)
 273 {
 274         return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
 275                              INTR_INFO_VALID_MASK)) ==
 276                 (INTR_TYPE_HARD_EXCEPTION | PF_VECTOR | INTR_INFO_VALID_MASK);
 277 }
 278
 279 static inline bool is_no_device(u32 intr_info)
 280 {
 281         return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
 282                              INTR_INFO_VALID_MASK)) ==
 283                 (INTR_TYPE_HARD_EXCEPTION | NM_VECTOR | INTR_INFO_VALID_MASK);
 284 }
 285
 286 static inline bool is_invalid_opcode(u32 intr_info)
 287 {
 288         return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
 289                              INTR_INFO_VALID_MASK)) ==
 290                 (INTR_TYPE_HARD_EXCEPTION | UD_VECTOR | INTR_INFO_VALID_MASK);
 291 }
 292
 293 static inline bool is_external_interrupt(u32 intr_info)
 294 {
 295         return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VALID_MASK))
 296                 == (INTR_TYPE_EXT_INTR | INTR_INFO_VALID_MASK);
 297 }
 298
 299 static inline bool is_machine_check(u32 intr_info)
 300 {
 301         return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
 302                              INTR_INFO_VALID_MASK)) ==
 303                 (INTR_TYPE_HARD_EXCEPTION | MC_VECTOR | INTR_INFO_VALID_MASK);
 304 }
 305
 306 static inline bool cpu_has_vmx_msr_bitmap(void)
 307 {
 308         return vmcs_config.cpu_based_exec_ctrl & CPU_BASED_USE_MSR_BITMAPS;
 309 }
 310
 311 static inline bool cpu_has_vmx_tpr_shadow(void)
 312 {
 313         return vmcs_config.cpu_based_exec_ctrl & CPU_BASED_TPR_SHADOW;
 314 }
 315
 316 static inline bool vm_need_tpr_shadow(struct kvm *kvm)
 317 {
 318         return (cpu_has_vmx_tpr_shadow()) && (irqchip_in_kernel(kvm));
 319 }
 320
 321 static inline bool cpu_has_secondary_exec_ctrls(void)
 322 {
 323         return vmcs_config.cpu_based_exec_ctrl &
 324                 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
 325 }
 326
 327 static inline bool cpu_has_vmx_virtualize_apic_accesses(void)
 328 {
 329         return vmcs_config.cpu_based_2nd_exec_ctrl &
 330                 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
 331 }
 332
 333 static inline bool cpu_has_vmx_flexpriority(void)
 334 {
 335         return cpu_has_vmx_tpr_shadow() &&
 336                 cpu_has_vmx_virtualize_apic_accesses();
 337 }
 338
 339 static inline bool cpu_has_vmx_ept_execute_only(void)
 340 {
 341         return vmx_capability.ept & VMX_EPT_EXECUTE_ONLY_BIT;
 342 }
 343
 344 static inline bool cpu_has_vmx_eptp_uncacheable(void)
 345 {
 346         return vmx_capability.ept & VMX_EPTP_UC_BIT;
 347 }
 348
 349 static inline bool cpu_has_vmx_eptp_writeback(void)
 350 {
 351         return vmx_capability.ept & VMX_EPTP_WB_BIT;
 352 }
 353
 354 static inline bool cpu_has_vmx_ept_2m_page(void)
 355 {
 356         return vmx_capability.ept & VMX_EPT_2MB_PAGE_BIT;
 357 }
 358
 359 static inline bool cpu_has_vmx_ept_1g_page(void)
 360 {
 361         return vmx_capability.ept & VMX_EPT_1GB_PAGE_BIT;
 362 }
 363
 364 static inline bool cpu_has_vmx_ept_4levels(void)
 365 {
 366         return vmx_capability.ept & VMX_EPT_PAGE_WALK_4_BIT;
 367 }
 368
 369 static inline bool cpu_has_vmx_invept_individual_addr(void)
 370 {
 371         return vmx_capability.ept & VMX_EPT_EXTENT_INDIVIDUAL_BIT;
 372 }
 373
 374 static inline bool cpu_has_vmx_invept_context(void)
 375 {
 376         return vmx_capability.ept & VMX_EPT_EXTENT_CONTEXT_BIT;
 377 }
 378
 379 static inline bool cpu_has_vmx_invept_global(void)
 380 {
 381         return vmx_capability.ept & VMX_EPT_EXTENT_GLOBAL_BIT;
 382 }
 383
 384 static inline bool cpu_has_vmx_invvpid_single(void)
 385 {
 386         return vmx_capability.vpid & VMX_VPID_EXTENT_SINGLE_CONTEXT_BIT;
 387 }
 388
 389 static inline bool cpu_has_vmx_invvpid_global(void)
 390 {
 391         return vmx_capability.vpid & VMX_VPID_EXTENT_GLOBAL_CONTEXT_BIT;
 392 }
 393
 394 static inline bool cpu_has_vmx_ept(void)
 395 {
 396         return vmcs_config.cpu_based_2nd_exec_ctrl &
 397                 SECONDARY_EXEC_ENABLE_EPT;
 398 }
 399
 400 static inline bool cpu_has_vmx_unrestricted_guest(void)
 401 {
 402         return vmcs_config.cpu_based_2nd_exec_ctrl &
 403                 SECONDARY_EXEC_UNRESTRICTED_GUEST;
 404 }
 405
 406 static inline bool cpu_has_vmx_ple(void)
 407 {
 408         return vmcs_config.cpu_based_2nd_exec_ctrl &
 409                 SECONDARY_EXEC_PAUSE_LOOP_EXITING;
 410 }
 411
 412 static inline bool vm_need_virtualize_apic_accesses(struct kvm *kvm)
 413 {
 414         return flexpriority_enabled && irqchip_in_kernel(kvm);
 415 }
 416
 417 static inline bool cpu_has_vmx_vpid(void)
 418 {
 419         return vmcs_config.cpu_based_2nd_exec_ctrl &
 420                 SECONDARY_EXEC_ENABLE_VPID;
 421 }
 422
 423 static inline bool cpu_has_vmx_rdtscp(void)
 424 {
 425         return vmcs_config.cpu_based_2nd_exec_ctrl &
 426                 SECONDARY_EXEC_RDTSCP;
 427 }
 428
 429 static inline bool cpu_has_virtual_nmis(void)
 430 {
 431         return vmcs_config.pin_based_exec_ctrl & PIN_BASED_VIRTUAL_NMIS;
 432 }
 433
 434 static inline bool cpu_has_vmx_wbinvd_exit(void)
 435 {
 436         return vmcs_config.cpu_based_2nd_exec_ctrl &
 437                 SECONDARY_EXEC_WBINVD_EXITING;
 438 }
 439
 440 static inline bool report_flexpriority(void)
 441 {
 442         return flexpriority_enabled;
 443 }
 444
 445 static int __find_msr_index(struct vcpu_vmx *vmx, u32 msr)
 446 {
 447         int i;
 448
 449         for (i = 0; i < vmx->nmsrs; ++i)
 450                 if (vmx_msr_index[vmx->guest_msrs[i].index] == msr)
 451                         return i;
 452         return -1;
 453 }
 454
 455 static inline void __invvpid(int ext, u16 vpid, gva_t gva)
 456 {
 457     struct {
 458         u64 vpid : 16;
 459         u64 rsvd : 48;
 460         u64 gva;
 461     } operand = { vpid, 0, gva };
 462
 463     asm volatile (__ex(ASM_VMX_INVVPID)
 464                   /* CF==1 or ZF==1 --> rc = -1 */
 465                   "; ja 1f ; ud2 ; 1:"
 466                   : : "a"(&operand), "c"(ext) : "cc", "memory");
 467 }
 468
 469 static inline void __invept(int ext, u64 eptp, gpa_t gpa)
 470 {
 471         struct {
 472                 u64 eptp, gpa;
 473         } operand = {eptp, gpa};
 474
 475         asm volatile (__ex(ASM_VMX_INVEPT)
 476                         /* CF==1 or ZF==1 --> rc = -1 */
 477                         "; ja 1f ; ud2 ; 1:\n"
 478                         : : "a" (&operand), "c" (ext) : "cc", "memory");
 479 }
 480
 481 static struct shared_msr_entry *find_msr_entry(struct vcpu_vmx *vmx, u32 msr)
 482 {
 483         int i;
 484
 485         i = __find_msr_index(vmx, msr);
 486         if (i >= 0)
 487                 return &vmx->guest_msrs[i];
 488         return NULL;
 489 }
 490
 491 static void vmcs_clear(struct vmcs *vmcs)
 492 {
 493         u64 phys_addr = __pa(vmcs);
 494         u8 error;
 495
 496         asm volatile (__ex(ASM_VMX_VMCLEAR_RAX) "; setna %0"
 497                       : "=qm"(error) : "a"(&phys_addr), "m"(phys_addr)
 498                       : "cc", "memory");
 499         if (error)
 500                 printk(KERN_ERR "kvm: vmclear fail: %p/%llx\n",
 501                        vmcs, phys_addr);
 502 }
 503
 504 static void vmcs_load(struct vmcs *vmcs)
 505 {
 506         u64 phys_addr = __pa(vmcs);
 507         u8 error;
 508
 509         asm volatile (__ex(ASM_VMX_VMPTRLD_RAX) "; setna %0"
 510                         : "=qm"(error) : "a"(&phys_addr), "m"(phys_addr)
 511                         : "cc", "memory");
 512         if (error)
 513                 printk(KERN_ERR "kvm: vmptrld %p/%llx fail\n",
 514                        vmcs, phys_addr);
 515 }
 516
 517 static void __vcpu_clear(void *arg)
 518 {
 519         struct vcpu_vmx *vmx = arg;
 520         int cpu = raw_smp_processor_id();
 521
 522         if (vmx->vcpu.cpu == cpu)
 523                 vmcs_clear(vmx->vmcs);
 524         if (per_cpu(current_vmcs, cpu) == vmx->vmcs)
 525                 per_cpu(current_vmcs, cpu) = NULL;
 526         list_del(&vmx->local_vcpus_link);
 527         vmx->vcpu.cpu = -1;
 528         vmx->launched = 0;
 529 }
 530
 531 static void vcpu_clear(struct vcpu_vmx *vmx)
 532 {
 533         if (vmx->vcpu.cpu == -1)
 534                 return;
 535         smp_call_function_single(vmx->vcpu.cpu, __vcpu_clear, vmx, 1);
 536 }
 537
 538 static inline void vpid_sync_vcpu_single(struct vcpu_vmx *vmx)
 539 {
 540         if (vmx->vpid == 0)
 541                 return;
 542
 543         if (cpu_has_vmx_invvpid_single())
 544                 __invvpid(VMX_VPID_EXTENT_SINGLE_CONTEXT, vmx->vpid, 0);
 545 }
 546
 547 static inline void vpid_sync_vcpu_global(void)
 548 {
 549         if (cpu_has_vmx_invvpid_global())
 550                 __invvpid(VMX_VPID_EXTENT_ALL_CONTEXT, 0, 0);
 551 }
 552
 553 static inline void vpid_sync_context(struct vcpu_vmx *vmx)
 554 {
 555         if (cpu_has_vmx_invvpid_single())
 556                 vpid_sync_vcpu_single(vmx);
 557         else
 558                 vpid_sync_vcpu_global();
 559 }
 560
 561 static inline void ept_sync_global(void)
 562 {
 563         if (cpu_has_vmx_invept_global())
 564                 __invept(VMX_EPT_EXTENT_GLOBAL, 0, 0);
 565 }
 566
 567 static inline void ept_sync_context(u64 eptp)
 568 {
 569         if (enable_ept) {
 570                 if (cpu_has_vmx_invept_context())
 571                         __invept(VMX_EPT_EXTENT_CONTEXT, eptp, 0);
 572                 else
 573                         ept_sync_global();
 574         }
 575 }
 576
 577 static inline void ept_sync_individual_addr(u64 eptp, gpa_t gpa)
 578 {
 579         if (enable_ept) {
 580                 if (cpu_has_vmx_invept_individual_addr())
 581                         __invept(VMX_EPT_EXTENT_INDIVIDUAL_ADDR,
 582                                         eptp, gpa);
 583                 else
 584                         ept_sync_context(eptp);
 585         }
 586 }
 587
 588 static unsigned long vmcs_readl(unsigned long field)
 589 {
 590         unsigned long value = 0;
 591
 592         asm volatile (__ex(ASM_VMX_VMREAD_RDX_RAX)
 593                       : "+a"(value) : "d"(field) : "cc");
 594         return value;
 595 }
 596
 597 static u16 vmcs_read16(unsigned long field)
 598 {
 599         return vmcs_readl(field);
 600 }
 601
 602 static u32 vmcs_read32(unsigned long field)
 603 {
 604         return vmcs_readl(field);
 605 }
 606
 607 static u64 vmcs_read64(unsigned long field)
 608 {
 609 #ifdef CONFIG_X86_64
 610         return vmcs_readl(field);
 611 #else
 612         return vmcs_readl(field) | ((u64)vmcs_readl(field+1) << 32);
 613 #endif
 614 }
 615
 616 static noinline void vmwrite_error(unsigned long field, unsigned long value)
 617 {
 618         printk(KERN_ERR "vmwrite error: reg %lx value %lx (err %d)\n",
 619                field, value, vmcs_read32(VM_INSTRUCTION_ERROR));
 620         dump_stack();
 621 }
 622
 623 static void vmcs_writel(unsigned long field, unsigned long value)
 624 {
 625         u8 error;
 626
 627         asm volatile (__ex(ASM_VMX_VMWRITE_RAX_RDX) "; setna %0"
 628                        : "=q"(error) : "a"(value), "d"(field) : "cc");
 629         if (unlikely(error))
 630                 vmwrite_error(field, value);
 631 }
 632
 633 static void vmcs_write16(unsigned long field, u16 value)
 634 {
 635         vmcs_writel(field, value);
 636 }
 637
 638 static void vmcs_write32(unsigned long field, u32 value)
 639 {
 640         vmcs_writel(field, value);
 641 }
 642
 643 static void vmcs_write64(unsigned long field, u64 value)
 644 {
 645         vmcs_writel(field, value);
 646 #ifndef CONFIG_X86_64
 647         asm volatile ("");
 648         vmcs_writel(field+1, value >> 32);
 649 #endif
 650 }
 651
 652 static void vmcs_clear_bits(unsigned long field, u32 mask)
 653 {
 654         vmcs_writel(field, vmcs_readl(field) & ~mask);
 655 }
 656
 657 static void vmcs_set_bits(unsigned long field, u32 mask)
 658 {
 659         vmcs_writel(field, vmcs_readl(field) | mask);
 660 }
 661
 662 static void vmx_segment_cache_clear(struct vcpu_vmx *vmx)
 663 {
 664         vmx->segment_cache.bitmask = 0;
 665 }
 666
 667 static bool vmx_segment_cache_test_set(struct vcpu_vmx *vmx, unsigned seg,
 668                                        unsigned field)
 669 {
 670         bool ret;
 671         u32 mask = 1 << (seg * SEG_FIELD_NR + field);
 672
 673         if (!(vmx->vcpu.arch.regs_avail & (1 << VCPU_EXREG_SEGMENTS))) {
 674                 vmx->vcpu.arch.regs_avail |= (1 << VCPU_EXREG_SEGMENTS);
 675                 vmx->segment_cache.bitmask = 0;
 676         }
 677         ret = vmx->segment_cache.bitmask & mask;
 678         vmx->segment_cache.bitmask |= mask;
 679         return ret;
 680 }
 681
 682 static u16 vmx_read_guest_seg_selector(struct vcpu_vmx *vmx, unsigned seg)
 683 {
 684         u16 *p = &vmx->segment_cache.seg[seg].selector;
 685
 686         if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_SEL))
 687                 *p = vmcs_read16(kvm_vmx_segment_fields[seg].selector);
 688         return *p;
 689 }
 690
 691 static ulong vmx_read_guest_seg_base(struct vcpu_vmx *vmx, unsigned seg)
 692 {
 693         ulong *p = &vmx->segment_cache.seg[seg].base;
 694
 695         if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_BASE))
 696                 *p = vmcs_readl(kvm_vmx_segment_fields[seg].base);
 697         return *p;
 698 }
 699
 700 static u32 vmx_read_guest_seg_limit(struct vcpu_vmx *vmx, unsigned seg)
 701 {
 702         u32 *p = &vmx->segment_cache.seg[seg].limit;
 703
 704         if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_LIMIT))
 705                 *p = vmcs_read32(kvm_vmx_segment_fields[seg].limit);
 706         return *p;
 707 }
 708
 709 static u32 vmx_read_guest_seg_ar(struct vcpu_vmx *vmx, unsigned seg)
 710 {
 711         u32 *p = &vmx->segment_cache.seg[seg].ar;
 712
 713         if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_AR))
 714                 *p = vmcs_read32(kvm_vmx_segment_fields[seg].ar_bytes);
 715         return *p;
 716 }
 717
 718 static void update_exception_bitmap(struct kvm_vcpu *vcpu)
 719 {
 720         u32 eb;
 721
 722         eb = (1u << PF_VECTOR) | (1u << UD_VECTOR) | (1u << MC_VECTOR) |
 723              (1u << NM_VECTOR) | (1u << DB_VECTOR);
 724         if ((vcpu->guest_debug &
 725              (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP)) ==
 726             (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP))
 727                 eb |= 1u << BP_VECTOR;
 728         if (to_vmx(vcpu)->rmode.vm86_active)
 729                 eb = ~0;
 730         if (enable_ept)
 731                 eb &= ~(1u << PF_VECTOR); /* bypass_guest_pf = 0 */
 732         if (vcpu->fpu_active)
 733                 eb &= ~(1u << NM_VECTOR);
 734         vmcs_write32(EXCEPTION_BITMAP, eb);
 735 }
 736
 737 static void clear_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr)
 738 {
 739         unsigned i;
 740         struct msr_autoload *m = &vmx->msr_autoload;
 741
 742         if (msr == MSR_EFER && cpu_has_load_ia32_efer) {
 743                 vmcs_clear_bits(VM_ENTRY_CONTROLS, VM_ENTRY_LOAD_IA32_EFER);
 744                 vmcs_clear_bits(VM_EXIT_CONTROLS, VM_EXIT_LOAD_IA32_EFER);
 745                 return;
 746         }
 747
 748         for (i = 0; i < m->nr; ++i)
 749                 if (m->guest[i].index == msr)
 750                         break;
 751
 752         if (i == m->nr)
 753                 return;
 754         --m->nr;
 755         m->guest[i] = m->guest[m->nr];
 756         m->host[i] = m->host[m->nr];
 757         vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->nr);
 758         vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->nr);
 759 }
 760
 761 static void add_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr,
 762                                   u64 guest_val, u64 host_val)
 763 {
 764         unsigned i;
 765         struct msr_autoload *m = &vmx->msr_autoload;
 766
 767         if (msr == MSR_EFER && cpu_has_load_ia32_efer) {
 768                 vmcs_write64(GUEST_IA32_EFER, guest_val);
 769                 vmcs_write64(HOST_IA32_EFER, host_val);
 770                 vmcs_set_bits(VM_ENTRY_CONTROLS, VM_ENTRY_LOAD_IA32_EFER);
 771                 vmcs_set_bits(VM_EXIT_CONTROLS, VM_EXIT_LOAD_IA32_EFER);
 772                 return;
 773         }
 774
 775         for (i = 0; i < m->nr; ++i)
 776                 if (m->guest[i].index == msr)
 777                         break;
 778
 779         if (i == m->nr) {
 780                 ++m->nr;
 781                 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->nr);
 782                 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->nr);
 783         }
 784
 785         m->guest[i].index = msr;
 786         m->guest[i].value = guest_val;
 787         m->host[i].index = msr;
 788         m->host[i].value = host_val;
 789 }
 790
 791 static void reload_tss(void)
 792 {
 793         /*
 794          * VT restores TR but not its size.  Useless.
 795          */
 796         struct desc_ptr *gdt = &__get_cpu_var(host_gdt);
 797         struct desc_struct *descs;
 798
 799         descs = (void *)gdt->address;
 800         descs[GDT_ENTRY_TSS].type = 9; /* available TSS */
 801         load_TR_desc();
 802 }
 803
 804 static bool update_transition_efer(struct vcpu_vmx *vmx, int efer_offset)
 805 {
 806         u64 guest_efer;
 807         u64 ignore_bits;
 808
 809         guest_efer = vmx->vcpu.arch.efer;
 810
 811         /*
 812          * NX is emulated; LMA and LME handled by hardware; SCE meaninless
 813          * outside long mode
 814          */
 815         ignore_bits = EFER_NX | EFER_SCE;
 816 #ifdef CONFIG_X86_64
 817         ignore_bits |= EFER_LMA | EFER_LME;
 818         /* SCE is meaningful only in long mode on Intel */
 819         if (guest_efer & EFER_LMA)
 820                 ignore_bits &= ~(u64)EFER_SCE;
 821 #endif
 822         guest_efer &= ~ignore_bits;
 823         guest_efer |= host_efer & ignore_bits;
 824         vmx->guest_msrs[efer_offset].data = guest_efer;
 825         vmx->guest_msrs[efer_offset].mask = ~ignore_bits;
 826
 827         clear_atomic_switch_msr(vmx, MSR_EFER);
 828         /* On ept, can't emulate nx, and must switch nx atomically */
 829         if (enable_ept && ((vmx->vcpu.arch.efer ^ host_efer) & EFER_NX)) {
 830                 guest_efer = vmx->vcpu.arch.efer;
 831                 if (!(guest_efer & EFER_LMA))
 832                         guest_efer &= ~EFER_LME;
 833                 add_atomic_switch_msr(vmx, MSR_EFER, guest_efer, host_efer);
 834                 return false;
 835         }
 836
 837         return true;
 838 }
 839
 840 static unsigned long segment_base(u16 selector)
 841 {
 842         struct desc_ptr *gdt = &__get_cpu_var(host_gdt);
 843         struct desc_struct *d;
 844         unsigned long table_base;
 845         unsigned long v;
 846
 847         if (!(selector & ~3))
 848                 return 0;
 849
 850         table_base = gdt->address;
 851
 852         if (selector & 4) {           /* from ldt */
 853                 u16 ldt_selector = kvm_read_ldt();
 854
 855                 if (!(ldt_selector & ~3))
 856                         return 0;
 857
 858                 table_base = segment_base(ldt_selector);
 859         }
 860         d = (struct desc_struct *)(table_base + (selector & ~7));
 861         v = get_desc_base(d);
 862 #ifdef CONFIG_X86_64
 863        if (d->s == 0 && (d->type == 2 || d->type == 9 || d->type == 11))
 864                v |= ((unsigned long)((struct ldttss_desc64 *)d)->base3) << 32;
 865 #endif
 866         return v;
 867 }
 868
 869 static inline unsigned long kvm_read_tr_base(void)
 870 {
 871         u16 tr;
 872         asm("str %0" : "=g"(tr));
 873         return segment_base(tr);
 874 }
 875
 876 static void vmx_save_host_state(struct kvm_vcpu *vcpu)
 877 {
 878         struct vcpu_vmx *vmx = to_vmx(vcpu);
 879         int i;
 880
 881         if (vmx->host_state.loaded)
 882                 return;
 883
 884         vmx->host_state.loaded = 1;
 885         /*
 886          * Set host fs and gs selectors.  Unfortunately, 22.2.3 does not
 887          * allow segment selectors with cpl > 0 or ti == 1.
 888          */
 889         vmx->host_state.ldt_sel = kvm_read_ldt();
 890         vmx->host_state.gs_ldt_reload_needed = vmx->host_state.ldt_sel;
 891         savesegment(fs, vmx->host_state.fs_sel);
 892         if (!(vmx->host_state.fs_sel & 7)) {
 893                 vmcs_write16(HOST_FS_SELECTOR, vmx->host_state.fs_sel);
 894                 vmx->host_state.fs_reload_needed = 0;
 895         } else {
 896                 vmcs_write16(HOST_FS_SELECTOR, 0);
 897                 vmx->host_state.fs_reload_needed = 1;
 898         }
 899         savesegment(gs, vmx->host_state.gs_sel);
 900         if (!(vmx->host_state.gs_sel & 7))
 901                 vmcs_write16(HOST_GS_SELECTOR, vmx->host_state.gs_sel);
 902         else {
 903                 vmcs_write16(HOST_GS_SELECTOR, 0);
 904                 vmx->host_state.gs_ldt_reload_needed = 1;
 905         }
 906
 907 #ifdef CONFIG_X86_64
 908         vmcs_writel(HOST_FS_BASE, read_msr(MSR_FS_BASE));
 909         vmcs_writel(HOST_GS_BASE, read_msr(MSR_GS_BASE));
 910 #else
 911         vmcs_writel(HOST_FS_BASE, segment_base(vmx->host_state.fs_sel));
 912         vmcs_writel(HOST_GS_BASE, segment_base(vmx->host_state.gs_sel));
 913 #endif
 914
 915 #ifdef CONFIG_X86_64
 916         rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
 917         if (is_long_mode(&vmx->vcpu))
 918                 wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
 919 #endif
 920         for (i = 0; i < vmx->save_nmsrs; ++i)
 921                 kvm_set_shared_msr(vmx->guest_msrs[i].index,
 922                                    vmx->guest_msrs[i].data,
 923                                    vmx->guest_msrs[i].mask);
 924 }
 925
 926 static void __vmx_load_host_state(struct vcpu_vmx *vmx)
 927 {
 928         if (!vmx->host_state.loaded)
 929                 return;
 930
 931         ++vmx->vcpu.stat.host_state_reload;
 932         vmx->host_state.loaded = 0;
 933 #ifdef CONFIG_X86_64
 934         if (is_long_mode(&vmx->vcpu))
 935                 rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
 936 #endif
 937         if (vmx->host_state.gs_ldt_reload_needed) {
 938                 kvm_load_ldt(vmx->host_state.ldt_sel);
 939 #ifdef CONFIG_X86_64
 940                 load_gs_index(vmx->host_state.gs_sel);
 941 #else
 942                 loadsegment(gs, vmx->host_state.gs_sel);
 943 #endif
 944         }
 945         if (vmx->host_state.fs_reload_needed)
 946                 loadsegment(fs, vmx->host_state.fs_sel);
 947         reload_tss();
 948 #ifdef CONFIG_X86_64
 949         wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
 950 #endif
 951         if (current_thread_info()->status & TS_USEDFPU)
 952                 clts();
 953         load_gdt(&__get_cpu_var(host_gdt));
 954 }
 955
 956 static void vmx_load_host_state(struct vcpu_vmx *vmx)
 957 {
 958         preempt_disable();
 959         __vmx_load_host_state(vmx);
 960         preempt_enable();
 961 }
 962
 963 /*
 964  * Switches to specified vcpu, until a matching vcpu_put(), but assumes
 965  * vcpu mutex is already taken.
 966  */
 967 static void vmx_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
 968 {
 969         struct vcpu_vmx *vmx = to_vmx(vcpu);
 970         u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
 971
 972         if (!vmm_exclusive)
 973                 kvm_cpu_vmxon(phys_addr);
 974         else if (vcpu->cpu != cpu)
 975                 vcpu_clear(vmx);
 976
 977         if (per_cpu(current_vmcs, cpu) != vmx->vmcs) {
 978                 per_cpu(current_vmcs, cpu) = vmx->vmcs;
 979                 vmcs_load(vmx->vmcs);
 980         }
 981
 982         if (vcpu->cpu != cpu) {
 983                 struct desc_ptr *gdt = &__get_cpu_var(host_gdt);
 984                 unsigned long sysenter_esp;
 985
 986                 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
 987                 local_irq_disable();
 988                 list_add(&vmx->local_vcpus_link,
 989                          &per_cpu(vcpus_on_cpu, cpu));
 990                 local_irq_enable();
 991
 992                 /*
 993                  * Linux uses per-cpu TSS and GDT, so set these when switching
 994                  * processors.
 995                  */
 996                 vmcs_writel(HOST_TR_BASE, kvm_read_tr_base()); /* 22.2.4 */
 997                 vmcs_writel(HOST_GDTR_BASE, gdt->address);   /* 22.2.4 */
 998
 999                 rdmsrl(MSR_IA32_SYSENTER_ESP, sysenter_esp);
1000                 vmcs_writel(HOST_IA32_SYSENTER_ESP, sysenter_esp); /* 22.2.3 */
1001         }
1002 }
1003
1004 static void vmx_vcpu_put(struct kvm_vcpu *vcpu)
1005 {
1006         __vmx_load_host_state(to_vmx(vcpu));
1007         if (!vmm_exclusive) {
1008                 __vcpu_clear(to_vmx(vcpu));
1009                 kvm_cpu_vmxoff();
1010         }
1011 }
1012
1013 static void vmx_fpu_activate(struct kvm_vcpu *vcpu)
1014 {
1015         ulong cr0;
1016
1017         if (vcpu->fpu_active)
1018                 return;
1019         vcpu->fpu_active = 1;
1020         cr0 = vmcs_readl(GUEST_CR0);
1021         cr0 &= ~(X86_CR0_TS | X86_CR0_MP);
1022         cr0 |= kvm_read_cr0_bits(vcpu, X86_CR0_TS | X86_CR0_MP);
1023         vmcs_writel(GUEST_CR0, cr0);
1024         update_exception_bitmap(vcpu);
1025         vcpu->arch.cr0_guest_owned_bits = X86_CR0_TS;
1026         vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
1027 }
1028
1029 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu);
1030
1031 static void vmx_fpu_deactivate(struct kvm_vcpu *vcpu)
1032 {
1033         vmx_decache_cr0_guest_bits(vcpu);
1034         vmcs_set_bits(GUEST_CR0, X86_CR0_TS | X86_CR0_MP);
1035         update_exception_bitmap(vcpu);
1036         vcpu->arch.cr0_guest_owned_bits = 0;
1037         vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
1038         vmcs_writel(CR0_READ_SHADOW, vcpu->arch.cr0);
1039 }
1040
1041 static unsigned long vmx_get_rflags(struct kvm_vcpu *vcpu)
1042 {
1043         unsigned long rflags, save_rflags;
1044
1045         if (!test_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail)) {
1046                 __set_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail);
1047                 rflags = vmcs_readl(GUEST_RFLAGS);
1048                 if (to_vmx(vcpu)->rmode.vm86_active) {
1049                         rflags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
1050                         save_rflags = to_vmx(vcpu)->rmode.save_rflags;
1051                         rflags |= save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
1052                 }
1053                 to_vmx(vcpu)->rflags = rflags;
1054         }
1055         return to_vmx(vcpu)->rflags;
1056 }
1057
1058 static void vmx_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
1059 {
1060         __set_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail);
1061         __clear_bit(VCPU_EXREG_CPL, (ulong *)&vcpu->arch.regs_avail);
1062         to_vmx(vcpu)->rflags = rflags;
1063         if (to_vmx(vcpu)->rmode.vm86_active) {
1064                 to_vmx(vcpu)->rmode.save_rflags = rflags;
1065                 rflags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
1066         }
1067         vmcs_writel(GUEST_RFLAGS, rflags);
1068 }
1069
1070 static u32 vmx_get_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
1071 {
1072         u32 interruptibility = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
1073         int ret = 0;
1074
1075         if (interruptibility & GUEST_INTR_STATE_STI)
1076                 ret |= KVM_X86_SHADOW_INT_STI;
1077         if (interruptibility & GUEST_INTR_STATE_MOV_SS)
1078                 ret |= KVM_X86_SHADOW_INT_MOV_SS;
1079
1080         return ret & mask;
1081 }
1082
1083 static void vmx_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
1084 {
1085         u32 interruptibility_old = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
1086         u32 interruptibility = interruptibility_old;
1087
1088         interruptibility &= ~(GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS);
1089
1090         if (mask & KVM_X86_SHADOW_INT_MOV_SS)
1091                 interruptibility |= GUEST_INTR_STATE_MOV_SS;
1092         else if (mask & KVM_X86_SHADOW_INT_STI)
1093                 interruptibility |= GUEST_INTR_STATE_STI;
1094
1095         if ((interruptibility != interruptibility_old))
1096                 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, interruptibility);
1097 }
1098
1099 static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
1100 {
1101         unsigned long rip;
1102
1103         rip = kvm_rip_read(vcpu);
1104         rip += vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
1105         kvm_rip_write(vcpu, rip);
1106
1107         /* skipping an emulated instruction also counts */
1108         vmx_set_interrupt_shadow(vcpu, 0);
1109 }
1110
1111 static void vmx_clear_hlt(struct kvm_vcpu *vcpu)
1112 {
1113         /* Ensure that we clear the HLT state in the VMCS.  We don't need to
1114          * explicitly skip the instruction because if the HLT state is set, then
1115          * the instruction is already executing and RIP has already been
1116          * advanced. */
1117         if (!yield_on_hlt &&
1118             vmcs_read32(GUEST_ACTIVITY_STATE) == GUEST_ACTIVITY_HLT)
1119                 vmcs_write32(GUEST_ACTIVITY_STATE, GUEST_ACTIVITY_ACTIVE);
1120 }
1121
1122 static void vmx_queue_exception(struct kvm_vcpu *vcpu, unsigned nr,
1123                                 bool has_error_code, u32 error_code,
1124                                 bool reinject)
1125 {
1126         struct vcpu_vmx *vmx = to_vmx(vcpu);
1127         u32 intr_info = nr | INTR_INFO_VALID_MASK;
1128
1129         if (has_error_code) {
1130                 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE, error_code);
1131                 intr_info |= INTR_INFO_DELIVER_CODE_MASK;
1132         }
1133
1134         if (vmx->rmode.vm86_active) {
1135                 int inc_eip = 0;
1136                 if (kvm_exception_is_soft(nr))
1137                         inc_eip = vcpu->arch.event_exit_inst_len;
1138                 if (kvm_inject_realmode_interrupt(vcpu, nr, inc_eip) != EMULATE_DONE)
1139                         kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
1140                 return;
1141         }
1142
1143         if (kvm_exception_is_soft(nr)) {
1144                 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
1145                              vmx->vcpu.arch.event_exit_inst_len);
1146                 intr_info |= INTR_TYPE_SOFT_EXCEPTION;
1147         } else
1148                 intr_info |= INTR_TYPE_HARD_EXCEPTION;
1149
1150         vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr_info);
1151         vmx_clear_hlt(vcpu);
1152 }
1153
1154 static bool vmx_rdtscp_supported(void)
1155 {
1156         return cpu_has_vmx_rdtscp();
1157 }
1158
1159 /*
1160  * Swap MSR entry in host/guest MSR entry array.
1161  */
1162 static void move_msr_up(struct vcpu_vmx *vmx, int from, int to)
1163 {
1164         struct shared_msr_entry tmp;
1165
1166         tmp = vmx->guest_msrs[to];
1167         vmx->guest_msrs[to] = vmx->guest_msrs[from];
1168         vmx->guest_msrs[from] = tmp;
1169 }
1170
1171 /*
1172  * Set up the vmcs to automatically save and restore system
1173  * msrs.  Don't touch the 64-bit msrs if the guest is in legacy
1174  * mode, as fiddling with msrs is very expensive.
1175  */
1176 static void setup_msrs(struct vcpu_vmx *vmx)
1177 {
1178         int save_nmsrs, index;
1179         unsigned long *msr_bitmap;
1180
1181         vmx_load_host_state(vmx);
1182         save_nmsrs = 0;
1183 #ifdef CONFIG_X86_64
1184         if (is_long_mode(&vmx->vcpu)) {
1185                 index = __find_msr_index(vmx, MSR_SYSCALL_MASK);
1186                 if (index >= 0)
1187                         move_msr_up(vmx, index, save_nmsrs++);
1188                 index = __find_msr_index(vmx, MSR_LSTAR);
1189                 if (index >= 0)
1190                         move_msr_up(vmx, index, save_nmsrs++);
1191                 index = __find_msr_index(vmx, MSR_CSTAR);
1192                 if (index >= 0)
1193                         move_msr_up(vmx, index, save_nmsrs++);
1194                 index = __find_msr_index(vmx, MSR_TSC_AUX);
1195                 if (index >= 0 && vmx->rdtscp_enabled)
1196                         move_msr_up(vmx, index, save_nmsrs++);
1197                 /*
1198                  * MSR_STAR is only needed on long mode guests, and only
1199                  * if efer.sce is enabled.
1200                  */
1201                 index = __find_msr_index(vmx, MSR_STAR);
1202                 if ((index >= 0) && (vmx->vcpu.arch.efer & EFER_SCE))
1203                         move_msr_up(vmx, index, save_nmsrs++);
1204         }
1205 #endif
1206         index = __find_msr_index(vmx, MSR_EFER);
1207         if (index >= 0 && update_transition_efer(vmx, index))
1208                 move_msr_up(vmx, index, save_nmsrs++);
1209
1210         vmx->save_nmsrs = save_nmsrs;
1211
1212         if (cpu_has_vmx_msr_bitmap()) {
1213                 if (is_long_mode(&vmx->vcpu))
1214                         msr_bitmap = vmx_msr_bitmap_longmode;
1215                 else
1216                         msr_bitmap = vmx_msr_bitmap_legacy;
1217
1218                 vmcs_write64(MSR_BITMAP, __pa(msr_bitmap));
1219         }
1220 }
1221
1222 /*
1223  * reads and returns guest's timestamp counter "register"
1224  * guest_tsc = host_tsc + tsc_offset    -- 21.3
1225  */
1226 static u64 guest_read_tsc(void)
1227 {
1228         u64 host_tsc, tsc_offset;
1229
1230         rdtscll(host_tsc);
1231         tsc_offset = vmcs_read64(TSC_OFFSET);
1232         return host_tsc + tsc_offset;
1233 }
1234
1235 /*
1236  * Empty call-back. Needs to be implemented when VMX enables the SET_TSC_KHZ
1237  * ioctl. In this case the call-back should update internal vmx state to make
1238  * the changes effective.
1239  */
1240 static void vmx_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz)
1241 {
1242         /* Nothing to do here */
1243 }
1244
1245 /*
1246  * writes 'offset' into guest's timestamp counter offset register
1247  */
1248 static void vmx_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
1249 {
1250         vmcs_write64(TSC_OFFSET, offset);
1251 }
1252
1253 static void vmx_adjust_tsc_offset(struct kvm_vcpu *vcpu, s64 adjustment)
1254 {
1255         u64 offset = vmcs_read64(TSC_OFFSET);
1256         vmcs_write64(TSC_OFFSET, offset + adjustment);
1257 }
1258
1259 static u64 vmx_compute_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
1260 {
1261         return target_tsc - native_read_tsc();
1262 }
1263
1264 /*
1265  * Reads an msr value (of 'msr_index') into 'pdata'.
1266  * Returns 0 on success, non-0 otherwise.
1267  * Assumes vcpu_load() was already called.
1268  */
1269 static int vmx_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1270 {
1271         u64 data;
1272         struct shared_msr_entry *msr;
1273
1274         if (!pdata) {
1275                 printk(KERN_ERR "BUG: get_msr called with NULL pdata\n");
1276                 return -EINVAL;
1277         }
1278
1279         switch (msr_index) {
1280 #ifdef CONFIG_X86_64
1281         case MSR_FS_BASE:
1282                 data = vmcs_readl(GUEST_FS_BASE);
1283                 break;
1284         case MSR_GS_BASE:
1285                 data = vmcs_readl(GUEST_GS_BASE);
1286                 break;
1287         case MSR_KERNEL_GS_BASE:
1288                 vmx_load_host_state(to_vmx(vcpu));
1289                 data = to_vmx(vcpu)->msr_guest_kernel_gs_base;
1290                 break;
1291 #endif
1292         case MSR_EFER:
1293                 return kvm_get_msr_common(vcpu, msr_index, pdata);
1294         case MSR_IA32_TSC:
1295                 data = guest_read_tsc();
1296                 break;
1297         case MSR_IA32_SYSENTER_CS:
1298                 data = vmcs_read32(GUEST_SYSENTER_CS);
1299                 break;
1300         case MSR_IA32_SYSENTER_EIP:
1301                 data = vmcs_readl(GUEST_SYSENTER_EIP);
1302                 break;
1303         case MSR_IA32_SYSENTER_ESP:
1304                 data = vmcs_readl(GUEST_SYSENTER_ESP);
1305                 break;
1306         case MSR_TSC_AUX:
1307                 if (!to_vmx(vcpu)->rdtscp_enabled)
1308                         return 1;
1309                 /* Otherwise falls through */
1310         default:
1311                 vmx_load_host_state(to_vmx(vcpu));
1312                 msr = find_msr_entry(to_vmx(vcpu), msr_index);
1313                 if (msr) {
1314                         vmx_load_host_state(to_vmx(vcpu));
1315                         data = msr->data;
1316                         break;
1317                 }
1318                 return kvm_get_msr_common(vcpu, msr_index, pdata);
1319         }
1320
1321         *pdata = data;
1322         return 0;
1323 }
1324
1325 /*
1326  * Writes msr value into into the appropriate "register".
1327  * Returns 0 on success, non-0 otherwise.
1328  * Assumes vcpu_load() was already called.
1329  */
1330 static int vmx_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
1331 {
1332         struct vcpu_vmx *vmx = to_vmx(vcpu);
1333         struct shared_msr_entry *msr;
1334         int ret = 0;
1335
1336         switch (msr_index) {
1337         case MSR_EFER:
1338                 vmx_load_host_state(vmx);
1339                 ret = kvm_set_msr_common(vcpu, msr_index, data);
1340                 break;
1341 #ifdef CONFIG_X86_64
1342         case MSR_FS_BASE:
1343                 vmx_segment_cache_clear(vmx);
1344                 vmcs_writel(GUEST_FS_BASE, data);
1345                 break;
1346         case MSR_GS_BASE:
1347                 vmx_segment_cache_clear(vmx);
1348                 vmcs_writel(GUEST_GS_BASE, data);
1349                 break;
1350         case MSR_KERNEL_GS_BASE:
1351                 vmx_load_host_state(vmx);
1352                 vmx->msr_guest_kernel_gs_base = data;
1353                 break;
1354 #endif
1355         case MSR_IA32_SYSENTER_CS:
1356                 vmcs_write32(GUEST_SYSENTER_CS, data);
1357                 break;
1358         case MSR_IA32_SYSENTER_EIP:
1359                 vmcs_writel(GUEST_SYSENTER_EIP, data);
1360                 break;
1361         case MSR_IA32_SYSENTER_ESP:
1362                 vmcs_writel(GUEST_SYSENTER_ESP, data);
1363                 break;
1364         case MSR_IA32_TSC:
1365                 kvm_write_tsc(vcpu, data);
1366                 break;
1367         case MSR_IA32_CR_PAT:
1368                 if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
1369                         vmcs_write64(GUEST_IA32_PAT, data);
1370                         vcpu->arch.pat = data;
1371                         break;
1372                 }
1373                 ret = kvm_set_msr_common(vcpu, msr_index, data);
1374                 break;
1375         case MSR_TSC_AUX:
1376                 if (!vmx->rdtscp_enabled)
1377                         return 1;
1378                 /* Check reserved bit, higher 32 bits should be zero */
1379                 if ((data >> 32) != 0)
1380                         return 1;
1381                 /* Otherwise falls through */
1382         default:
1383                 msr = find_msr_entry(vmx, msr_index);
1384                 if (msr) {
1385                         vmx_load_host_state(vmx);
1386                         msr->data = data;
1387                         break;
1388                 }
1389                 ret = kvm_set_msr_common(vcpu, msr_index, data);
1390         }
1391
1392         return ret;
1393 }
1394
1395 static void vmx_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
1396 {
1397         __set_bit(reg, (unsigned long *)&vcpu->arch.regs_avail);
1398         switch (reg) {
1399         case VCPU_REGS_RSP:
1400                 vcpu->arch.regs[VCPU_REGS_RSP] = vmcs_readl(GUEST_RSP);
1401                 break;
1402         case VCPU_REGS_RIP:
1403                 vcpu->arch.regs[VCPU_REGS_RIP] = vmcs_readl(GUEST_RIP);
1404                 break;
1405         case VCPU_EXREG_PDPTR:
1406                 if (enable_ept)
1407                         ept_save_pdptrs(vcpu);
1408                 break;
1409         default:
1410                 break;
1411         }
1412 }
1413
1414 static void set_guest_debug(struct kvm_vcpu *vcpu, struct kvm_guest_debug *dbg)
1415 {
1416         if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1417                 vmcs_writel(GUEST_DR7, dbg->arch.debugreg[7]);
1418         else
1419                 vmcs_writel(GUEST_DR7, vcpu->arch.dr7);
1420
1421         update_exception_bitmap(vcpu);
1422 }
1423
1424 static __init int cpu_has_kvm_support(void)
1425 {
1426         return cpu_has_vmx();
1427 }
1428
1429 static __init int vmx_disabled_by_bios(void)
1430 {
1431         u64 msr;
1432
1433         rdmsrl(MSR_IA32_FEATURE_CONTROL, msr);
1434         if (msr & FEATURE_CONTROL_LOCKED) {
1435                 /* launched w/ TXT and VMX disabled */
1436                 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX)
1437                         && tboot_enabled())
1438                         return 1;
1439                 /* launched w/o TXT and VMX only enabled w/ TXT */
1440                 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX)
1441                         && (msr & FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX)
1442                         && !tboot_enabled()) {
1443                         printk(KERN_WARNING "kvm: disable TXT in the BIOS or "
1444                                 "activate TXT before enabling KVM\n");
1445                         return 1;
1446                 }
1447                 /* launched w/o TXT and VMX disabled */
1448                 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX)
1449                         && !tboot_enabled())
1450                         return 1;
1451         }
1452
1453         return 0;
1454 }
1455
1456 static void kvm_cpu_vmxon(u64 addr)
1457 {
1458         asm volatile (ASM_VMX_VMXON_RAX
1459                         : : "a"(&addr), "m"(addr)
1460                         : "memory", "cc");
1461 }
1462
1463 static int hardware_enable(void *garbage)
1464 {
1465         int cpu = raw_smp_processor_id();
1466         u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
1467         u64 old, test_bits;
1468
1469         if (read_cr4() & X86_CR4_VMXE)
1470                 return -EBUSY;
1471
1472         INIT_LIST_HEAD(&per_cpu(vcpus_on_cpu, cpu));
1473         rdmsrl(MSR_IA32_FEATURE_CONTROL, old);
1474
1475         test_bits = FEATURE_CONTROL_LOCKED;
1476         test_bits |= FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
1477         if (tboot_enabled())
1478                 test_bits |= FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX;
1479
1480         if ((old & test_bits) != test_bits) {
1481                 /* enable and lock */
1482                 wrmsrl(MSR_IA32_FEATURE_CONTROL, old | test_bits);
1483         }
1484         write_cr4(read_cr4() | X86_CR4_VMXE); /* FIXME: not cpu hotplug safe */
1485
1486         if (vmm_exclusive) {
1487                 kvm_cpu_vmxon(phys_addr);
1488                 ept_sync_global();
1489         }
1490
1491         store_gdt(&__get_cpu_var(host_gdt));
1492
1493         return 0;
1494 }
1495
1496 static void vmclear_local_vcpus(void)
1497 {
1498         int cpu = raw_smp_processor_id();
1499         struct vcpu_vmx *vmx, *n;
1500
1501         list_for_each_entry_safe(vmx, n, &per_cpu(vcpus_on_cpu, cpu),
1502                                  local_vcpus_link)
1503                 __vcpu_clear(vmx);
1504 }
1505
1506
1507 /* Just like cpu_vmxoff(), but with the __kvm_handle_fault_on_reboot()
1508  * tricks.
1509  */
1510 static void kvm_cpu_vmxoff(void)
1511 {
1512         asm volatile (__ex(ASM_VMX_VMXOFF) : : : "cc");
1513 }
1514
1515 static void hardware_disable(void *garbage)
1516 {
1517         if (vmm_exclusive) {
1518                 vmclear_local_vcpus();
1519                 kvm_cpu_vmxoff();
1520         }
1521         write_cr4(read_cr4() & ~X86_CR4_VMXE);
1522 }
1523
1524 static __init int adjust_vmx_controls(u32 ctl_min, u32 ctl_opt,
1525                                       u32 msr, u32 *result)
1526 {
1527         u32 vmx_msr_low, vmx_msr_high;
1528         u32 ctl = ctl_min | ctl_opt;
1529
1530         rdmsr(msr, vmx_msr_low, vmx_msr_high);
1531
1532         ctl &= vmx_msr_high; /* bit == 0 in high word ==> must be zero */
1533         ctl |= vmx_msr_low;  /* bit == 1 in low word  ==> must be one  */
1534
1535         /* Ensure minimum (required) set of control bits are supported. */
1536         if (ctl_min & ~ctl)
1537                 return -EIO;
1538
1539         *result = ctl;
1540         return 0;
1541 }
1542
1543 static __init bool allow_1_setting(u32 msr, u32 ctl)
1544 {
1545         u32 vmx_msr_low, vmx_msr_high;
1546
1547         rdmsr(msr, vmx_msr_low, vmx_msr_high);
1548         return vmx_msr_high & ctl;
1549 }
1550
1551 static __init int setup_vmcs_config(struct vmcs_config *vmcs_conf)
1552 {
1553         u32 vmx_msr_low, vmx_msr_high;
1554         u32 min, opt, min2, opt2;
1555         u32 _pin_based_exec_control = 0;
1556         u32 _cpu_based_exec_control = 0;
1557         u32 _cpu_based_2nd_exec_control = 0;
1558         u32 _vmexit_control = 0;
1559         u32 _vmentry_control = 0;
1560
1561         min = PIN_BASED_EXT_INTR_MASK | PIN_BASED_NMI_EXITING;
1562         opt = PIN_BASED_VIRTUAL_NMIS;
1563         if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PINBASED_CTLS,
1564                                 &_pin_based_exec_control) < 0)
1565                 return -EIO;
1566
1567         min =
1568 #ifdef CONFIG_X86_64
1569               CPU_BASED_CR8_LOAD_EXITING |
1570               CPU_BASED_CR8_STORE_EXITING |
1571 #endif
1572               CPU_BASED_CR3_LOAD_EXITING |
1573               CPU_BASED_CR3_STORE_EXITING |
1574               CPU_BASED_USE_IO_BITMAPS |
1575               CPU_BASED_MOV_DR_EXITING |
1576               CPU_BASED_USE_TSC_OFFSETING |
1577               CPU_BASED_MWAIT_EXITING |
1578               CPU_BASED_MONITOR_EXITING |
1579               CPU_BASED_INVLPG_EXITING;
1580
1581         if (yield_on_hlt)
1582                 min |= CPU_BASED_HLT_EXITING;
1583
1584         opt = CPU_BASED_TPR_SHADOW |
1585               CPU_BASED_USE_MSR_BITMAPS |
1586               CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
1587         if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PROCBASED_CTLS,
1588                                 &_cpu_based_exec_control) < 0)
1589                 return -EIO;
1590 #ifdef CONFIG_X86_64
1591         if ((_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
1592                 _cpu_based_exec_control &= ~CPU_BASED_CR8_LOAD_EXITING &
1593                                            ~CPU_BASED_CR8_STORE_EXITING;
1594 #endif
1595         if (_cpu_based_exec_control & CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) {
1596                 min2 = 0;
1597                 opt2 = SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
1598                         SECONDARY_EXEC_WBINVD_EXITING |
1599                         SECONDARY_EXEC_ENABLE_VPID |
1600                         SECONDARY_EXEC_ENABLE_EPT |
1601                         SECONDARY_EXEC_UNRESTRICTED_GUEST |
1602                         SECONDARY_EXEC_PAUSE_LOOP_EXITING |
1603                         SECONDARY_EXEC_RDTSCP;
1604                 if (adjust_vmx_controls(min2, opt2,
1605                                         MSR_IA32_VMX_PROCBASED_CTLS2,
1606                                         &_cpu_based_2nd_exec_control) < 0)
1607                         return -EIO;
1608         }
1609 #ifndef CONFIG_X86_64
1610         if (!(_cpu_based_2nd_exec_control &
1611                                 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
1612                 _cpu_based_exec_control &= ~CPU_BASED_TPR_SHADOW;
1613 #endif
1614         if (_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_EPT) {
1615                 /* CR3 accesses and invlpg don't need to cause VM Exits when EPT
1616                    enabled */
1617                 _cpu_based_exec_control &= ~(CPU_BASED_CR3_LOAD_EXITING |
1618                                              CPU_BASED_CR3_STORE_EXITING |
1619                                              CPU_BASED_INVLPG_EXITING);
1620                 rdmsr(MSR_IA32_VMX_EPT_VPID_CAP,
1621                       vmx_capability.ept, vmx_capability.vpid);
1622         }
1623
1624         min = 0;
1625 #ifdef CONFIG_X86_64
1626         min |= VM_EXIT_HOST_ADDR_SPACE_SIZE;
1627 #endif
1628         opt = VM_EXIT_SAVE_IA32_PAT | VM_EXIT_LOAD_IA32_PAT;
1629         if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_EXIT_CTLS,
1630                                 &_vmexit_control) < 0)
1631                 return -EIO;
1632
1633         min = 0;
1634         opt = VM_ENTRY_LOAD_IA32_PAT;
1635         if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_ENTRY_CTLS,
1636                                 &_vmentry_control) < 0)
1637                 return -EIO;
1638
1639         rdmsr(MSR_IA32_VMX_BASIC, vmx_msr_low, vmx_msr_high);
1640
1641         /* IA-32 SDM Vol 3B: VMCS size is never greater than 4kB. */
1642         if ((vmx_msr_high & 0x1fff) > PAGE_SIZE)
1643                 return -EIO;
1644
1645 #ifdef CONFIG_X86_64
1646         /* IA-32 SDM Vol 3B: 64-bit CPUs always have VMX_BASIC_MSR[48]==0. */
1647         if (vmx_msr_high & (1u<<16))
1648                 return -EIO;
1649 #endif
1650
1651         /* Require Write-Back (WB) memory type for VMCS accesses. */
1652         if (((vmx_msr_high >> 18) & 15) != 6)
1653                 return -EIO;
1654
1655         vmcs_conf->size = vmx_msr_high & 0x1fff;
1656         vmcs_conf->order = get_order(vmcs_config.size);
1657         vmcs_conf->revision_id = vmx_msr_low;
1658
1659         vmcs_conf->pin_based_exec_ctrl = _pin_based_exec_control;
1660         vmcs_conf->cpu_based_exec_ctrl = _cpu_based_exec_control;
1661         vmcs_conf->cpu_based_2nd_exec_ctrl = _cpu_based_2nd_exec_control;
1662         vmcs_conf->vmexit_ctrl         = _vmexit_control;
1663         vmcs_conf->vmentry_ctrl        = _vmentry_control;
1664
1665         cpu_has_load_ia32_efer =
1666                 allow_1_setting(MSR_IA32_VMX_ENTRY_CTLS,
1667                                 VM_ENTRY_LOAD_IA32_EFER)
1668                 && allow_1_setting(MSR_IA32_VMX_EXIT_CTLS,
1669                                    VM_EXIT_LOAD_IA32_EFER);
1670
1671         return 0;
1672 }
1673
1674 static struct vmcs *alloc_vmcs_cpu(int cpu)
1675 {
1676         int node = cpu_to_node(cpu);
1677         struct page *pages;
1678         struct vmcs *vmcs;
1679
1680         pages = alloc_pages_exact_node(node, GFP_KERNEL, vmcs_config.order);
1681         if (!pages)
1682                 return NULL;
1683         vmcs = page_address(pages);
1684         memset(vmcs, 0, vmcs_config.size);
1685         vmcs->revision_id = vmcs_config.revision_id; /* vmcs revision id */
1686         return vmcs;
1687 }
1688
1689 static struct vmcs *alloc_vmcs(void)
1690 {
1691         return alloc_vmcs_cpu(raw_smp_processor_id());
1692 }
1693
1694 static void free_vmcs(struct vmcs *vmcs)
1695 {
1696         free_pages((unsigned long)vmcs, vmcs_config.order);
1697 }
1698
1699 static void free_kvm_area(void)
1700 {
1701         int cpu;
1702
1703         for_each_possible_cpu(cpu) {
1704                 free_vmcs(per_cpu(vmxarea, cpu));
1705                 per_cpu(vmxarea, cpu) = NULL;
1706         }
1707 }
1708
1709 static __init int alloc_kvm_area(void)
1710 {
1711         int cpu;
1712
1713         for_each_possible_cpu(cpu) {
1714                 struct vmcs *vmcs;
1715
1716                 vmcs = alloc_vmcs_cpu(cpu);
1717                 if (!vmcs) {
1718                         free_kvm_area();
1719                         return -ENOMEM;
1720                 }
1721
1722                 per_cpu(vmxarea, cpu) = vmcs;
1723         }
1724         return 0;
1725 }
1726
1727 static __init int hardware_setup(void)
1728 {
1729         if (setup_vmcs_config(&vmcs_config) < 0)
1730                 return -EIO;
1731
1732         if (boot_cpu_has(X86_FEATURE_NX))
1733                 kvm_enable_efer_bits(EFER_NX);
1734
1735         if (!cpu_has_vmx_vpid())
1736                 enable_vpid = 0;
1737
1738         if (!cpu_has_vmx_ept() ||
1739             !cpu_has_vmx_ept_4levels()) {
1740                 enable_ept = 0;
1741                 enable_unrestricted_guest = 0;
1742         }
1743
1744         if (!cpu_has_vmx_unrestricted_guest())
1745                 enable_unrestricted_guest = 0;
1746
1747         if (!cpu_has_vmx_flexpriority())
1748                 flexpriority_enabled = 0;
1749
1750         if (!cpu_has_vmx_tpr_shadow())
1751                 kvm_x86_ops->update_cr8_intercept = NULL;
1752
1753         if (enable_ept && !cpu_has_vmx_ept_2m_page())
1754                 kvm_disable_largepages();
1755
1756         if (!cpu_has_vmx_ple())
1757                 ple_gap = 0;
1758
1759         return alloc_kvm_area();
1760 }
1761
1762 static __exit void hardware_unsetup(void)
1763 {
1764         free_kvm_area();
1765 }
1766
1767 static void fix_pmode_dataseg(int seg, struct kvm_save_segment *save)
1768 {
1769         struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1770
1771         if (vmcs_readl(sf->base) == save->base && (save->base & AR_S_MASK)) {
1772                 vmcs_write16(sf->selector, save->selector);
1773                 vmcs_writel(sf->base, save->base);
1774                 vmcs_write32(sf->limit, save->limit);
1775                 vmcs_write32(sf->ar_bytes, save->ar);
1776         } else {
1777                 u32 dpl = (vmcs_read16(sf->selector) & SELECTOR_RPL_MASK)
1778                         << AR_DPL_SHIFT;
1779                 vmcs_write32(sf->ar_bytes, 0x93 | dpl);
1780         }
1781 }
1782
1783 static void enter_pmode(struct kvm_vcpu *vcpu)
1784 {
1785         unsigned long flags;
1786         struct vcpu_vmx *vmx = to_vmx(vcpu);
1787
1788         vmx->emulation_required = 1;
1789         vmx->rmode.vm86_active = 0;
1790
1791         vmx_segment_cache_clear(vmx);
1792
1793         vmcs_write16(GUEST_TR_SELECTOR, vmx->rmode.tr.selector);
1794         vmcs_writel(GUEST_TR_BASE, vmx->rmode.tr.base);
1795         vmcs_write32(GUEST_TR_LIMIT, vmx->rmode.tr.limit);
1796         vmcs_write32(GUEST_TR_AR_BYTES, vmx->rmode.tr.ar);
1797
1798         flags = vmcs_readl(GUEST_RFLAGS);
1799         flags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
1800         flags |= vmx->rmode.save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
1801         vmcs_writel(GUEST_RFLAGS, flags);
1802
1803         vmcs_writel(GUEST_CR4, (vmcs_readl(GUEST_CR4) & ~X86_CR4_VME) |
1804                         (vmcs_readl(CR4_READ_SHADOW) & X86_CR4_VME));
1805
1806         update_exception_bitmap(vcpu);
1807
1808         if (emulate_invalid_guest_state)
1809                 return;
1810
1811         fix_pmode_dataseg(VCPU_SREG_ES, &vmx->rmode.es);
1812         fix_pmode_dataseg(VCPU_SREG_DS, &vmx->rmode.ds);
1813         fix_pmode_dataseg(VCPU_SREG_GS, &vmx->rmode.gs);
1814         fix_pmode_dataseg(VCPU_SREG_FS, &vmx->rmode.fs);
1815
1816         vmx_segment_cache_clear(vmx);
1817
1818         vmcs_write16(GUEST_SS_SELECTOR, 0);
1819         vmcs_write32(GUEST_SS_AR_BYTES, 0x93);
1820
1821         vmcs_write16(GUEST_CS_SELECTOR,
1822                      vmcs_read16(GUEST_CS_SELECTOR) & ~SELECTOR_RPL_MASK);
1823         vmcs_write32(GUEST_CS_AR_BYTES, 0x9b);
1824 }
1825
1826 static gva_t rmode_tss_base(struct kvm *kvm)
1827 {
1828         if (!kvm->arch.tss_addr) {
1829                 struct kvm_memslots *slots;
1830                 gfn_t base_gfn;
1831
1832                 slots = kvm_memslots(kvm);
1833                 base_gfn = slots->memslots[0].base_gfn +
1834                                  kvm->memslots->memslots[0].npages - 3;
1835                 return base_gfn << PAGE_SHIFT;
1836         }
1837         return kvm->arch.tss_addr;
1838 }
1839
1840 static void fix_rmode_seg(int seg, struct kvm_save_segment *save)
1841 {
1842         struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1843
1844         save->selector = vmcs_read16(sf->selector);
1845         save->base = vmcs_readl(sf->base);
1846         save->limit = vmcs_read32(sf->limit);
1847         save->ar = vmcs_read32(sf->ar_bytes);
1848         vmcs_write16(sf->selector, save->base >> 4);
1849         vmcs_write32(sf->base, save->base & 0xffff0);
1850         vmcs_write32(sf->limit, 0xffff);
1851         vmcs_write32(sf->ar_bytes, 0xf3);
1852         if (save->base & 0xf)
1853                 printk_once(KERN_WARNING "kvm: segment base is not paragraph"
1854                             " aligned when entering protected mode (seg=%d)",
1855                             seg);
1856 }
1857
1858 static void enter_rmode(struct kvm_vcpu *vcpu)
1859 {
1860         unsigned long flags;
1861         struct vcpu_vmx *vmx = to_vmx(vcpu);
1862
1863         if (enable_unrestricted_guest)
1864                 return;
1865
1866         vmx->emulation_required = 1;
1867         vmx->rmode.vm86_active = 1;
1868
1869         /*
1870          * Very old userspace does not call KVM_SET_TSS_ADDR before entering
1871          * vcpu. Call it here with phys address pointing 16M below 4G.
1872          */
1873         if (!vcpu->kvm->arch.tss_addr) {
1874                 printk_once(KERN_WARNING "kvm: KVM_SET_TSS_ADDR need to be "
1875                              "called before entering vcpu\n");
1876                 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
1877                 vmx_set_tss_addr(vcpu->kvm, 0xfeffd000);
1878                 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1879         }
1880
1881         vmx_segment_cache_clear(vmx);
1882
1883         vmx->rmode.tr.selector = vmcs_read16(GUEST_TR_SELECTOR);
1884         vmx->rmode.tr.base = vmcs_readl(GUEST_TR_BASE);
1885         vmcs_writel(GUEST_TR_BASE, rmode_tss_base(vcpu->kvm));
1886
1887         vmx->rmode.tr.limit = vmcs_read32(GUEST_TR_LIMIT);
1888         vmcs_write32(GUEST_TR_LIMIT, RMODE_TSS_SIZE - 1);
1889
1890         vmx->rmode.tr.ar = vmcs_read32(GUEST_TR_AR_BYTES);
1891         vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
1892
1893         flags = vmcs_readl(GUEST_RFLAGS);
1894         vmx->rmode.save_rflags = flags;
1895
1896         flags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
1897
1898         vmcs_writel(GUEST_RFLAGS, flags);
1899         vmcs_writel(GUEST_CR4, vmcs_readl(GUEST_CR4) | X86_CR4_VME);
1900         update_exception_bitmap(vcpu);
1901
1902         if (emulate_invalid_guest_state)
1903                 goto continue_rmode;
1904
1905         vmcs_write16(GUEST_SS_SELECTOR, vmcs_readl(GUEST_SS_BASE) >> 4);
1906         vmcs_write32(GUEST_SS_LIMIT, 0xffff);
1907         vmcs_write32(GUEST_SS_AR_BYTES, 0xf3);
1908
1909         vmcs_write32(GUEST_CS_AR_BYTES, 0xf3);
1910         vmcs_write32(GUEST_CS_LIMIT, 0xffff);
1911         if (vmcs_readl(GUEST_CS_BASE) == 0xffff0000)
1912                 vmcs_writel(GUEST_CS_BASE, 0xf0000);
1913         vmcs_write16(GUEST_CS_SELECTOR, vmcs_readl(GUEST_CS_BASE) >> 4);
1914
1915         fix_rmode_seg(VCPU_SREG_ES, &vmx->rmode.es);
1916         fix_rmode_seg(VCPU_SREG_DS, &vmx->rmode.ds);
1917         fix_rmode_seg(VCPU_SREG_GS, &vmx->rmode.gs);
1918         fix_rmode_seg(VCPU_SREG_FS, &vmx->rmode.fs);
1919
1920 continue_rmode:
1921         kvm_mmu_reset_context(vcpu);
1922 }
1923
1924 static void vmx_set_efer(struct kvm_vcpu *vcpu, u64 efer)
1925 {
1926         struct vcpu_vmx *vmx = to_vmx(vcpu);
1927         struct shared_msr_entry *msr = find_msr_entry(vmx, MSR_EFER);
1928
1929         if (!msr)
1930                 return;
1931
1932         /*
1933          * Force kernel_gs_base reloading before EFER changes, as control
1934          * of this msr depends on is_long_mode().
1935          */
1936         vmx_load_host_state(to_vmx(vcpu));
1937         vcpu->arch.efer = efer;
1938         if (efer & EFER_LMA) {
1939                 vmcs_write32(VM_ENTRY_CONTROLS,
1940                              vmcs_read32(VM_ENTRY_CONTROLS) |
1941                              VM_ENTRY_IA32E_MODE);
1942                 msr->data = efer;
1943         } else {
1944                 vmcs_write32(VM_ENTRY_CONTROLS,
1945                              vmcs_read32(VM_ENTRY_CONTROLS) &
1946                              ~VM_ENTRY_IA32E_MODE);
1947
1948                 msr->data = efer & ~EFER_LME;
1949         }
1950         setup_msrs(vmx);
1951 }
1952
1953 #ifdef CONFIG_X86_64
1954
1955 static void enter_lmode(struct kvm_vcpu *vcpu)
1956 {
1957         u32 guest_tr_ar;
1958
1959         vmx_segment_cache_clear(to_vmx(vcpu));
1960
1961         guest_tr_ar = vmcs_read32(GUEST_TR_AR_BYTES);
1962         if ((guest_tr_ar & AR_TYPE_MASK) != AR_TYPE_BUSY_64_TSS) {
1963                 printk(KERN_DEBUG "%s: tss fixup for long mode. \n",
1964                        __func__);
1965                 vmcs_write32(GUEST_TR_AR_BYTES,
1966                              (guest_tr_ar & ~AR_TYPE_MASK)
1967                              | AR_TYPE_BUSY_64_TSS);
1968         }
1969         vmx_set_efer(vcpu, vcpu->arch.efer | EFER_LMA);
1970 }
1971
1972 static void exit_lmode(struct kvm_vcpu *vcpu)
1973 {
1974         vmcs_write32(VM_ENTRY_CONTROLS,
1975                      vmcs_read32(VM_ENTRY_CONTROLS)
1976                      & ~VM_ENTRY_IA32E_MODE);
1977         vmx_set_efer(vcpu, vcpu->arch.efer & ~EFER_LMA);
1978 }
1979
1980 #endif
1981
1982 static void vmx_flush_tlb(struct kvm_vcpu *vcpu)
1983 {
1984         vpid_sync_context(to_vmx(vcpu));
1985         if (enable_ept) {
1986                 if (!VALID_PAGE(vcpu->arch.mmu.root_hpa))
1987                         return;
1988                 ept_sync_context(construct_eptp(vcpu->arch.mmu.root_hpa));
1989         }
1990 }
1991
1992 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu)
1993 {
1994         ulong cr0_guest_owned_bits = vcpu->arch.cr0_guest_owned_bits;
1995
1996         vcpu->arch.cr0 &= ~cr0_guest_owned_bits;
1997         vcpu->arch.cr0 |= vmcs_readl(GUEST_CR0) & cr0_guest_owned_bits;
1998 }
1999
2000 static void vmx_decache_cr3(struct kvm_vcpu *vcpu)
2001 {
2002         if (enable_ept && is_paging(vcpu))
2003                 vcpu->arch.cr3 = vmcs_readl(GUEST_CR3);
2004         __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
2005 }
2006
2007 static void vmx_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
2008 {
2009         ulong cr4_guest_owned_bits = vcpu->arch.cr4_guest_owned_bits;
2010
2011         vcpu->arch.cr4 &= ~cr4_guest_owned_bits;
2012         vcpu->arch.cr4 |= vmcs_readl(GUEST_CR4) & cr4_guest_owned_bits;
2013 }
2014
2015 static void ept_load_pdptrs(struct kvm_vcpu *vcpu)
2016 {
2017         if (!test_bit(VCPU_EXREG_PDPTR,
2018                       (unsigned long *)&vcpu->arch.regs_dirty))
2019                 return;
2020
2021         if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
2022                 vmcs_write64(GUEST_PDPTR0, vcpu->arch.mmu.pdptrs[0]);
2023                 vmcs_write64(GUEST_PDPTR1, vcpu->arch.mmu.pdptrs[1]);
2024                 vmcs_write64(GUEST_PDPTR2, vcpu->arch.mmu.pdptrs[2]);
2025                 vmcs_write64(GUEST_PDPTR3, vcpu->arch.mmu.pdptrs[3]);
2026         }
2027 }
2028
2029 static void ept_save_pdptrs(struct kvm_vcpu *vcpu)
2030 {
2031         if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
2032                 vcpu->arch.mmu.pdptrs[0] = vmcs_read64(GUEST_PDPTR0);
2033                 vcpu->arch.mmu.pdptrs[1] = vmcs_read64(GUEST_PDPTR1);
2034                 vcpu->arch.mmu.pdptrs[2] = vmcs_read64(GUEST_PDPTR2);
2035                 vcpu->arch.mmu.pdptrs[3] = vmcs_read64(GUEST_PDPTR3);
2036         }
2037
2038         __set_bit(VCPU_EXREG_PDPTR,
2039                   (unsigned long *)&vcpu->arch.regs_avail);
2040         __set_bit(VCPU_EXREG_PDPTR,
2041                   (unsigned long *)&vcpu->arch.regs_dirty);
2042 }
2043
2044 static void vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4);
2045
2046 static void ept_update_paging_mode_cr0(unsigned long *hw_cr0,
2047                                         unsigned long cr0,
2048                                         struct kvm_vcpu *vcpu)
2049 {
2050         if (!test_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail))
2051                 vmx_decache_cr3(vcpu);
2052         if (!(cr0 & X86_CR0_PG)) {
2053                 /* From paging/starting to nonpaging */
2054                 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
2055                              vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) |
2056                              (CPU_BASED_CR3_LOAD_EXITING |
2057                               CPU_BASED_CR3_STORE_EXITING));
2058                 vcpu->arch.cr0 = cr0;
2059                 vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
2060         } else if (!is_paging(vcpu)) {
2061                 /* From nonpaging to paging */
2062                 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
2063                              vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) &
2064                              ~(CPU_BASED_CR3_LOAD_EXITING |
2065                                CPU_BASED_CR3_STORE_EXITING));
2066                 vcpu->arch.cr0 = cr0;
2067                 vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
2068         }
2069
2070         if (!(cr0 & X86_CR0_WP))
2071                 *hw_cr0 &= ~X86_CR0_WP;
2072 }
2073
2074 static void vmx_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
2075 {
2076         struct vcpu_vmx *vmx = to_vmx(vcpu);
2077         unsigned long hw_cr0;
2078
2079         if (enable_unrestricted_guest)
2080                 hw_cr0 = (cr0 & ~KVM_GUEST_CR0_MASK_UNRESTRICTED_GUEST)
2081                         | KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST;
2082         else
2083                 hw_cr0 = (cr0 & ~KVM_GUEST_CR0_MASK) | KVM_VM_CR0_ALWAYS_ON;
2084
2085         if (vmx->rmode.vm86_active && (cr0 & X86_CR0_PE))
2086                 enter_pmode(vcpu);
2087
2088         if (!vmx->rmode.vm86_active && !(cr0 & X86_CR0_PE))
2089                 enter_rmode(vcpu);
2090
2091 #ifdef CONFIG_X86_64
2092         if (vcpu->arch.efer & EFER_LME) {
2093                 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG))
2094                         enter_lmode(vcpu);
2095                 if (is_paging(vcpu) && !(cr0 & X86_CR0_PG))
2096                         exit_lmode(vcpu);
2097         }
2098 #endif
2099
2100         if (enable_ept)
2101                 ept_update_paging_mode_cr0(&hw_cr0, cr0, vcpu);
2102
2103         if (!vcpu->fpu_active)
2104                 hw_cr0 |= X86_CR0_TS | X86_CR0_MP;
2105
2106         vmcs_writel(CR0_READ_SHADOW, cr0);
2107         vmcs_writel(GUEST_CR0, hw_cr0);
2108         vcpu->arch.cr0 = cr0;
2109         __clear_bit(VCPU_EXREG_CPL, (ulong *)&vcpu->arch.regs_avail);
2110 }
2111
2112 static u64 construct_eptp(unsigned long root_hpa)
2113 {
2114         u64 eptp;
2115
2116         /* TODO write the value reading from MSR */
2117         eptp = VMX_EPT_DEFAULT_MT |
2118                 VMX_EPT_DEFAULT_GAW << VMX_EPT_GAW_EPTP_SHIFT;
2119         eptp |= (root_hpa & PAGE_MASK);
2120
2121         return eptp;
2122 }
2123
2124 static void vmx_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
2125 {
2126         unsigned long guest_cr3;
2127         u64 eptp;
2128
2129         guest_cr3 = cr3;
2130         if (enable_ept) {
2131                 eptp = construct_eptp(cr3);
2132                 vmcs_write64(EPT_POINTER, eptp);
2133                 guest_cr3 = is_paging(vcpu) ? kvm_read_cr3(vcpu) :
2134                         vcpu->kvm->arch.ept_identity_map_addr;
2135                 ept_load_pdptrs(vcpu);
2136         }
2137
2138         vmx_flush_tlb(vcpu);
2139         vmcs_writel(GUEST_CR3, guest_cr3);
2140 }
2141
2142 static void vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
2143 {
2144         unsigned long hw_cr4 = cr4 | (to_vmx(vcpu)->rmode.vm86_active ?
2145                     KVM_RMODE_VM_CR4_ALWAYS_ON : KVM_PMODE_VM_CR4_ALWAYS_ON);
2146
2147         vcpu->arch.cr4 = cr4;
2148         if (enable_ept) {
2149                 if (!is_paging(vcpu)) {
2150                         hw_cr4 &= ~X86_CR4_PAE;
2151                         hw_cr4 |= X86_CR4_PSE;
2152                 } else if (!(cr4 & X86_CR4_PAE)) {
2153                         hw_cr4 &= ~X86_CR4_PAE;
2154                 }
2155         }
2156
2157         vmcs_writel(CR4_READ_SHADOW, cr4);
2158         vmcs_writel(GUEST_CR4, hw_cr4);
2159 }
2160
2161 static void vmx_get_segment(struct kvm_vcpu *vcpu,
2162                             struct kvm_segment *var, int seg)
2163 {
2164         struct vcpu_vmx *vmx = to_vmx(vcpu);
2165         struct kvm_save_segment *save;
2166         u32 ar;
2167
2168         if (vmx->rmode.vm86_active
2169             && (seg == VCPU_SREG_TR || seg == VCPU_SREG_ES
2170                 || seg == VCPU_SREG_DS || seg == VCPU_SREG_FS
2171                 || seg == VCPU_SREG_GS)
2172             && !emulate_invalid_guest_state) {
2173                 switch (seg) {
2174                 case VCPU_SREG_TR: save = &vmx->rmode.tr; break;
2175                 case VCPU_SREG_ES: save = &vmx->rmode.es; break;
2176                 case VCPU_SREG_DS: save = &vmx->rmode.ds; break;
2177                 case VCPU_SREG_FS: save = &vmx->rmode.fs; break;
2178                 case VCPU_SREG_GS: save = &vmx->rmode.gs; break;
2179                 default: BUG();
2180                 }
2181                 var->selector = save->selector;
2182                 var->base = save->base;
2183                 var->limit = save->limit;
2184                 ar = save->ar;
2185                 if (seg == VCPU_SREG_TR
2186                     || var->selector == vmx_read_guest_seg_selector(vmx, seg))
2187                         goto use_saved_rmode_seg;
2188         }
2189         var->base = vmx_read_guest_seg_base(vmx, seg);
2190         var->limit = vmx_read_guest_seg_limit(vmx, seg);
2191         var->selector = vmx_read_guest_seg_selector(vmx, seg);
2192         ar = vmx_read_guest_seg_ar(vmx, seg);
2193 use_saved_rmode_seg:
2194         if ((ar & AR_UNUSABLE_MASK) && !emulate_invalid_guest_state)
2195                 ar = 0;
2196         var->type = ar & 15;
2197         var->s = (ar >> 4) & 1;
2198         var->dpl = (ar >> 5) & 3;
2199         var->present = (ar >> 7) & 1;
2200         var->avl = (ar >> 12) & 1;
2201         var->l = (ar >> 13) & 1;
2202         var->db = (ar >> 14) & 1;
2203         var->g = (ar >> 15) & 1;
2204         var->unusable = (ar >> 16) & 1;
2205 }
2206
2207 static u64 vmx_get_segment_base(struct kvm_vcpu *vcpu, int seg)
2208 {
2209         struct kvm_segment s;
2210
2211         if (to_vmx(vcpu)->rmode.vm86_active) {
2212                 vmx_get_segment(vcpu, &s, seg);
2213                 return s.base;
2214         }
2215         return vmx_read_guest_seg_base(to_vmx(vcpu), seg);
2216 }
2217
2218 static int __vmx_get_cpl(struct kvm_vcpu *vcpu)
2219 {
2220         if (!is_protmode(vcpu))
2221                 return 0;
2222
2223         if (!is_long_mode(vcpu)
2224             && (kvm_get_rflags(vcpu) & X86_EFLAGS_VM)) /* if virtual 8086 */
2225                 return 3;
2226
2227         return vmx_read_guest_seg_selector(to_vmx(vcpu), VCPU_SREG_CS) & 3;
2228 }
2229
2230 static int vmx_get_cpl(struct kvm_vcpu *vcpu)
2231 {
2232         if (!test_bit(VCPU_EXREG_CPL, (ulong *)&vcpu->arch.regs_avail)) {
2233                 __set_bit(VCPU_EXREG_CPL, (ulong *)&vcpu->arch.regs_avail);
2234                 to_vmx(vcpu)->cpl = __vmx_get_cpl(vcpu);
2235         }
2236         return to_vmx(vcpu)->cpl;
2237 }
2238
2239
2240 static u32 vmx_segment_access_rights(struct kvm_segment *var)
2241 {
2242         u32 ar;
2243
2244         if (var->unusable)
2245                 ar = 1 << 16;
2246         else {
2247                 ar = var->type & 15;
2248                 ar |= (var->s & 1) << 4;
2249                 ar |= (var->dpl & 3) << 5;
2250                 ar |= (var->present & 1) << 7;
2251                 ar |= (var->avl & 1) << 12;
2252                 ar |= (var->l & 1) << 13;
2253                 ar |= (var->db & 1) << 14;
2254                 ar |= (var->g & 1) << 15;
2255         }
2256         if (ar == 0) /* a 0 value means unusable */
2257                 ar = AR_UNUSABLE_MASK;
2258
2259         return ar;
2260 }
2261
2262 static void vmx_set_segment(struct kvm_vcpu *vcpu,
2263                             struct kvm_segment *var, int seg)
2264 {
2265         struct vcpu_vmx *vmx = to_vmx(vcpu);
2266         struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
2267         u32 ar;
2268
2269         vmx_segment_cache_clear(vmx);
2270
2271         if (vmx->rmode.vm86_active && seg == VCPU_SREG_TR) {
2272                 vmcs_write16(sf->selector, var->selector);
2273                 vmx->rmode.tr.selector = var->selector;
2274                 vmx->rmode.tr.base = var->base;
2275                 vmx->rmode.tr.limit = var->limit;
2276                 vmx->rmode.tr.ar = vmx_segment_access_rights(var);
2277                 return;
2278         }
2279         vmcs_writel(sf->base, var->base);
2280         vmcs_write32(sf->limit, var->limit);
2281         vmcs_write16(sf->selector, var->selector);
2282         if (vmx->rmode.vm86_active && var->s) {
2283                 /*
2284                  * Hack real-mode segments into vm86 compatibility.
2285                  */
2286                 if (var->base == 0xffff0000 && var->selector == 0xf000)
2287                         vmcs_writel(sf->base, 0xf0000);
2288                 ar = 0xf3;
2289         } else
2290                 ar = vmx_segment_access_rights(var);
2291
2292         /*
2293          *   Fix the "Accessed" bit in AR field of segment registers for older
2294          * qemu binaries.
2295          *   IA32 arch specifies that at the time of processor reset the
2296          * "Accessed" bit in the AR field of segment registers is 1. And qemu
2297          * is setting it to 0 in the usedland code. This causes invalid guest
2298          * state vmexit when "unrestricted guest" mode is turned on.
2299          *    Fix for this setup issue in cpu_reset is being pushed in the qemu
2300          * tree. Newer qemu binaries with that qemu fix would not need this
2301          * kvm hack.
2302          */
2303         if (enable_unrestricted_guest && (seg != VCPU_SREG_LDTR))
2304                 ar |= 0x1; /* Accessed */
2305
2306         vmcs_write32(sf->ar_bytes, ar);
2307         __clear_bit(VCPU_EXREG_CPL, (ulong *)&vcpu->arch.regs_avail);
2308 }
2309
2310 static void vmx_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
2311 {
2312         u32 ar = vmx_read_guest_seg_ar(to_vmx(vcpu), VCPU_SREG_CS);
2313
2314         *db = (ar >> 14) & 1;
2315         *l = (ar >> 13) & 1;
2316 }
2317
2318 static void vmx_get_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
2319 {
2320         dt->size = vmcs_read32(GUEST_IDTR_LIMIT);
2321         dt->address = vmcs_readl(GUEST_IDTR_BASE);
2322 }
2323
2324 static void vmx_set_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
2325 {
2326         vmcs_write32(GUEST_IDTR_LIMIT, dt->size);
2327         vmcs_writel(GUEST_IDTR_BASE, dt->address);
2328 }
2329
2330 static void vmx_get_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
2331 {
2332         dt->size = vmcs_read32(GUEST_GDTR_LIMIT);
2333         dt->address = vmcs_readl(GUEST_GDTR_BASE);
2334 }
2335
2336 static void vmx_set_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
2337 {
2338         vmcs_write32(GUEST_GDTR_LIMIT, dt->size);
2339         vmcs_writel(GUEST_GDTR_BASE, dt->address);
2340 }
2341
2342 static bool rmode_segment_valid(struct kvm_vcpu *vcpu, int seg)
2343 {
2344         struct kvm_segment var;
2345         u32 ar;
2346
2347         vmx_get_segment(vcpu, &var, seg);
2348         ar = vmx_segment_access_rights(&var);
2349
2350         if (var.base != (var.selector << 4))
2351                 return false;
2352         if (var.limit != 0xffff)
2353                 return false;
2354         if (ar != 0xf3)
2355                 return false;
2356
2357         return true;
2358 }
2359
2360 static bool code_segment_valid(struct kvm_vcpu *vcpu)
2361 {
2362         struct kvm_segment cs;
2363         unsigned int cs_rpl;
2364
2365         vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
2366         cs_rpl = cs.selector & SELECTOR_RPL_MASK;
2367
2368         if (cs.unusable)
2369                 return false;
2370         if (~cs.type & (AR_TYPE_CODE_MASK|AR_TYPE_ACCESSES_MASK))
2371                 return false;
2372         if (!cs.s)
2373                 return false;
2374         if (cs.type & AR_TYPE_WRITEABLE_MASK) {
2375                 if (cs.dpl > cs_rpl)
2376                         return false;
2377         } else {
2378                 if (cs.dpl != cs_rpl)
2379                         return false;
2380         }
2381         if (!cs.present)
2382                 return false;
2383
2384         /* TODO: Add Reserved field check, this'll require a new member in the kvm_segment_field structure */
2385         return true;
2386 }
2387
2388 static bool stack_segment_valid(struct kvm_vcpu *vcpu)
2389 {
2390         struct kvm_segment ss;
2391         unsigned int ss_rpl;
2392
2393         vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
2394         ss_rpl = ss.selector & SELECTOR_RPL_MASK;
2395
2396         if (ss.unusable)
2397                 return true;
2398         if (ss.type != 3 && ss.type != 7)
2399                 return false;
2400         if (!ss.s)
2401                 return false;
2402         if (ss.dpl != ss_rpl) /* DPL != RPL */
2403                 return false;
2404         if (!ss.present)
2405                 return false;
2406
2407         return true;
2408 }
2409
2410 static bool data_segment_valid(struct kvm_vcpu *vcpu, int seg)
2411 {
2412         struct kvm_segment var;
2413         unsigned int rpl;
2414
2415         vmx_get_segment(vcpu, &var, seg);
2416         rpl = var.selector & SELECTOR_RPL_MASK;
2417
2418         if (var.unusable)
2419                 return true;
2420         if (!var.s)
2421                 return false;
2422         if (!var.present)
2423                 return false;
2424         if (~var.type & (AR_TYPE_CODE_MASK|AR_TYPE_WRITEABLE_MASK)) {
2425                 if (var.dpl < rpl) /* DPL < RPL */
2426                         return false;
2427         }
2428
2429         /* TODO: Add other members to kvm_segment_field to allow checking for other access
2430          * rights flags
2431          */
2432         return true;
2433 }
2434
2435 static bool tr_valid(struct kvm_vcpu *vcpu)
2436 {
2437         struct kvm_segment tr;
2438
2439         vmx_get_segment(vcpu, &tr, VCPU_SREG_TR);
2440
2441         if (tr.unusable)
2442                 return false;
2443         if (tr.selector & SELECTOR_TI_MASK)     /* TI = 1 */
2444                 return false;
2445         if (tr.type != 3 && tr.type != 11) /* TODO: Check if guest is in IA32e mode */
2446                 return false;
2447         if (!tr.present)
2448                 return false;
2449
2450         return true;
2451 }
2452
2453 static bool ldtr_valid(struct kvm_vcpu *vcpu)
2454 {
2455         struct kvm_segment ldtr;
2456
2457         vmx_get_segment(vcpu, &ldtr, VCPU_SREG_LDTR);
2458
2459         if (ldtr.unusable)
2460                 return true;
2461         if (ldtr.selector & SELECTOR_TI_MASK)   /* TI = 1 */
2462                 return false;
2463         if (ldtr.type != 2)
2464                 return false;
2465         if (!ldtr.present)
2466                 return false;
2467
2468         return true;
2469 }
2470
2471 static bool cs_ss_rpl_check(struct kvm_vcpu *vcpu)
2472 {
2473         struct kvm_segment cs, ss;
2474
2475         vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
2476         vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
2477
2478         return ((cs.selector & SELECTOR_RPL_MASK) ==
2479                  (ss.selector & SELECTOR_RPL_MASK));
2480 }
2481
2482 /*
2483  * Check if guest state is valid. Returns true if valid, false if
2484  * not.
2485  * We assume that registers are always usable
2486  */
2487 static bool guest_state_valid(struct kvm_vcpu *vcpu)
2488 {
2489         /* real mode guest state checks */
2490         if (!is_protmode(vcpu)) {
2491                 if (!rmode_segment_valid(vcpu, VCPU_SREG_CS))
2492                         return false;
2493                 if (!rmode_segment_valid(vcpu, VCPU_SREG_SS))
2494                         return false;
2495                 if (!rmode_segment_valid(vcpu, VCPU_SREG_DS))
2496                         return false;
2497                 if (!rmode_segment_valid(vcpu, VCPU_SREG_ES))
2498                         return false;
2499                 if (!rmode_segment_valid(vcpu, VCPU_SREG_FS))
2500                         return false;
2501                 if (!rmode_segment_valid(vcpu, VCPU_SREG_GS))
2502                         return false;
2503         } else {
2504         /* protected mode guest state checks */
2505                 if (!cs_ss_rpl_check(vcpu))
2506                         return false;
2507                 if (!code_segment_valid(vcpu))
2508                         return false;
2509                 if (!stack_segment_valid(vcpu))
2510                         return false;
2511                 if (!data_segment_valid(vcpu, VCPU_SREG_DS))
2512                         return false;
2513                 if (!data_segment_valid(vcpu, VCPU_SREG_ES))
2514                         return false;
2515                 if (!data_segment_valid(vcpu, VCPU_SREG_FS))
2516                         return false;
2517                 if (!data_segment_valid(vcpu, VCPU_SREG_GS))
2518                         return false;
2519                 if (!tr_valid(vcpu))
2520                         return false;
2521                 if (!ldtr_valid(vcpu))
2522                         return false;
2523         }
2524         /* TODO:
2525          * - Add checks on RIP
2526          * - Add checks on RFLAGS
2527          */
2528
2529         return true;
2530 }
2531
2532 static int init_rmode_tss(struct kvm *kvm)
2533 {
2534         gfn_t fn;
2535         u16 data = 0;
2536         int r, idx, ret = 0;
2537
2538         idx = srcu_read_lock(&kvm->srcu);
2539         fn = rmode_tss_base(kvm) >> PAGE_SHIFT;
2540         r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
2541         if (r < 0)
2542                 goto out;
2543         data = TSS_BASE_SIZE + TSS_REDIRECTION_SIZE;
2544         r = kvm_write_guest_page(kvm, fn++, &data,
2545                         TSS_IOPB_BASE_OFFSET, sizeof(u16));
2546         if (r < 0)
2547                 goto out;
2548         r = kvm_clear_guest_page(kvm, fn++, 0, PAGE_SIZE);
2549         if (r < 0)
2550                 goto out;
2551         r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
2552         if (r < 0)
2553                 goto out;
2554         data = ~0;
2555         r = kvm_write_guest_page(kvm, fn, &data,
2556                                  RMODE_TSS_SIZE - 2 * PAGE_SIZE - 1,
2557                                  sizeof(u8));
2558         if (r < 0)
2559                 goto out;
2560
2561         ret = 1;
2562 out:
2563         srcu_read_unlock(&kvm->srcu, idx);
2564         return ret;
2565 }
2566
2567 static int init_rmode_identity_map(struct kvm *kvm)
2568 {
2569         int i, idx, r, ret;
2570         pfn_t identity_map_pfn;
2571         u32 tmp;
2572
2573         if (!enable_ept)
2574                 return 1;
2575         if (unlikely(!kvm->arch.ept_identity_pagetable)) {
2576                 printk(KERN_ERR "EPT: identity-mapping pagetable "
2577                         "haven't been allocated!\n");
2578                 return 0;
2579         }
2580         if (likely(kvm->arch.ept_identity_pagetable_done))
2581                 return 1;
2582         ret = 0;
2583         identity_map_pfn = kvm->arch.ept_identity_map_addr >> PAGE_SHIFT;
2584         idx = srcu_read_lock(&kvm->srcu);
2585         r = kvm_clear_guest_page(kvm, identity_map_pfn, 0, PAGE_SIZE);
2586         if (r < 0)
2587                 goto out;
2588         /* Set up identity-mapping pagetable for EPT in real mode */
2589         for (i = 0; i < PT32_ENT_PER_PAGE; i++) {
2590                 tmp = (i << 22) + (_PAGE_PRESENT | _PAGE_RW | _PAGE_USER |
2591                         _PAGE_ACCESSED | _PAGE_DIRTY | _PAGE_PSE);
2592                 r = kvm_write_guest_page(kvm, identity_map_pfn,
2593                                 &tmp, i * sizeof(tmp), sizeof(tmp));
2594                 if (r < 0)
2595                         goto out;
2596         }
2597         kvm->arch.ept_identity_pagetable_done = true;
2598         ret = 1;
2599 out:
2600         srcu_read_unlock(&kvm->srcu, idx);
2601         return ret;
2602 }
2603
2604 static void seg_setup(int seg)
2605 {
2606         struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
2607         unsigned int ar;
2608
2609         vmcs_write16(sf->selector, 0);
2610         vmcs_writel(sf->base, 0);
2611         vmcs_write32(sf->limit, 0xffff);
2612         if (enable_unrestricted_guest) {
2613                 ar = 0x93;
2614                 if (seg == VCPU_SREG_CS)
2615                         ar |= 0x08; /* code segment */
2616         } else
2617                 ar = 0xf3;
2618
2619         vmcs_write32(sf->ar_bytes, ar);
2620 }
2621
2622 static int alloc_apic_access_page(struct kvm *kvm)
2623 {
2624         struct kvm_userspace_memory_region kvm_userspace_mem;
2625         int r = 0;
2626
2627         mutex_lock(&kvm->slots_lock);
2628         if (kvm->arch.apic_access_page)
2629                 goto out;
2630         kvm_userspace_mem.slot = APIC_ACCESS_PAGE_PRIVATE_MEMSLOT;
2631         kvm_userspace_mem.flags = 0;
2632         kvm_userspace_mem.guest_phys_addr = 0xfee00000ULL;
2633         kvm_userspace_mem.memory_size = PAGE_SIZE;
2634         r = __kvm_set_memory_region(kvm, &kvm_userspace_mem, 0);
2635         if (r)
2636                 goto out;
2637
2638         kvm->arch.apic_access_page = gfn_to_page(kvm, 0xfee00);
2639 out:
2640         mutex_unlock(&kvm->slots_lock);
2641         return r;
2642 }
2643
2644 static int alloc_identity_pagetable(struct kvm *kvm)
2645 {
2646         struct kvm_userspace_memory_region kvm_userspace_mem;
2647         int r = 0;
2648
2649         mutex_lock(&kvm->slots_lock);
2650         if (kvm->arch.ept_identity_pagetable)
2651                 goto out;
2652         kvm_userspace_mem.slot = IDENTITY_PAGETABLE_PRIVATE_MEMSLOT;
2653         kvm_userspace_mem.flags = 0;
2654         kvm_userspace_mem.guest_phys_addr =
2655                 kvm->arch.ept_identity_map_addr;
2656         kvm_userspace_mem.memory_size = PAGE_SIZE;
2657         r = __kvm_set_memory_region(kvm, &kvm_userspace_mem, 0);
2658         if (r)
2659                 goto out;
2660
2661         kvm->arch.ept_identity_pagetable = gfn_to_page(kvm,
2662                         kvm->arch.ept_identity_map_addr >> PAGE_SHIFT);
2663 out:
2664         mutex_unlock(&kvm->slots_lock);
2665         return r;
2666 }
2667
2668 static void allocate_vpid(struct vcpu_vmx *vmx)
2669 {
2670         int vpid;
2671
2672         vmx->vpid = 0;
2673         if (!enable_vpid)
2674                 return;
2675         spin_lock(&vmx_vpid_lock);
2676         vpid = find_first_zero_bit(vmx_vpid_bitmap, VMX_NR_VPIDS);
2677         if (vpid < VMX_NR_VPIDS) {
2678                 vmx->vpid = vpid;
2679                 __set_bit(vpid, vmx_vpid_bitmap);
2680         }
2681         spin_unlock(&vmx_vpid_lock);
2682 }
2683
2684 static void free_vpid(struct vcpu_vmx *vmx)
2685 {
2686         if (!enable_vpid)
2687                 return;
2688         spin_lock(&vmx_vpid_lock);
2689         if (vmx->vpid != 0)
2690                 __clear_bit(vmx->vpid, vmx_vpid_bitmap);
2691         spin_unlock(&vmx_vpid_lock);
2692 }
2693
2694 static void __vmx_disable_intercept_for_msr(unsigned long *msr_bitmap, u32 msr)
2695 {
2696         int f = sizeof(unsigned long);
2697
2698         if (!cpu_has_vmx_msr_bitmap())
2699                 return;
2700
2701         /*
2702          * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
2703          * have the write-low and read-high bitmap offsets the wrong way round.
2704          * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
2705          */
2706         if (msr <= 0x1fff) {
2707                 __clear_bit(msr, msr_bitmap + 0x000 / f); /* read-low */
2708                 __clear_bit(msr, msr_bitmap + 0x800 / f); /* write-low */
2709         } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
2710                 msr &= 0x1fff;
2711                 __clear_bit(msr, msr_bitmap + 0x400 / f); /* read-high */
2712                 __clear_bit(msr, msr_bitmap + 0xc00 / f); /* write-high */
2713         }
2714 }
2715
2716 static void vmx_disable_intercept_for_msr(u32 msr, bool longmode_only)
2717 {
2718         if (!longmode_only)
2719                 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_legacy, msr);
2720         __vmx_disable_intercept_for_msr(vmx_msr_bitmap_longmode, msr);
2721 }
2722
2723 /*
2724  * Sets up the vmcs for emulated real mode.
2725  */
2726 static int vmx_vcpu_setup(struct vcpu_vmx *vmx)
2727 {
2728         u32 host_sysenter_cs, msr_low, msr_high;
2729         u32 junk;
2730         u64 host_pat;
2731         unsigned long a;
2732         struct desc_ptr dt;
2733         int i;
2734         unsigned long kvm_vmx_return;
2735         u32 exec_control;
2736
2737         /* I/O */
2738         vmcs_write64(IO_BITMAP_A, __pa(vmx_io_bitmap_a));
2739         vmcs_write64(IO_BITMAP_B, __pa(vmx_io_bitmap_b));
2740
2741         if (cpu_has_vmx_msr_bitmap())
2742                 vmcs_write64(MSR_BITMAP, __pa(vmx_msr_bitmap_legacy));
2743
2744         vmcs_write64(VMCS_LINK_POINTER, -1ull); /* 22.3.1.5 */
2745
2746         /* Control */
2747         vmcs_write32(PIN_BASED_VM_EXEC_CONTROL,
2748                 vmcs_config.pin_based_exec_ctrl);
2749
2750         exec_control = vmcs_config.cpu_based_exec_ctrl;
2751         if (!vm_need_tpr_shadow(vmx->vcpu.kvm)) {
2752                 exec_control &= ~CPU_BASED_TPR_SHADOW;
2753 #ifdef CONFIG_X86_64
2754                 exec_control |= CPU_BASED_CR8_STORE_EXITING |
2755                                 CPU_BASED_CR8_LOAD_EXITING;
2756 #endif
2757         }
2758         if (!enable_ept)
2759                 exec_control |= CPU_BASED_CR3_STORE_EXITING |
2760                                 CPU_BASED_CR3_LOAD_EXITING  |
2761                                 CPU_BASED_INVLPG_EXITING;
2762         vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, exec_control);
2763
2764         if (cpu_has_secondary_exec_ctrls()) {
2765                 exec_control = vmcs_config.cpu_based_2nd_exec_ctrl;
2766                 if (!vm_need_virtualize_apic_accesses(vmx->vcpu.kvm))
2767                         exec_control &=
2768                                 ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
2769                 if (vmx->vpid == 0)
2770                         exec_control &= ~SECONDARY_EXEC_ENABLE_VPID;
2771                 if (!enable_ept) {
2772                         exec_control &= ~SECONDARY_EXEC_ENABLE_EPT;
2773                         enable_unrestricted_guest = 0;
2774                 }
2775                 if (!enable_unrestricted_guest)
2776                         exec_control &= ~SECONDARY_EXEC_UNRESTRICTED_GUEST;
2777                 if (!ple_gap)
2778                         exec_control &= ~SECONDARY_EXEC_PAUSE_LOOP_EXITING;
2779                 vmcs_write32(SECONDARY_VM_EXEC_CONTROL, exec_control);
2780         }
2781
2782         if (ple_gap) {
2783                 vmcs_write32(PLE_GAP, ple_gap);
2784                 vmcs_write32(PLE_WINDOW, ple_window);
2785         }
2786
2787         vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, !!bypass_guest_pf);
2788         vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, !!bypass_guest_pf);
2789         vmcs_write32(CR3_TARGET_COUNT, 0);           /* 22.2.1 */
2790
2791         vmcs_writel(HOST_CR0, read_cr0() | X86_CR0_TS);  /* 22.2.3 */
2792         vmcs_writel(HOST_CR4, read_cr4());  /* 22.2.3, 22.2.5 */
2793         vmcs_writel(HOST_CR3, read_cr3());  /* 22.2.3  FIXME: shadow tables */
2794
2795         vmcs_write16(HOST_CS_SELECTOR, __KERNEL_CS);  /* 22.2.4 */
2796         vmcs_write16(HOST_DS_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
2797         vmcs_write16(HOST_ES_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
2798         vmcs_write16(HOST_FS_SELECTOR, 0);            /* 22.2.4 */
2799         vmcs_write16(HOST_GS_SELECTOR, 0);            /* 22.2.4 */
2800         vmcs_write16(HOST_SS_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
2801 #ifdef CONFIG_X86_64
2802         rdmsrl(MSR_FS_BASE, a);
2803         vmcs_writel(HOST_FS_BASE, a); /* 22.2.4 */
2804         rdmsrl(MSR_GS_BASE, a);
2805         vmcs_writel(HOST_GS_BASE, a); /* 22.2.4 */
2806 #else
2807         vmcs_writel(HOST_FS_BASE, 0); /* 22.2.4 */
2808         vmcs_writel(HOST_GS_BASE, 0); /* 22.2.4 */
2809 #endif
2810
2811         vmcs_write16(HOST_TR_SELECTOR, GDT_ENTRY_TSS*8);  /* 22.2.4 */
2812
2813         native_store_idt(&dt);
2814         vmcs_writel(HOST_IDTR_BASE, dt.address);   /* 22.2.4 */
2815
2816         asm("mov $.Lkvm_vmx_return, %0" : "=r"(kvm_vmx_return));
2817         vmcs_writel(HOST_RIP, kvm_vmx_return); /* 22.2.5 */
2818         vmcs_write32(VM_EXIT_MSR_STORE_COUNT, 0);
2819         vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, 0);
2820         vmcs_write64(VM_EXIT_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.host));
2821         vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, 0);
2822         vmcs_write64(VM_ENTRY_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.guest));
2823
2824         rdmsr(MSR_IA32_SYSENTER_CS, host_sysenter_cs, junk);
2825         vmcs_write32(HOST_IA32_SYSENTER_CS, host_sysenter_cs);
2826         rdmsrl(MSR_IA32_SYSENTER_ESP, a);
2827         vmcs_writel(HOST_IA32_SYSENTER_ESP, a);   /* 22.2.3 */
2828         rdmsrl(MSR_IA32_SYSENTER_EIP, a);
2829         vmcs_writel(HOST_IA32_SYSENTER_EIP, a);   /* 22.2.3 */
2830
2831         if (vmcs_config.vmexit_ctrl & VM_EXIT_LOAD_IA32_PAT) {
2832                 rdmsr(MSR_IA32_CR_PAT, msr_low, msr_high);
2833                 host_pat = msr_low | ((u64) msr_high << 32);
2834                 vmcs_write64(HOST_IA32_PAT, host_pat);
2835         }
2836         if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
2837                 rdmsr(MSR_IA32_CR_PAT, msr_low, msr_high);
2838                 host_pat = msr_low | ((u64) msr_high << 32);
2839                 /* Write the default value follow host pat */
2840                 vmcs_write64(GUEST_IA32_PAT, host_pat);
2841                 /* Keep arch.pat sync with GUEST_IA32_PAT */
2842                 vmx->vcpu.arch.pat = host_pat;
2843         }
2844
2845         for (i = 0; i < NR_VMX_MSR; ++i) {
2846                 u32 index = vmx_msr_index[i];
2847                 u32 data_low, data_high;
2848                 int j = vmx->nmsrs;
2849
2850                 if (rdmsr_safe(index, &data_low, &data_high) < 0)
2851                         continue;
2852                 if (wrmsr_safe(index, data_low, data_high) < 0)
2853                         continue;
2854                 vmx->guest_msrs[j].index = i;
2855                 vmx->guest_msrs[j].data = 0;
2856                 vmx->guest_msrs[j].mask = -1ull;
2857                 ++vmx->nmsrs;
2858         }
2859
2860         vmcs_write32(VM_EXIT_CONTROLS, vmcs_config.vmexit_ctrl);
2861
2862         /* 22.2.1, 20.8.1 */
2863         vmcs_write32(VM_ENTRY_CONTROLS, vmcs_config.vmentry_ctrl);
2864
2865         vmcs_writel(CR0_GUEST_HOST_MASK, ~0UL);
2866         vmx->vcpu.arch.cr4_guest_owned_bits = KVM_CR4_GUEST_OWNED_BITS;
2867         if (enable_ept)
2868                 vmx->vcpu.arch.cr4_guest_owned_bits |= X86_CR4_PGE;
2869         vmcs_writel(CR4_GUEST_HOST_MASK, ~vmx->vcpu.arch.cr4_guest_owned_bits);
2870
2871         kvm_write_tsc(&vmx->vcpu, 0);
2872
2873         return 0;
2874 }
2875
2876 static int vmx_vcpu_reset(struct kvm_vcpu *vcpu)
2877 {
2878         struct vcpu_vmx *vmx = to_vmx(vcpu);
2879         u64 msr;
2880         int ret;
2881
2882         vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP));
2883
2884         vmx->rmode.vm86_active = 0;
2885
2886         vmx->soft_vnmi_blocked = 0;
2887
2888         vmx->vcpu.arch.regs[VCPU_REGS_RDX] = get_rdx_init_val();
2889         kvm_set_cr8(&vmx->vcpu, 0);
2890         msr = 0xfee00000 | MSR_IA32_APICBASE_ENABLE;
2891         if (kvm_vcpu_is_bsp(&vmx->vcpu))
2892                 msr |= MSR_IA32_APICBASE_BSP;
2893         kvm_set_apic_base(&vmx->vcpu, msr);
2894
2895         ret = fx_init(&vmx->vcpu);
2896         if (ret != 0)
2897                 goto out;
2898
2899         vmx_segment_cache_clear(vmx);
2900
2901         seg_setup(VCPU_SREG_CS);
2902         /*
2903          * GUEST_CS_BASE should really be 0xffff0000, but VT vm86 mode
2904          * insists on having GUEST_CS_BASE == GUEST_CS_SELECTOR << 4.  Sigh.
2905          */
2906         if (kvm_vcpu_is_bsp(&vmx->vcpu)) {
2907                 vmcs_write16(GUEST_CS_SELECTOR, 0xf000);
2908                 vmcs_writel(GUEST_CS_BASE, 0x000f0000);
2909         } else {
2910                 vmcs_write16(GUEST_CS_SELECTOR, vmx->vcpu.arch.sipi_vector << 8);
2911                 vmcs_writel(GUEST_CS_BASE, vmx->vcpu.arch.sipi_vector << 12);
2912         }
2913
2914         seg_setup(VCPU_SREG_DS);
2915         seg_setup(VCPU_SREG_ES);
2916         seg_setup(VCPU_SREG_FS);
2917         seg_setup(VCPU_SREG_GS);
2918         seg_setup(VCPU_SREG_SS);
2919
2920         vmcs_write16(GUEST_TR_SELECTOR, 0);
2921         vmcs_writel(GUEST_TR_BASE, 0);
2922         vmcs_write32(GUEST_TR_LIMIT, 0xffff);
2923         vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
2924
2925         vmcs_write16(GUEST_LDTR_SELECTOR, 0);
2926         vmcs_writel(GUEST_LDTR_BASE, 0);
2927         vmcs_write32(GUEST_LDTR_LIMIT, 0xffff);
2928         vmcs_write32(GUEST_LDTR_AR_BYTES, 0x00082);
2929
2930         vmcs_write32(GUEST_SYSENTER_CS, 0);
2931         vmcs_writel(GUEST_SYSENTER_ESP, 0);
2932         vmcs_writel(GUEST_SYSENTER_EIP, 0);
2933
2934         vmcs_writel(GUEST_RFLAGS, 0x02);
2935         if (kvm_vcpu_is_bsp(&vmx->vcpu))
2936                 kvm_rip_write(vcpu, 0xfff0);
2937         else
2938                 kvm_rip_write(vcpu, 0);
2939         kvm_register_write(vcpu, VCPU_REGS_RSP, 0);
2940
2941         vmcs_writel(GUEST_DR7, 0x400);
2942
2943         vmcs_writel(GUEST_GDTR_BASE, 0);
2944         vmcs_write32(GUEST_GDTR_LIMIT, 0xffff);
2945
2946         vmcs_writel(GUEST_IDTR_BASE, 0);
2947         vmcs_write32(GUEST_IDTR_LIMIT, 0xffff);
2948
2949         vmcs_write32(GUEST_ACTIVITY_STATE, GUEST_ACTIVITY_ACTIVE);
2950         vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, 0);
2951         vmcs_write32(GUEST_PENDING_DBG_EXCEPTIONS, 0);
2952
2953         /* Special registers */
2954         vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
2955
2956         setup_msrs(vmx);
2957
2958         vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);  /* 22.2.1 */
2959
2960         if (cpu_has_vmx_tpr_shadow()) {
2961                 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, 0);
2962                 if (vm_need_tpr_shadow(vmx->vcpu.kvm))
2963                         vmcs_write64(VIRTUAL_APIC_PAGE_ADDR,
2964                                      __pa(vmx->vcpu.arch.apic->regs));
2965                 vmcs_write32(TPR_THRESHOLD, 0);
2966         }
2967
2968         if (vm_need_virtualize_apic_accesses(vmx->vcpu.kvm))
2969                 vmcs_write64(APIC_ACCESS_ADDR,
2970                              page_to_phys(vmx->vcpu.kvm->arch.apic_access_page));
2971
2972         if (vmx->vpid != 0)
2973                 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
2974
2975         vmx->vcpu.arch.cr0 = X86_CR0_NW | X86_CR0_CD | X86_CR0_ET;
2976         vmx_set_cr0(&vmx->vcpu, kvm_read_cr0(vcpu)); /* enter rmode */
2977         vmx_set_cr4(&vmx->vcpu, 0);
2978         vmx_set_efer(&vmx->vcpu, 0);
2979         vmx_fpu_activate(&vmx->vcpu);
2980         update_exception_bitmap(&vmx->vcpu);
2981
2982         vpid_sync_context(vmx);
2983
2984         ret = 0;
2985
2986         /* HACK: Don't enable emulation on guest boot/reset */
2987         vmx->emulation_required = 0;
2988
2989 out:
2990         return ret;
2991 }
2992
2993 static void enable_irq_window(struct kvm_vcpu *vcpu)
2994 {
2995         u32 cpu_based_vm_exec_control;
2996
2997         cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
2998         cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_INTR_PENDING;
2999         vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
3000 }
3001
3002 static void enable_nmi_window(struct kvm_vcpu *vcpu)
3003 {
3004         u32 cpu_based_vm_exec_control;
3005
3006         if (!cpu_has_virtual_nmis()) {
3007                 enable_irq_window(vcpu);
3008                 return;
3009         }
3010
3011         if (vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_STI) {
3012                 enable_irq_window(vcpu);
3013                 return;
3014         }
3015         cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
3016         cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_NMI_PENDING;
3017         vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
3018 }
3019
3020 static void vmx_inject_irq(struct kvm_vcpu *vcpu)
3021 {
3022         struct vcpu_vmx *vmx = to_vmx(vcpu);
3023         uint32_t intr;
3024         int irq = vcpu->arch.interrupt.nr;
3025
3026         trace_kvm_inj_virq(irq);
3027
3028         ++vcpu->stat.irq_injections;
3029         if (vmx->rmode.vm86_active) {
3030                 int inc_eip = 0;
3031                 if (vcpu->arch.interrupt.soft)
3032                         inc_eip = vcpu->arch.event_exit_inst_len;
3033                 if (kvm_inject_realmode_interrupt(vcpu, irq, inc_eip) != EMULATE_DONE)
3034                         kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
3035                 return;
3036         }
3037         intr = irq | INTR_INFO_VALID_MASK;
3038         if (vcpu->arch.interrupt.soft) {
3039                 intr |= INTR_TYPE_SOFT_INTR;
3040                 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
3041                              vmx->vcpu.arch.event_exit_inst_len);
3042         } else
3043                 intr |= INTR_TYPE_EXT_INTR;
3044         vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr);
3045         vmx_clear_hlt(vcpu);
3046 }
3047
3048 static void vmx_inject_nmi(struct kvm_vcpu *vcpu)
3049 {
3050         struct vcpu_vmx *vmx = to_vmx(vcpu);
3051
3052         if (!cpu_has_virtual_nmis()) {
3053                 /*
3054                  * Tracking the NMI-blocked state in software is built upon
3055                  * finding the next open IRQ window. This, in turn, depends on
3056                  * well-behaving guests: They have to keep IRQs disabled at
3057                  * least as long as the NMI handler runs. Otherwise we may
3058                  * cause NMI nesting, maybe breaking the guest. But as this is
3059                  * highly unlikely, we can live with the residual risk.
3060                  */
3061                 vmx->soft_vnmi_blocked = 1;
3062                 vmx->vnmi_blocked_time = 0;
3063         }
3064
3065         ++vcpu->stat.nmi_injections;
3066         vmx->nmi_known_unmasked = false;
3067         if (vmx->rmode.vm86_active) {
3068                 if (kvm_inject_realmode_interrupt(vcpu, NMI_VECTOR, 0) != EMULATE_DONE)
3069                         kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
3070                 return;
3071         }
3072         vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
3073                         INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR);
3074         vmx_clear_hlt(vcpu);
3075 }
3076
3077 static int vmx_nmi_allowed(struct kvm_vcpu *vcpu)
3078 {
3079         if (!cpu_has_virtual_nmis() && to_vmx(vcpu)->soft_vnmi_blocked)
3080                 return 0;
3081
3082         return  !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
3083                   (GUEST_INTR_STATE_MOV_SS | GUEST_INTR_STATE_STI
3084                    | GUEST_INTR_STATE_NMI));
3085 }
3086
3087 static bool vmx_get_nmi_mask(struct kvm_vcpu *vcpu)
3088 {
3089         if (!cpu_has_virtual_nmis())
3090                 return to_vmx(vcpu)->soft_vnmi_blocked;
3091         if (to_vmx(vcpu)->nmi_known_unmasked)
3092                 return false;
3093         return vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_NMI;
3094 }
3095
3096 static void vmx_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked)
3097 {
3098         struct vcpu_vmx *vmx = to_vmx(vcpu);
3099
3100         if (!cpu_has_virtual_nmis()) {
3101                 if (vmx->soft_vnmi_blocked != masked) {
3102                         vmx->soft_vnmi_blocked = masked;
3103                         vmx->vnmi_blocked_time = 0;
3104                 }
3105         } else {
3106                 vmx->nmi_known_unmasked = !masked;
3107                 if (masked)
3108                         vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
3109                                       GUEST_INTR_STATE_NMI);
3110                 else
3111                         vmcs_clear_bits(GUEST_INTERRUPTIBILITY_INFO,
3112                                         GUEST_INTR_STATE_NMI);
3113         }
3114 }
3115
3116 static int vmx_interrupt_allowed(struct kvm_vcpu *vcpu)
3117 {
3118         return (vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_IF) &&
3119                 !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
3120                         (GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS));
3121 }
3122
3123 static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr)
3124 {
3125         int ret;
3126         struct kvm_userspace_memory_region tss_mem = {
3127                 .slot = TSS_PRIVATE_MEMSLOT,
3128                 .guest_phys_addr = addr,
3129                 .memory_size = PAGE_SIZE * 3,
3130                 .flags = 0,
3131         };
3132
3133         ret = kvm_set_memory_region(kvm, &tss_mem, 0);
3134         if (ret)
3135                 return ret;
3136         kvm->arch.tss_addr = addr;
3137         if (!init_rmode_tss(kvm))
3138                 return  -ENOMEM;
3139
3140         return 0;
3141 }
3142
3143 static int handle_rmode_exception(struct kvm_vcpu *vcpu,
3144                                   int vec, u32 err_code)
3145 {
3146         /*
3147          * Instruction with address size override prefix opcode 0x67
3148          * Cause the #SS fault with 0 error code in VM86 mode.
3149          */
3150         if (((vec == GP_VECTOR) || (vec == SS_VECTOR)) && err_code == 0)
3151                 if (emulate_instruction(vcpu, 0) == EMULATE_DONE)
3152                         return 1;
3153         /*
3154          * Forward all other exceptions that are valid in real mode.
3155          * FIXME: Breaks guest debugging in real mode, needs to be fixed with
3156          *        the required debugging infrastructure rework.
3157          */
3158         switch (vec) {
3159         case DB_VECTOR:
3160                 if (vcpu->guest_debug &
3161                     (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
3162                         return 0;
3163                 kvm_queue_exception(vcpu, vec);
3164                 return 1;
3165         case BP_VECTOR:
3166                 /*
3167                  * Update instruction length as we may reinject the exception
3168                  * from user space while in guest debugging mode.
3169                  */
3170                 to_vmx(vcpu)->vcpu.arch.event_exit_inst_len =
3171                         vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
3172                 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
3173                         return 0;
3174                 /* fall through */
3175         case DE_VECTOR:
3176         case OF_VECTOR:
3177         case BR_VECTOR:
3178         case UD_VECTOR:
3179         case DF_VECTOR:
3180         case SS_VECTOR:
3181         case GP_VECTOR:
3182         case MF_VECTOR:
3183                 kvm_queue_exception(vcpu, vec);
3184                 return 1;
3185         }
3186         return 0;
3187 }
3188
3189 /*
3190  * Trigger machine check on the host. We assume all the MSRs are already set up
3191  * by the CPU and that we still run on the same CPU as the MCE occurred on.
3192  * We pass a fake environment to the machine check handler because we want
3193  * the guest to be always treated like user space, no matter what context
3194  * it used internally.
3195  */
3196 static void kvm_machine_check(void)
3197 {
3198 #if defined(CONFIG_X86_MCE) && defined(CONFIG_X86_64)
3199         struct pt_regs regs = {
3200                 .cs = 3, /* Fake ring 3 no matter what the guest ran on */
3201                 .flags = X86_EFLAGS_IF,
3202         };
3203
3204         do_machine_check(&regs, 0);
3205 #endif
3206 }
3207
3208 static int handle_machine_check(struct kvm_vcpu *vcpu)
3209 {
3210         /* already handled by vcpu_run */
3211         return 1;
3212 }
3213
3214 static int handle_exception(struct kvm_vcpu *vcpu)
3215 {
3216         struct vcpu_vmx *vmx = to_vmx(vcpu);
3217         struct kvm_run *kvm_run = vcpu->run;
3218         u32 intr_info, ex_no, error_code;
3219         unsigned long cr2, rip, dr6;
3220         u32 vect_info;
3221         enum emulation_result er;
3222
3223         vect_info = vmx->idt_vectoring_info;
3224         intr_info = vmx->exit_intr_info;
3225
3226         if (is_machine_check(intr_info))
3227                 return handle_machine_check(vcpu);
3228
3229         if ((vect_info & VECTORING_INFO_VALID_MASK) &&
3230             !is_page_fault(intr_info)) {
3231                 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
3232                 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_SIMUL_EX;
3233                 vcpu->run->internal.ndata = 2;
3234                 vcpu->run->internal.data[0] = vect_info;
3235                 vcpu->run->internal.data[1] = intr_info;
3236                 return 0;
3237         }
3238
3239         if ((intr_info & INTR_INFO_INTR_TYPE_MASK) == INTR_TYPE_NMI_INTR)
3240                 return 1;  /* already handled by vmx_vcpu_run() */
3241
3242         if (is_no_device(intr_info)) {
3243                 vmx_fpu_activate(vcpu);
3244                 return 1;
3245         }
3246
3247         if (is_invalid_opcode(intr_info)) {
3248                 er = emulate_instruction(vcpu, EMULTYPE_TRAP_UD);
3249                 if (er != EMULATE_DONE)
3250                         kvm_queue_exception(vcpu, UD_VECTOR);
3251                 return 1;
3252         }
3253
3254         error_code = 0;
3255         if (intr_info & INTR_INFO_DELIVER_CODE_MASK)
3256                 error_code = vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
3257         if (is_page_fault(intr_info)) {
3258                 /* EPT won't cause page fault directly */
3259                 if (enable_ept)
3260                         BUG();
3261                 cr2 = vmcs_readl(EXIT_QUALIFICATION);
3262                 trace_kvm_page_fault(cr2, error_code);
3263
3264                 if (kvm_event_needs_reinjection(vcpu))
3265                         kvm_mmu_unprotect_page_virt(vcpu, cr2);
3266                 return kvm_mmu_page_fault(vcpu, cr2, error_code, NULL, 0);
3267         }
3268
3269         if (vmx->rmode.vm86_active &&
3270             handle_rmode_exception(vcpu, intr_info & INTR_INFO_VECTOR_MASK,
3271                                                                 error_code)) {
3272                 if (vcpu->arch.halt_request) {
3273                         vcpu->arch.halt_request = 0;
3274                         return kvm_emulate_halt(vcpu);
3275                 }
3276                 return 1;
3277         }
3278
3279         ex_no = intr_info & INTR_INFO_VECTOR_MASK;
3280         switch (ex_no) {
3281         case DB_VECTOR:
3282                 dr6 = vmcs_readl(EXIT_QUALIFICATION);
3283                 if (!(vcpu->guest_debug &
3284                       (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))) {
3285                         vcpu->arch.dr6 = dr6 | DR6_FIXED_1;
3286                         kvm_queue_exception(vcpu, DB_VECTOR);
3287                         return 1;
3288                 }
3289                 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1;
3290                 kvm_run->debug.arch.dr7 = vmcs_readl(GUEST_DR7);
3291                 /* fall through */
3292         case BP_VECTOR:
3293                 /*
3294                  * Update instruction length as we may reinject #BP from
3295                  * user space while in guest debugging mode. Reading it for
3296                  * #DB as well causes no harm, it is not used in that case.
3297                  */
3298                 vmx->vcpu.arch.event_exit_inst_len =
3299                         vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
3300                 kvm_run->exit_reason = KVM_EXIT_DEBUG;
3301                 rip = kvm_rip_read(vcpu);
3302                 kvm_run->debug.arch.pc = vmcs_readl(GUEST_CS_BASE) + rip;
3303                 kvm_run->debug.arch.exception = ex_no;
3304                 break;
3305         default:
3306                 kvm_run->exit_reason = KVM_EXIT_EXCEPTION;
3307                 kvm_run->ex.exception = ex_no;
3308                 kvm_run->ex.error_code = error_code;
3309                 break;
3310         }
3311         return 0;
3312 }
3313
3314 static int handle_external_interrupt(struct kvm_vcpu *vcpu)
3315 {
3316         ++vcpu->stat.irq_exits;
3317         return 1;
3318 }
3319
3320 static int handle_triple_fault(struct kvm_vcpu *vcpu)
3321 {
3322         vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
3323         return 0;
3324 }
3325
3326 static int handle_io(struct kvm_vcpu *vcpu)
3327 {
3328         unsigned long exit_qualification;
3329         int size, in, string;
3330         unsigned port;
3331
3332         exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
3333         string = (exit_qualification & 16) != 0;
3334         in = (exit_qualification & 8) != 0;
3335
3336         ++vcpu->stat.io_exits;
3337
3338         if (string || in)
3339                 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
3340
3341         port = exit_qualification >> 16;
3342         size = (exit_qualification & 7) + 1;
3343         skip_emulated_instruction(vcpu);
3344
3345         return kvm_fast_pio_out(vcpu, size, port);
3346 }
3347
3348 static void
3349 vmx_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
3350 {
3351         /*
3352          * Patch in the VMCALL instruction:
3353          */
3354         hypercall[0] = 0x0f;
3355         hypercall[1] = 0x01;
3356         hypercall[2] = 0xc1;
3357 }
3358
3359 static int handle_cr(struct kvm_vcpu *vcpu)
3360 {
3361         unsigned long exit_qualification, val;
3362         int cr;
3363         int reg;
3364         int err;
3365
3366         exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
3367         cr = exit_qualification & 15;
3368         reg = (exit_qualification >> 8) & 15;
3369         switch ((exit_qualification >> 4) & 3) {
3370         case 0: /* mov to cr */
3371                 val = kvm_register_read(vcpu, reg);
3372                 trace_kvm_cr_write(cr, val);
3373                 switch (cr) {
3374                 case 0:
3375                         err = kvm_set_cr0(vcpu, val);
3376                         kvm_complete_insn_gp(vcpu, err);
3377                         return 1;
3378                 case 3:
3379                         err = kvm_set_cr3(vcpu, val);
3380                         kvm_complete_insn_gp(vcpu, err);
3381                         return 1;
3382                 case 4:
3383                         err = kvm_set_cr4(vcpu, val);
3384                         kvm_complete_insn_gp(vcpu, err);
3385                         return 1;
3386                 case 8: {
3387                                 u8 cr8_prev = kvm_get_cr8(vcpu);
3388                                 u8 cr8 = kvm_register_read(vcpu, reg);
3389                                 err = kvm_set_cr8(vcpu, cr8);
3390                                 kvm_complete_insn_gp(vcpu, err);
3391                                 if (irqchip_in_kernel(vcpu->kvm))
3392                                         return 1;
3393                                 if (cr8_prev <= cr8)
3394                                         return 1;
3395                                 vcpu->run->exit_reason = KVM_EXIT_SET_TPR;
3396                                 return 0;
3397                         }
3398                 };
3399                 break;
3400         case 2: /* clts */
3401                 vmx_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~X86_CR0_TS));
3402                 trace_kvm_cr_write(0, kvm_read_cr0(vcpu));
3403                 skip_emulated_instruction(vcpu);
3404                 vmx_fpu_activate(vcpu);
3405                 return 1;
3406         case 1: /*mov from cr*/
3407                 switch (cr) {
3408                 case 3:
3409                         val = kvm_read_cr3(vcpu);
3410                         kvm_register_write(vcpu, reg, val);
3411                         trace_kvm_cr_read(cr, val);
3412                         skip_emulated_instruction(vcpu);
3413                         return 1;
3414                 case 8:
3415                         val = kvm_get_cr8(vcpu);
3416                         kvm_register_write(vcpu, reg, val);
3417                         trace_kvm_cr_read(cr, val);
3418                         skip_emulated_instruction(vcpu);
3419                         return 1;
3420                 }
3421                 break;
3422         case 3: /* lmsw */
3423                 val = (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f;
3424                 trace_kvm_cr_write(0, (kvm_read_cr0(vcpu) & ~0xful) | val);
3425                 kvm_lmsw(vcpu, val);
3426
3427                 skip_emulated_instruction(vcpu);
3428                 return 1;
3429         default:
3430                 break;
3431         }
3432         vcpu->run->exit_reason = 0;
3433         pr_unimpl(vcpu, "unhandled control register: op %d cr %d\n",
3434                (int)(exit_qualification >> 4) & 3, cr);
3435         return 0;
3436 }
3437
3438 static int handle_dr(struct kvm_vcpu *vcpu)
3439 {
3440         unsigned long exit_qualification;
3441         int dr, reg;
3442
3443         /* Do not handle if the CPL > 0, will trigger GP on re-entry */
3444         if (!kvm_require_cpl(vcpu, 0))
3445                 return 1;
3446         dr = vmcs_readl(GUEST_DR7);
3447         if (dr & DR7_GD) {
3448                 /*
3449                  * As the vm-exit takes precedence over the debug trap, we
3450                  * need to emulate the latter, either for the host or the
3451                  * guest debugging itself.
3452                  */
3453                 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
3454                         vcpu->run->debug.arch.dr6 = vcpu->arch.dr6;
3455                         vcpu->run->debug.arch.dr7 = dr;
3456                         vcpu->run->debug.arch.pc =
3457                                 vmcs_readl(GUEST_CS_BASE) +
3458                                 vmcs_readl(GUEST_RIP);
3459                         vcpu->run->debug.arch.exception = DB_VECTOR;
3460                         vcpu->run->exit_reason = KVM_EXIT_DEBUG;
3461                         return 0;
3462                 } else {
3463                         vcpu->arch.dr7 &= ~DR7_GD;
3464                         vcpu->arch.dr6 |= DR6_BD;
3465                         vmcs_writel(GUEST_DR7, vcpu->arch.dr7);
3466                         kvm_queue_exception(vcpu, DB_VECTOR);
3467                         return 1;
3468                 }
3469         }
3470
3471         exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
3472         dr = exit_qualification & DEBUG_REG_ACCESS_NUM;
3473         reg = DEBUG_REG_ACCESS_REG(exit_qualification);
3474         if (exit_qualification & TYPE_MOV_FROM_DR) {
3475                 unsigned long val;
3476                 if (!kvm_get_dr(vcpu, dr, &val))
3477                         kvm_register_write(vcpu, reg, val);
3478         } else
3479                 kvm_set_dr(vcpu, dr, vcpu->arch.regs[reg]);
3480         skip_emulated_instruction(vcpu);
3481         return 1;
3482 }
3483
3484 static void vmx_set_dr7(struct kvm_vcpu *vcpu, unsigned long val)
3485 {
3486         vmcs_writel(GUEST_DR7, val);
3487 }
3488
3489 static int handle_cpuid(struct kvm_vcpu *vcpu)
3490 {
3491         kvm_emulate_cpuid(vcpu);
3492         return 1;
3493 }
3494
3495 static int handle_rdmsr(struct kvm_vcpu *vcpu)
3496 {
3497         u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
3498         u64 data;
3499
3500         if (vmx_get_msr(vcpu, ecx, &data)) {
3501                 trace_kvm_msr_read_ex(ecx);
3502                 kvm_inject_gp(vcpu, 0);
3503                 return 1;
3504         }
3505
3506         trace_kvm_msr_read(ecx, data);
3507
3508         /* FIXME: handling of bits 32:63 of rax, rdx */
3509         vcpu->arch.regs[VCPU_REGS_RAX] = data & -1u;
3510         vcpu->arch.regs[VCPU_REGS_RDX] = (data >> 32) & -1u;
3511         skip_emulated_instruction(vcpu);
3512         return 1;
3513 }
3514
3515 static int handle_wrmsr(struct kvm_vcpu *vcpu)
3516 {
3517         u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
3518         u64 data = (vcpu->arch.regs[VCPU_REGS_RAX] & -1u)
3519                 | ((u64)(vcpu->arch.regs[VCPU_REGS_RDX] & -1u) << 32);
3520
3521         if (vmx_set_msr(vcpu, ecx, data) != 0) {
3522                 trace_kvm_msr_write_ex(ecx, data);
3523                 kvm_inject_gp(vcpu, 0);
3524                 return 1;
3525         }
3526
3527         trace_kvm_msr_write(ecx, data);
3528         skip_emulated_instruction(vcpu);
3529         return 1;
3530 }
3531
3532 static int handle_tpr_below_threshold(struct kvm_vcpu *vcpu)
3533 {
3534         kvm_make_request(KVM_REQ_EVENT, vcpu);
3535         return 1;
3536 }
3537
3538 static int handle_interrupt_window(struct kvm_vcpu *vcpu)
3539 {
3540         u32 cpu_based_vm_exec_control;
3541
3542         /* clear pending irq */
3543         cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
3544         cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_INTR_PENDING;
3545         vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
3546
3547         kvm_make_request(KVM_REQ_EVENT, vcpu);
3548
3549         ++vcpu->stat.irq_window_exits;
3550
3551         /*
3552          * If the user space waits to inject interrupts, exit as soon as
3553          * possible
3554          */
3555         if (!irqchip_in_kernel(vcpu->kvm) &&
3556             vcpu->run->request_interrupt_window &&
3557             !kvm_cpu_has_interrupt(vcpu)) {
3558                 vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
3559                 return 0;
3560         }
3561         return 1;
3562 }
3563
3564 static int handle_halt(struct kvm_vcpu *vcpu)
3565 {
3566         skip_emulated_instruction(vcpu);
3567         return kvm_emulate_halt(vcpu);
3568 }
3569
3570 static int handle_vmcall(struct kvm_vcpu *vcpu)
3571 {
3572         skip_emulated_instruction(vcpu);
3573         kvm_emulate_hypercall(vcpu);
3574         return 1;
3575 }
3576
3577 static int handle_vmx_insn(struct kvm_vcpu *vcpu)
3578 {
3579         kvm_queue_exception(vcpu, UD_VECTOR);
3580         return 1;
3581 }
3582
3583 static int handle_invd(struct kvm_vcpu *vcpu)
3584 {
3585         return emulate_instruction(vcpu, 0) == EMULATE_DONE;
3586 }
3587
3588 static int handle_invlpg(struct kvm_vcpu *vcpu)
3589 {
3590         unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
3591
3592         kvm_mmu_invlpg(vcpu, exit_qualification);
3593         skip_emulated_instruction(vcpu);
3594         return 1;
3595 }
3596
3597 static int handle_wbinvd(struct kvm_vcpu *vcpu)
3598 {
3599         skip_emulated_instruction(vcpu);
3600         kvm_emulate_wbinvd(vcpu);
3601         return 1;
3602 }
3603
3604 static int handle_xsetbv(struct kvm_vcpu *vcpu)
3605 {
3606         u64 new_bv = kvm_read_edx_eax(vcpu);
3607         u32 index = kvm_register_read(vcpu, VCPU_REGS_RCX);
3608
3609         if (kvm_set_xcr(vcpu, index, new_bv) == 0)
3610                 skip_emulated_instruction(vcpu);
3611         return 1;
3612 }
3613
3614 static int handle_apic_access(struct kvm_vcpu *vcpu)
3615 {
3616         return emulate_instruction(vcpu, 0) == EMULATE_DONE;
3617 }
3618
3619 static int handle_task_switch(struct kvm_vcpu *vcpu)
3620 {
3621         struct vcpu_vmx *vmx = to_vmx(vcpu);
3622         unsigned long exit_qualification;
3623         bool has_error_code = false;
3624         u32 error_code = 0;
3625         u16 tss_selector;
3626         int reason, type, idt_v;
3627
3628         idt_v = (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK);
3629         type = (vmx->idt_vectoring_info & VECTORING_INFO_TYPE_MASK);
3630
3631         exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
3632
3633         reason = (u32)exit_qualification >> 30;
3634         if (reason == TASK_SWITCH_GATE && idt_v) {
3635                 switch (type) {
3636                 case INTR_TYPE_NMI_INTR:
3637                         vcpu->arch.nmi_injected = false;
3638                         vmx_set_nmi_mask(vcpu, true);
3639                         break;
3640                 case INTR_TYPE_EXT_INTR:
3641                 case INTR_TYPE_SOFT_INTR:
3642                         kvm_clear_interrupt_queue(vcpu);
3643                         break;
3644                 case INTR_TYPE_HARD_EXCEPTION:
3645                         if (vmx->idt_vectoring_info &
3646                             VECTORING_INFO_DELIVER_CODE_MASK) {
3647                                 has_error_code = true;
3648                                 error_code =
3649                                         vmcs_read32(IDT_VECTORING_ERROR_CODE);
3650                         }
3651                         /* fall through */
3652                 case INTR_TYPE_SOFT_EXCEPTION:
3653                         kvm_clear_exception_queue(vcpu);
3654                         break;
3655                 default:
3656                         break;
3657                 }
3658         }
3659         tss_selector = exit_qualification;
3660
3661         if (!idt_v || (type != INTR_TYPE_HARD_EXCEPTION &&
3662                        type != INTR_TYPE_EXT_INTR &&
3663                        type != INTR_TYPE_NMI_INTR))
3664                 skip_emulated_instruction(vcpu);
3665
3666         if (kvm_task_switch(vcpu, tss_selector, reason,
3667                                 has_error_code, error_code) == EMULATE_FAIL) {
3668                 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
3669                 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
3670                 vcpu->run->internal.ndata = 0;
3671                 return 0;
3672         }
3673
3674         /* clear all local breakpoint enable flags */
3675         vmcs_writel(GUEST_DR7, vmcs_readl(GUEST_DR7) & ~55);
3676
3677         /*
3678          * TODO: What about debug traps on tss switch?
3679          *       Are we supposed to inject them and update dr6?
3680          */
3681
3682         return 1;
3683 }
3684
3685 static int handle_ept_violation(struct kvm_vcpu *vcpu)
3686 {
3687         unsigned long exit_qualification;
3688         gpa_t gpa;
3689         int gla_validity;
3690
3691         exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
3692
3693         if (exit_qualification & (1 << 6)) {
3694                 printk(KERN_ERR "EPT: GPA exceeds GAW!\n");
3695                 return -EINVAL;
3696         }
3697
3698         gla_validity = (exit_qualification >> 7) & 0x3;
3699         if (gla_validity != 0x3 && gla_validity != 0x1 && gla_validity != 0) {
3700                 printk(KERN_ERR "EPT: Handling EPT violation failed!\n");
3701                 printk(KERN_ERR "EPT: GPA: 0x%lx, GVA: 0x%lx\n",
3702                         (long unsigned int)vmcs_read64(GUEST_PHYSICAL_ADDRESS),
3703                         vmcs_readl(GUEST_LINEAR_ADDRESS));
3704                 printk(KERN_ERR "EPT: Exit qualification is 0x%lx\n",
3705                         (long unsigned int)exit_qualification);
3706                 vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
3707                 vcpu->run->hw.hardware_exit_reason = EXIT_REASON_EPT_VIOLATION;
3708                 return 0;
3709         }
3710
3711         gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
3712         trace_kvm_page_fault(gpa, exit_qualification);
3713         return kvm_mmu_page_fault(vcpu, gpa, exit_qualification & 0x3, NULL, 0);
3714 }
3715
3716 static u64 ept_rsvd_mask(u64 spte, int level)
3717 {
3718         int i;
3719         u64 mask = 0;
3720
3721         for (i = 51; i > boot_cpu_data.x86_phys_bits; i--)
3722                 mask |= (1ULL << i);
3723
3724         if (level > 2)
3725                 /* bits 7:3 reserved */
3726                 mask |= 0xf8;
3727         else if (level == 2) {
3728                 if (spte & (1ULL << 7))
3729                         /* 2MB ref, bits 20:12 reserved */
3730                         mask |= 0x1ff000;
3731                 else
3732                         /* bits 6:3 reserved */
3733                         mask |= 0x78;
3734         }
3735
3736         return mask;
3737 }
3738
3739 static void ept_misconfig_inspect_spte(struct kvm_vcpu *vcpu, u64 spte,
3740                                        int level)
3741 {
3742         printk(KERN_ERR "%s: spte 0x%llx level %d\n", __func__, spte, level);
3743
3744         /* 010b (write-only) */
3745         WARN_ON((spte & 0x7) == 0x2);
3746
3747         /* 110b (write/execute) */
3748         WARN_ON((spte & 0x7) == 0x6);
3749
3750         /* 100b (execute-only) and value not supported by logical processor */
3751         if (!cpu_has_vmx_ept_execute_only())
3752                 WARN_ON((spte & 0x7) == 0x4);
3753
3754         /* not 000b */
3755         if ((spte & 0x7)) {
3756                 u64 rsvd_bits = spte & ept_rsvd_mask(spte, level);
3757
3758                 if (rsvd_bits != 0) {
3759                         printk(KERN_ERR "%s: rsvd_bits = 0x%llx\n",
3760                                          __func__, rsvd_bits);
3761                         WARN_ON(1);
3762                 }
3763
3764                 if (level == 1 || (level == 2 && (spte & (1ULL << 7)))) {
3765                         u64 ept_mem_type = (spte & 0x38) >> 3;
3766
3767                         if (ept_mem_type == 2 || ept_mem_type == 3 ||
3768                             ept_mem_type == 7) {
3769                                 printk(KERN_ERR "%s: ept_mem_type=0x%llx\n",
3770                                                 __func__, ept_mem_type);
3771                                 WARN_ON(1);
3772                         }
3773                 }
3774         }
3775 }
3776
3777 static int handle_ept_misconfig(struct kvm_vcpu *vcpu)
3778 {
3779         u64 sptes[4];
3780         int nr_sptes, i;
3781         gpa_t gpa;
3782
3783         gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
3784
3785         printk(KERN_ERR "EPT: Misconfiguration.\n");
3786         printk(KERN_ERR "EPT: GPA: 0x%llx\n", gpa);
3787
3788         nr_sptes = kvm_mmu_get_spte_hierarchy(vcpu, gpa, sptes);
3789
3790         for (i = PT64_ROOT_LEVEL; i > PT64_ROOT_LEVEL - nr_sptes; --i)
3791                 ept_misconfig_inspect_spte(vcpu, sptes[i-1], i);
3792
3793         vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
3794         vcpu->run->hw.hardware_exit_reason = EXIT_REASON_EPT_MISCONFIG;
3795
3796         return 0;
3797 }
3798
3799 static int handle_nmi_window(struct kvm_vcpu *vcpu)
3800 {
3801         u32 cpu_based_vm_exec_control;
3802
3803         /* clear pending NMI */
3804         cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
3805         cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_NMI_PENDING;
3806         vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
3807         ++vcpu->stat.nmi_window_exits;
3808         kvm_make_request(KVM_REQ_EVENT, vcpu);
3809
3810         return 1;
3811 }
3812
3813 static int handle_invalid_guest_state(struct kvm_vcpu *vcpu)
3814 {
3815         struct vcpu_vmx *vmx = to_vmx(vcpu);
3816         enum emulation_result err = EMULATE_DONE;
3817         int ret = 1;
3818         u32 cpu_exec_ctrl;
3819         bool intr_window_requested;
3820
3821         cpu_exec_ctrl = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
3822         intr_window_requested = cpu_exec_ctrl & CPU_BASED_VIRTUAL_INTR_PENDING;
3823
3824         while (!guest_state_valid(vcpu)) {
3825                 if (intr_window_requested
3826                     && (kvm_get_rflags(&vmx->vcpu) & X86_EFLAGS_IF))
3827                         return handle_interrupt_window(&vmx->vcpu);
3828
3829                 err = emulate_instruction(vcpu, 0);
3830
3831                 if (err == EMULATE_DO_MMIO) {
3832                         ret = 0;
3833                         goto out;
3834                 }
3835
3836                 if (err != EMULATE_DONE)
3837                         return 0;
3838
3839                 if (signal_pending(current))
3840                         goto out;
3841                 if (need_resched())
3842                         schedule();
3843         }
3844
3845         vmx->emulation_required = 0;
3846 out:
3847         return ret;
3848 }
3849
3850 /*
3851  * Indicate a busy-waiting vcpu in spinlock. We do not enable the PAUSE
3852  * exiting, so only get here on cpu with PAUSE-Loop-Exiting.
3853  */
3854 static int handle_pause(struct kvm_vcpu *vcpu)
3855 {
3856         skip_emulated_instruction(vcpu);
3857         kvm_vcpu_on_spin(vcpu);
3858
3859         return 1;
3860 }
3861
3862 static int handle_invalid_op(struct kvm_vcpu *vcpu)
3863 {
3864         kvm_queue_exception(vcpu, UD_VECTOR);
3865         return 1;
3866 }
3867
3868 /*
3869  * The exit handlers return 1 if the exit was handled fully and guest execution
3870  * may resume.  Otherwise they set the kvm_run parameter to indicate what needs
3871  * to be done to userspace and return 0.
3872  */
3873 static int (*kvm_vmx_exit_handlers[])(struct kvm_vcpu *vcpu) = {
3874         [EXIT_REASON_EXCEPTION_NMI]           = handle_exception,
3875         [EXIT_REASON_EXTERNAL_INTERRUPT]      = handle_external_interrupt,
3876         [EXIT_REASON_TRIPLE_FAULT]            = handle_triple_fault,
3877         [EXIT_REASON_NMI_WINDOW]              = handle_nmi_window,
3878         [EXIT_REASON_IO_INSTRUCTION]          = handle_io,
3879         [EXIT_REASON_CR_ACCESS]               = handle_cr,
3880         [EXIT_REASON_DR_ACCESS]               = handle_dr,
3881         [EXIT_REASON_CPUID]                   = handle_cpuid,
3882         [EXIT_REASON_MSR_READ]                = handle_rdmsr,
3883         [EXIT_REASON_MSR_WRITE]               = handle_wrmsr,
3884         [EXIT_REASON_PENDING_INTERRUPT]       = handle_interrupt_window,
3885         [EXIT_REASON_HLT]                     = handle_halt,
3886         [EXIT_REASON_INVD]                    = handle_invd,
3887         [EXIT_REASON_INVLPG]                  = handle_invlpg,
3888         [EXIT_REASON_VMCALL]                  = handle_vmcall,
3889         [EXIT_REASON_VMCLEAR]                 = handle_vmx_insn,
3890         [EXIT_REASON_VMLAUNCH]                = handle_vmx_insn,
3891         [EXIT_REASON_VMPTRLD]                 = handle_vmx_insn,
3892         [EXIT_REASON_VMPTRST]                 = handle_vmx_insn,
3893         [EXIT_REASON_VMREAD]                  = handle_vmx_insn,
3894         [EXIT_REASON_VMRESUME]                = handle_vmx_insn,
3895         [EXIT_REASON_VMWRITE]                 = handle_vmx_insn,
3896         [EXIT_REASON_VMOFF]                   = handle_vmx_insn,
3897         [EXIT_REASON_VMON]                    = handle_vmx_insn,
3898         [EXIT_REASON_TPR_BELOW_THRESHOLD]     = handle_tpr_below_threshold,
3899         [EXIT_REASON_APIC_ACCESS]             = handle_apic_access,
3900         [EXIT_REASON_WBINVD]                  = handle_wbinvd,
3901         [EXIT_REASON_XSETBV]                  = handle_xsetbv,
3902         [EXIT_REASON_TASK_SWITCH]             = handle_task_switch,
3903         [EXIT_REASON_MCE_DURING_VMENTRY]      = handle_machine_check,
3904         [EXIT_REASON_EPT_VIOLATION]           = handle_ept_violation,
3905         [EXIT_REASON_EPT_MISCONFIG]           = handle_ept_misconfig,
3906         [EXIT_REASON_PAUSE_INSTRUCTION]       = handle_pause,
3907         [EXIT_REASON_MWAIT_INSTRUCTION]       = handle_invalid_op,
3908         [EXIT_REASON_MONITOR_INSTRUCTION]     = handle_invalid_op,
3909 };
3910
3911 static const int kvm_vmx_max_exit_handlers =
3912         ARRAY_SIZE(kvm_vmx_exit_handlers);
3913
3914 static void vmx_get_exit_info(struct kvm_vcpu *vcpu, u64 *info1, u64 *info2)
3915 {
3916         *info1 = vmcs_readl(EXIT_QUALIFICATION);
3917         *info2 = vmcs_read32(VM_EXIT_INTR_INFO);
3918 }
3919
3920 /*
3921  * The guest has exited.  See if we can fix it or if we need userspace
3922  * assistance.
3923  */
3924 static int vmx_handle_exit(struct kvm_vcpu *vcpu)
3925 {
3926         struct vcpu_vmx *vmx = to_vmx(vcpu);
3927         u32 exit_reason = vmx->exit_reason;
3928         u32 vectoring_info = vmx->idt_vectoring_info;
3929
3930         trace_kvm_exit(exit_reason, vcpu, KVM_ISA_VMX);
3931
3932         /* If guest state is invalid, start emulating */
3933         if (vmx->emulation_required && emulate_invalid_guest_state)
3934                 return handle_invalid_guest_state(vcpu);
3935
3936         if (exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY) {
3937                 vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
3938                 vcpu->run->fail_entry.hardware_entry_failure_reason
3939                         = exit_reason;
3940                 return 0;
3941         }
3942
3943         if (unlikely(vmx->fail)) {
3944                 vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
3945                 vcpu->run->fail_entry.hardware_entry_failure_reason
3946                         = vmcs_read32(VM_INSTRUCTION_ERROR);
3947                 return 0;
3948         }
3949
3950         if ((vectoring_info & VECTORING_INFO_VALID_MASK) &&
3951                         (exit_reason != EXIT_REASON_EXCEPTION_NMI &&
3952                         exit_reason != EXIT_REASON_EPT_VIOLATION &&
3953                         exit_reason != EXIT_REASON_TASK_SWITCH))
3954                 printk(KERN_WARNING "%s: unexpected, valid vectoring info "
3955                        "(0x%x) and exit reason is 0x%x\n",
3956                        __func__, vectoring_info, exit_reason);
3957
3958         if (unlikely(!cpu_has_virtual_nmis() && vmx->soft_vnmi_blocked)) {
3959                 if (vmx_interrupt_allowed(vcpu)) {
3960                         vmx->soft_vnmi_blocked = 0;
3961                 } else if (vmx->vnmi_blocked_time > 1000000000LL &&
3962                            vcpu->arch.nmi_pending) {
3963                         /*
3964                          * This CPU don't support us in finding the end of an
3965                          * NMI-blocked window if the guest runs with IRQs
3966                          * disabled. So we pull the trigger after 1 s of
3967                          * futile waiting, but inform the user about this.
3968                          */
3969                         printk(KERN_WARNING "%s: Breaking out of NMI-blocked "
3970                                "state on VCPU %d after 1 s timeout\n",
3971                                __func__, vcpu->vcpu_id);
3972                         vmx->soft_vnmi_blocked = 0;
3973                 }
3974         }
3975
3976         if (exit_reason < kvm_vmx_max_exit_handlers
3977             && kvm_vmx_exit_handlers[exit_reason])
3978                 return kvm_vmx_exit_handlers[exit_reason](vcpu);
3979         else {
3980                 vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
3981                 vcpu->run->hw.hardware_exit_reason = exit_reason;
3982         }
3983         return 0;
3984 }
3985
3986 static void update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
3987 {
3988         if (irr == -1 || tpr < irr) {
3989                 vmcs_write32(TPR_THRESHOLD, 0);
3990                 return;
3991         }
3992
3993         vmcs_write32(TPR_THRESHOLD, irr);
3994 }
3995
3996 static void vmx_complete_atomic_exit(struct vcpu_vmx *vmx)
3997 {
3998         u32 exit_intr_info;
3999
4000         if (!(vmx->exit_reason == EXIT_REASON_MCE_DURING_VMENTRY
4001               || vmx->exit_reason == EXIT_REASON_EXCEPTION_NMI))
4002                 return;
4003
4004         vmx->exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
4005         exit_intr_info = vmx->exit_intr_info;
4006
4007         /* Handle machine checks before interrupts are enabled */
4008         if (is_machine_check(exit_intr_info))
4009                 kvm_machine_check();
4010
4011         /* We need to handle NMIs before interrupts are enabled */
4012         if ((exit_intr_info & INTR_INFO_INTR_TYPE_MASK) == INTR_TYPE_NMI_INTR &&
4013             (exit_intr_info & INTR_INFO_VALID_MASK)) {
4014                 kvm_before_handle_nmi(&vmx->vcpu);
4015                 asm("int $2");
4016                 kvm_after_handle_nmi(&vmx->vcpu);
4017         }
4018 }
4019
4020 static void vmx_recover_nmi_blocking(struct vcpu_vmx *vmx)
4021 {
4022         u32 exit_intr_info;
4023         bool unblock_nmi;
4024         u8 vector;
4025         bool idtv_info_valid;
4026
4027         idtv_info_valid = vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK;
4028
4029         if (cpu_has_virtual_nmis()) {
4030                 if (vmx->nmi_known_unmasked)
4031                         return;
4032                 /*
4033                  * Can't use vmx->exit_intr_info since we're not sure what
4034                  * the exit reason is.
4035                  */
4036                 exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
4037                 unblock_nmi = (exit_intr_info & INTR_INFO_UNBLOCK_NMI) != 0;
4038                 vector = exit_intr_info & INTR_INFO_VECTOR_MASK;
4039                 /*
4040                  * SDM 3: 27.7.1.2 (September 2008)
4041                  * Re-set bit "block by NMI" before VM entry if vmexit caused by
4042                  * a guest IRET fault.
4043                  * SDM 3: 23.2.2 (September 2008)
4044                  * Bit 12 is undefined in any of the following cases:
4045                  *  If the VM exit sets the valid bit in the IDT-vectoring
4046                  *   information field.
4047                  *  If the VM exit is due to a double fault.
4048                  */
4049                 if ((exit_intr_info & INTR_INFO_VALID_MASK) && unblock_nmi &&
4050                     vector != DF_VECTOR && !idtv_info_valid)
4051                         vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
4052                                       GUEST_INTR_STATE_NMI);
4053                 else
4054                         vmx->nmi_known_unmasked =
4055                                 !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO)
4056                                   & GUEST_INTR_STATE_NMI);
4057         } else if (unlikely(vmx->soft_vnmi_blocked))
4058                 vmx->vnmi_blocked_time +=
4059                         ktime_to_ns(ktime_sub(ktime_get(), vmx->entry_time));
4060 }
4061
4062 static void __vmx_complete_interrupts(struct vcpu_vmx *vmx,
4063                                       u32 idt_vectoring_info,
4064                                       int instr_len_field,
4065                                       int error_code_field)
4066 {
4067         u8 vector;
4068         int type;
4069         bool idtv_info_valid;
4070
4071         idtv_info_valid = idt_vectoring_info & VECTORING_INFO_VALID_MASK;
4072
4073         vmx->vcpu.arch.nmi_injected = false;
4074         kvm_clear_exception_queue(&vmx->vcpu);
4075         kvm_clear_interrupt_queue(&vmx->vcpu);
4076
4077         if (!idtv_info_valid)
4078                 return;
4079
4080         kvm_make_request(KVM_REQ_EVENT, &vmx->vcpu);
4081
4082         vector = idt_vectoring_info & VECTORING_INFO_VECTOR_MASK;
4083         type = idt_vectoring_info & VECTORING_INFO_TYPE_MASK;
4084
4085         switch (type) {
4086         case INTR_TYPE_NMI_INTR:
4087                 vmx->vcpu.arch.nmi_injected = true;
4088                 /*
4089                  * SDM 3: 27.7.1.2 (September 2008)
4090                  * Clear bit "block by NMI" before VM entry if a NMI
4091                  * delivery faulted.
4092                  */
4093                 vmx_set_nmi_mask(&vmx->vcpu, false);
4094                 break;
4095         case INTR_TYPE_SOFT_EXCEPTION:
4096                 vmx->vcpu.arch.event_exit_inst_len =
4097                         vmcs_read32(instr_len_field);
4098                 /* fall through */
4099         case INTR_TYPE_HARD_EXCEPTION:
4100                 if (idt_vectoring_info & VECTORING_INFO_DELIVER_CODE_MASK) {
4101                         u32 err = vmcs_read32(error_code_field);
4102                         kvm_queue_exception_e(&vmx->vcpu, vector, err);
4103                 } else
4104                         kvm_queue_exception(&vmx->vcpu, vector);
4105                 break;
4106         case INTR_TYPE_SOFT_INTR:
4107                 vmx->vcpu.arch.event_exit_inst_len =
4108                         vmcs_read32(instr_len_field);
4109                 /* fall through */
4110         case INTR_TYPE_EXT_INTR:
4111                 kvm_queue_interrupt(&vmx->vcpu, vector,
4112                         type == INTR_TYPE_SOFT_INTR);
4113                 break;
4114         default:
4115                 break;
4116         }
4117 }
4118
4119 static void vmx_complete_interrupts(struct vcpu_vmx *vmx)
4120 {
4121         __vmx_complete_interrupts(vmx, vmx->idt_vectoring_info,
4122                                   VM_EXIT_INSTRUCTION_LEN,
4123                                   IDT_VECTORING_ERROR_CODE);
4124 }
4125
4126 static void vmx_cancel_injection(struct kvm_vcpu *vcpu)
4127 {
4128         __vmx_complete_interrupts(to_vmx(vcpu),
4129                                   vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
4130                                   VM_ENTRY_INSTRUCTION_LEN,
4131                                   VM_ENTRY_EXCEPTION_ERROR_CODE);
4132
4133         vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);
4134 }
4135
4136 #ifdef CONFIG_X86_64
4137 #define R "r"
4138 #define Q "q"
4139 #else
4140 #define R "e"
4141 #define Q "l"
4142 #endif
4143
4144 static void __noclone vmx_vcpu_run(struct kvm_vcpu *vcpu)
4145 {
4146         struct vcpu_vmx *vmx = to_vmx(vcpu);
4147
4148         /* Record the guest's net vcpu time for enforced NMI injections. */
4149         if (unlikely(!cpu_has_virtual_nmis() && vmx->soft_vnmi_blocked))
4150                 vmx->entry_time = ktime_get();
4151
4152         /* Don't enter VMX if guest state is invalid, let the exit handler
4153            start emulation until we arrive back to a valid state */
4154         if (vmx->emulation_required && emulate_invalid_guest_state)
4155                 return;
4156
4157         if (test_bit(VCPU_REGS_RSP, (unsigned long *)&vcpu->arch.regs_dirty))
4158                 vmcs_writel(GUEST_RSP, vcpu->arch.regs[VCPU_REGS_RSP]);
4159         if (test_bit(VCPU_REGS_RIP, (unsigned long *)&vcpu->arch.regs_dirty))
4160                 vmcs_writel(GUEST_RIP, vcpu->arch.regs[VCPU_REGS_RIP]);
4161
4162         /* When single-stepping over STI and MOV SS, we must clear the
4163          * corresponding interruptibility bits in the guest state. Otherwise
4164          * vmentry fails as it then expects bit 14 (BS) in pending debug
4165          * exceptions being set, but that's not correct for the guest debugging
4166          * case. */
4167         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
4168                 vmx_set_interrupt_shadow(vcpu, 0);
4169
4170         asm(
4171                 /* Store host registers */
4172                 "push %%"R"dx; push %%"R"bp;"
4173                 "push %%"R"cx \n\t" /* placeholder for guest rcx */
4174                 "push %%"R"cx \n\t"
4175                 "cmp %%"R"sp, %c[host_rsp](%0) \n\t"
4176                 "je 1f \n\t"
4177                 "mov %%"R"sp, %c[host_rsp](%0) \n\t"
4178                 __ex(ASM_VMX_VMWRITE_RSP_RDX) "\n\t"
4179                 "1: \n\t"
4180                 /* Reload cr2 if changed */
4181                 "mov %c[cr2](%0), %%"R"ax \n\t"
4182                 "mov %%cr2, %%"R"dx \n\t"
4183                 "cmp %%"R"ax, %%"R"dx \n\t"
4184                 "je 2f \n\t"
4185                 "mov %%"R"ax, %%cr2 \n\t"
4186                 "2: \n\t"
4187                 /* Check if vmlaunch of vmresume is needed */
4188                 "cmpl $0, %c[launched](%0) \n\t"
4189                 /* Load guest registers.  Don't clobber flags. */
4190                 "mov %c[rax](%0), %%"R"ax \n\t"
4191                 "mov %c[rbx](%0), %%"R"bx \n\t"
4192                 "mov %c[rdx](%0), %%"R"dx \n\t"
4193                 "mov %c[rsi](%0), %%"R"si \n\t"
4194                 "mov %c[rdi](%0), %%"R"di \n\t"
4195                 "mov %c[rbp](%0), %%"R"bp \n\t"
4196 #ifdef CONFIG_X86_64
4197                 "mov %c[r8](%0),  %%r8  \n\t"
4198                 "mov %c[r9](%0),  %%r9  \n\t"
4199                 "mov %c[r10](%0), %%r10 \n\t"
4200                 "mov %c[r11](%0), %%r11 \n\t"
4201                 "mov %c[r12](%0), %%r12 \n\t"
4202                 "mov %c[r13](%0), %%r13 \n\t"
4203                 "mov %c[r14](%0), %%r14 \n\t"
4204                 "mov %c[r15](%0), %%r15 \n\t"
4205 #endif
4206                 "mov %c[rcx](%0), %%"R"cx \n\t" /* kills %0 (ecx) */
4207
4208                 /* Enter guest mode */
4209                 "jne .Llaunched \n\t"
4210                 __ex(ASM_VMX_VMLAUNCH) "\n\t"
4211                 "jmp .Lkvm_vmx_return \n\t"
4212                 ".Llaunched: " __ex(ASM_VMX_VMRESUME) "\n\t"
4213                 ".Lkvm_vmx_return: "
4214                 /* Save guest registers, load host registers, keep flags */
4215                 "mov %0, %c[wordsize](%%"R"sp) \n\t"
4216                 "pop %0 \n\t"
4217                 "mov %%"R"ax, %c[rax](%0) \n\t"
4218                 "mov %%"R"bx, %c[rbx](%0) \n\t"
4219                 "pop"Q" %c[rcx](%0) \n\t"
4220                 "mov %%"R"dx, %c[rdx](%0) \n\t"
4221                 "mov %%"R"si, %c[rsi](%0) \n\t"
4222                 "mov %%"R"di, %c[rdi](%0) \n\t"
4223                 "mov %%"R"bp, %c[rbp](%0) \n\t"
4224 #ifdef CONFIG_X86_64
4225                 "mov %%r8,  %c[r8](%0) \n\t"
4226                 "mov %%r9,  %c[r9](%0) \n\t"
4227                 "mov %%r10, %c[r10](%0) \n\t"
4228                 "mov %%r11, %c[r11](%0) \n\t"
4229                 "mov %%r12, %c[r12](%0) \n\t"
4230                 "mov %%r13, %c[r13](%0) \n\t"
4231                 "mov %%r14, %c[r14](%0) \n\t"
4232                 "mov %%r15, %c[r15](%0) \n\t"
4233 #endif
4234                 "mov %%cr2, %%"R"ax   \n\t"
4235                 "mov %%"R"ax, %c[cr2](%0) \n\t"
4236
4237                 "pop  %%"R"bp; pop  %%"R"dx \n\t"
4238                 "setbe %c[fail](%0) \n\t"
4239               : : "c"(vmx), "d"((unsigned long)HOST_RSP),
4240                 [launched]"i"(offsetof(struct vcpu_vmx, launched)),
4241                 [fail]"i"(offsetof(struct vcpu_vmx, fail)),
4242                 [host_rsp]"i"(offsetof(struct vcpu_vmx, host_rsp)),
4243                 [rax]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RAX])),
4244                 [rbx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBX])),
4245                 [rcx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RCX])),
4246                 [rdx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDX])),
4247                 [rsi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RSI])),
4248                 [rdi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDI])),
4249                 [rbp]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBP])),
4250 #ifdef CONFIG_X86_64
4251                 [r8]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R8])),
4252                 [r9]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R9])),
4253                 [r10]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R10])),
4254                 [r11]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R11])),
4255                 [r12]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R12])),
4256                 [r13]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R13])),
4257                 [r14]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R14])),
4258                 [r15]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R15])),
4259 #endif
4260                 [cr2]"i"(offsetof(struct vcpu_vmx, vcpu.arch.cr2)),
4261                 [wordsize]"i"(sizeof(ulong))
4262               : "cc", "memory"
4263                 , R"ax", R"bx", R"di", R"si"
4264 #ifdef CONFIG_X86_64
4265                 , "r8", "r9", "r10", "r11", "r12", "r13", "r14", "r15"
4266 #endif
4267               );
4268
4269         vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP)
4270                                   | (1 << VCPU_EXREG_RFLAGS)
4271                                   | (1 << VCPU_EXREG_CPL)
4272                                   | (1 << VCPU_EXREG_PDPTR)
4273                                   | (1 << VCPU_EXREG_SEGMENTS)
4274                                   | (1 << VCPU_EXREG_CR3));
4275         vcpu->arch.regs_dirty = 0;
4276
4277         vmx->idt_vectoring_info = vmcs_read32(IDT_VECTORING_INFO_FIELD);
4278
4279         asm("mov %0, %%ds; mov %0, %%es" : : "r"(__USER_DS));
4280         vmx->launched = 1;
4281
4282         vmx->exit_reason = vmcs_read32(VM_EXIT_REASON);
4283
4284         vmx_complete_atomic_exit(vmx);
4285         vmx_recover_nmi_blocking(vmx);
4286         vmx_complete_interrupts(vmx);
4287 }
4288
4289 #undef R
4290 #undef Q
4291
4292 static void vmx_free_vmcs(struct kvm_vcpu *vcpu)
4293 {
4294         struct vcpu_vmx *vmx = to_vmx(vcpu);
4295
4296         if (vmx->vmcs) {
4297                 vcpu_clear(vmx);
4298                 free_vmcs(vmx->vmcs);
4299                 vmx->vmcs = NULL;
4300         }
4301 }
4302
4303 static void vmx_free_vcpu(struct kvm_vcpu *vcpu)
4304 {
4305         struct vcpu_vmx *vmx = to_vmx(vcpu);
4306
4307         free_vpid(vmx);
4308         vmx_free_vmcs(vcpu);
4309         kfree(vmx->guest_msrs);
4310         kvm_vcpu_uninit(vcpu);
4311         kmem_cache_free(kvm_vcpu_cache, vmx);
4312 }
4313
4314 static inline void vmcs_init(struct vmcs *vmcs)
4315 {
4316         u64 phys_addr = __pa(per_cpu(vmxarea, raw_smp_processor_id()));
4317
4318         if (!vmm_exclusive)
4319                 kvm_cpu_vmxon(phys_addr);
4320
4321         vmcs_clear(vmcs);
4322
4323         if (!vmm_exclusive)
4324                 kvm_cpu_vmxoff();
4325 }
4326
4327 static struct kvm_vcpu *vmx_create_vcpu(struct kvm *kvm, unsigned int id)
4328 {
4329         int err;
4330         struct vcpu_vmx *vmx = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
4331         int cpu;
4332
4333         if (!vmx)
4334                 return ERR_PTR(-ENOMEM);
4335
4336         allocate_vpid(vmx);
4337
4338         err = kvm_vcpu_init(&vmx->vcpu, kvm, id);
4339         if (err)
4340                 goto free_vcpu;
4341
4342         vmx->guest_msrs = kmalloc(PAGE_SIZE, GFP_KERNEL);
4343         err = -ENOMEM;
4344         if (!vmx->guest_msrs) {
4345                 goto uninit_vcpu;
4346         }
4347
4348         vmx->vmcs = alloc_vmcs();
4349         if (!vmx->vmcs)
4350                 goto free_msrs;
4351
4352         vmcs_init(vmx->vmcs);
4353
4354         cpu = get_cpu();
4355         vmx_vcpu_load(&vmx->vcpu, cpu);
4356         vmx->vcpu.cpu = cpu;
4357         err = vmx_vcpu_setup(vmx);
4358         vmx_vcpu_put(&vmx->vcpu);
4359         put_cpu();
4360         if (err)
4361                 goto free_vmcs;
4362         if (vm_need_virtualize_apic_accesses(kvm))
4363                 err = alloc_apic_access_page(kvm);
4364                 if (err)
4365                         goto free_vmcs;
4366
4367         if (enable_ept) {
4368                 if (!kvm->arch.ept_identity_map_addr)
4369                         kvm->arch.ept_identity_map_addr =
4370                                 VMX_EPT_IDENTITY_PAGETABLE_ADDR;
4371                 err = -ENOMEM;
4372                 if (alloc_identity_pagetable(kvm) != 0)
4373                         goto free_vmcs;
4374                 if (!init_rmode_identity_map(kvm))
4375                         goto free_vmcs;
4376         }
4377
4378         return &vmx->vcpu;
4379
4380 free_vmcs:
4381         free_vmcs(vmx->vmcs);
4382 free_msrs:
4383         kfree(vmx->guest_msrs);
4384 uninit_vcpu:
4385         kvm_vcpu_uninit(&vmx->vcpu);
4386 free_vcpu:
4387         free_vpid(vmx);
4388         kmem_cache_free(kvm_vcpu_cache, vmx);
4389         return ERR_PTR(err);
4390 }
4391
4392 static void __init vmx_check_processor_compat(void *rtn)
4393 {
4394         struct vmcs_config vmcs_conf;
4395
4396         *(int *)rtn = 0;
4397         if (setup_vmcs_config(&vmcs_conf) < 0)
4398                 *(int *)rtn = -EIO;
4399         if (memcmp(&vmcs_config, &vmcs_conf, sizeof(struct vmcs_config)) != 0) {
4400                 printk(KERN_ERR "kvm: CPU %d feature inconsistency!\n",
4401                                 smp_processor_id());
4402                 *(int *)rtn = -EIO;
4403         }
4404 }
4405
4406 static int get_ept_level(void)
4407 {
4408         return VMX_EPT_DEFAULT_GAW + 1;
4409 }
4410
4411 static u64 vmx_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio)
4412 {
4413         u64 ret;
4414
4415         /* For VT-d and EPT combination
4416          * 1. MMIO: always map as UC
4417          * 2. EPT with VT-d:
4418          *   a. VT-d without snooping control feature: can't guarantee the
4419          *      result, try to trust guest.
4420          *   b. VT-d with snooping control feature: snooping control feature of
4421          *      VT-d engine can guarantee the cache correctness. Just set it
4422          *      to WB to keep consistent with host. So the same as item 3.
4423          * 3. EPT without VT-d: always map as WB and set IPAT=1 to keep
4424          *    consistent with host MTRR
4425          */
4426         if (is_mmio)
4427                 ret = MTRR_TYPE_UNCACHABLE << VMX_EPT_MT_EPTE_SHIFT;
4428         else if (vcpu->kvm->arch.iommu_domain &&
4429                 !(vcpu->kvm->arch.iommu_flags & KVM_IOMMU_CACHE_COHERENCY))
4430                 ret = kvm_get_guest_memory_type(vcpu, gfn) <<
4431                       VMX_EPT_MT_EPTE_SHIFT;
4432         else
4433                 ret = (MTRR_TYPE_WRBACK << VMX_EPT_MT_EPTE_SHIFT)
4434                         | VMX_EPT_IPAT_BIT;
4435
4436         return ret;
4437 }
4438
4439 #define _ER(x) { EXIT_REASON_##x, #x }
4440
4441 static const struct trace_print_flags vmx_exit_reasons_str[] = {
4442         _ER(EXCEPTION_NMI),
4443         _ER(EXTERNAL_INTERRUPT),
4444         _ER(TRIPLE_FAULT),
4445         _ER(PENDING_INTERRUPT),
4446         _ER(NMI_WINDOW),
4447         _ER(TASK_SWITCH),
4448         _ER(CPUID),
4449         _ER(HLT),
4450         _ER(INVLPG),
4451         _ER(RDPMC),
4452         _ER(RDTSC),
4453         _ER(VMCALL),
4454         _ER(VMCLEAR),
4455         _ER(VMLAUNCH),
4456         _ER(VMPTRLD),
4457         _ER(VMPTRST),
4458         _ER(VMREAD),
4459         _ER(VMRESUME),
4460         _ER(VMWRITE),
4461         _ER(VMOFF),
4462         _ER(VMON),
4463         _ER(CR_ACCESS),
4464         _ER(DR_ACCESS),
4465         _ER(IO_INSTRUCTION),
4466         _ER(MSR_READ),
4467         _ER(MSR_WRITE),
4468         _ER(MWAIT_INSTRUCTION),
4469         _ER(MONITOR_INSTRUCTION),
4470         _ER(PAUSE_INSTRUCTION),
4471         _ER(MCE_DURING_VMENTRY),
4472         _ER(TPR_BELOW_THRESHOLD),
4473         _ER(APIC_ACCESS),
4474         _ER(EPT_VIOLATION),
4475         _ER(EPT_MISCONFIG),
4476         _ER(WBINVD),
4477         { -1, NULL }
4478 };
4479
4480 #undef _ER
4481
4482 static int vmx_get_lpage_level(void)
4483 {
4484         if (enable_ept && !cpu_has_vmx_ept_1g_page())
4485                 return PT_DIRECTORY_LEVEL;
4486         else
4487                 /* For shadow and EPT supported 1GB page */
4488                 return PT_PDPE_LEVEL;
4489 }
4490
4491 static void vmx_cpuid_update(struct kvm_vcpu *vcpu)
4492 {
4493         struct kvm_cpuid_entry2 *best;
4494         struct vcpu_vmx *vmx = to_vmx(vcpu);
4495         u32 exec_control;
4496
4497         vmx->rdtscp_enabled = false;
4498         if (vmx_rdtscp_supported()) {
4499                 exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
4500                 if (exec_control & SECONDARY_EXEC_RDTSCP) {
4501                         best = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
4502                         if (best && (best->edx & bit(X86_FEATURE_RDTSCP)))
4503                                 vmx->rdtscp_enabled = true;
4504                         else {
4505                                 exec_control &= ~SECONDARY_EXEC_RDTSCP;
4506                                 vmcs_write32(SECONDARY_VM_EXEC_CONTROL,
4507                                                 exec_control);
4508                         }
4509                 }
4510         }
4511 }
4512
4513 static void vmx_set_supported_cpuid(u32 func, struct kvm_cpuid_entry2 *entry)
4514 {
4515 }
4516
4517 static int vmx_check_intercept(struct kvm_vcpu *vcpu,
4518                                struct x86_instruction_info *info,
4519                                enum x86_intercept_stage stage)
4520 {
4521         return X86EMUL_CONTINUE;
4522 }
4523
4524 static struct kvm_x86_ops vmx_x86_ops = {
4525         .cpu_has_kvm_support = cpu_has_kvm_support,
4526         .disabled_by_bios = vmx_disabled_by_bios,
4527         .hardware_setup = hardware_setup,
4528         .hardware_unsetup = hardware_unsetup,
4529         .check_processor_compatibility = vmx_check_processor_compat,
4530         .hardware_enable = hardware_enable,
4531         .hardware_disable = hardware_disable,
4532         .cpu_has_accelerated_tpr = report_flexpriority,
4533
4534         .vcpu_create = vmx_create_vcpu,
4535         .vcpu_free = vmx_free_vcpu,
4536         .vcpu_reset = vmx_vcpu_reset,
4537
4538         .prepare_guest_switch = vmx_save_host_state,
4539         .vcpu_load = vmx_vcpu_load,
4540         .vcpu_put = vmx_vcpu_put,
4541
4542         .set_guest_debug = set_guest_debug,
4543         .get_msr = vmx_get_msr,
4544         .set_msr = vmx_set_msr,
4545         .get_segment_base = vmx_get_segment_base,
4546         .get_segment = vmx_get_segment,
4547         .set_segment = vmx_set_segment,
4548         .get_cpl = vmx_get_cpl,
4549         .get_cs_db_l_bits = vmx_get_cs_db_l_bits,
4550         .decache_cr0_guest_bits = vmx_decache_cr0_guest_bits,
4551         .decache_cr3 = vmx_decache_cr3,
4552         .decache_cr4_guest_bits = vmx_decache_cr4_guest_bits,
4553         .set_cr0 = vmx_set_cr0,
4554         .set_cr3 = vmx_set_cr3,
4555         .set_cr4 = vmx_set_cr4,
4556         .set_efer = vmx_set_efer,
4557         .get_idt = vmx_get_idt,
4558         .set_idt = vmx_set_idt,
4559         .get_gdt = vmx_get_gdt,
4560         .set_gdt = vmx_set_gdt,
4561         .set_dr7 = vmx_set_dr7,
4562         .cache_reg = vmx_cache_reg,
4563         .get_rflags = vmx_get_rflags,
4564         .set_rflags = vmx_set_rflags,
4565         .fpu_activate = vmx_fpu_activate,
4566         .fpu_deactivate = vmx_fpu_deactivate,
4567
4568         .tlb_flush = vmx_flush_tlb,
4569
4570         .run = vmx_vcpu_run,
4571         .handle_exit = vmx_handle_exit,
4572         .skip_emulated_instruction = skip_emulated_instruction,
4573         .set_interrupt_shadow = vmx_set_interrupt_shadow,
4574         .get_interrupt_shadow = vmx_get_interrupt_shadow,
4575         .patch_hypercall = vmx_patch_hypercall,
4576         .set_irq = vmx_inject_irq,
4577         .set_nmi = vmx_inject_nmi,
4578         .queue_exception = vmx_queue_exception,
4579         .cancel_injection = vmx_cancel_injection,
4580         .interrupt_allowed = vmx_interrupt_allowed,
4581         .nmi_allowed = vmx_nmi_allowed,
4582         .get_nmi_mask = vmx_get_nmi_mask,
4583         .set_nmi_mask = vmx_set_nmi_mask,
4584         .enable_nmi_window = enable_nmi_window,
4585         .enable_irq_window = enable_irq_window,
4586         .update_cr8_intercept = update_cr8_intercept,
4587
4588         .set_tss_addr = vmx_set_tss_addr,
4589         .get_tdp_level = get_ept_level,
4590         .get_mt_mask = vmx_get_mt_mask,
4591
4592         .get_exit_info = vmx_get_exit_info,
4593         .exit_reasons_str = vmx_exit_reasons_str,
4594
4595         .get_lpage_level = vmx_get_lpage_level,
4596
4597         .cpuid_update = vmx_cpuid_update,
4598
4599         .rdtscp_supported = vmx_rdtscp_supported,
4600
4601         .set_supported_cpuid = vmx_set_supported_cpuid,
4602
4603         .has_wbinvd_exit = cpu_has_vmx_wbinvd_exit,
4604
4605         .set_tsc_khz = vmx_set_tsc_khz,
4606         .write_tsc_offset = vmx_write_tsc_offset,
4607         .adjust_tsc_offset = vmx_adjust_tsc_offset,
4608         .compute_tsc_offset = vmx_compute_tsc_offset,
4609
4610         .set_tdp_cr3 = vmx_set_cr3,
4611
4612         .check_intercept = vmx_check_intercept,
4613 };
4614
4615 static int __init vmx_init(void)
4616 {
4617         int r, i;
4618
4619         rdmsrl_safe(MSR_EFER, &host_efer);
4620
4621         for (i = 0; i < NR_VMX_MSR; ++i)
4622                 kvm_define_shared_msr(i, vmx_msr_index[i]);
4623
4624         vmx_io_bitmap_a = (unsigned long *)__get_free_page(GFP_KERNEL);
4625         if (!vmx_io_bitmap_a)
4626                 return -ENOMEM;
4627
4628         vmx_io_bitmap_b = (unsigned long *)__get_free_page(GFP_KERNEL);
4629         if (!vmx_io_bitmap_b) {
4630                 r = -ENOMEM;
4631                 goto out;
4632         }
4633
4634         vmx_msr_bitmap_legacy = (unsigned long *)__get_free_page(GFP_KERNEL);
4635         if (!vmx_msr_bitmap_legacy) {
4636                 r = -ENOMEM;
4637                 goto out1;
4638         }
4639
4640         vmx_msr_bitmap_longmode = (unsigned long *)__get_free_page(GFP_KERNEL);
4641         if (!vmx_msr_bitmap_longmode) {
4642                 r = -ENOMEM;
4643                 goto out2;
4644         }
4645
4646         /*
4647          * Allow direct access to the PC debug port (it is often used for I/O
4648          * delays, but the vmexits simply slow things down).
4649          */
4650         memset(vmx_io_bitmap_a, 0xff, PAGE_SIZE);
4651         clear_bit(0x80, vmx_io_bitmap_a);
4652
4653         memset(vmx_io_bitmap_b, 0xff, PAGE_SIZE);
4654
4655         memset(vmx_msr_bitmap_legacy, 0xff, PAGE_SIZE);
4656         memset(vmx_msr_bitmap_longmode, 0xff, PAGE_SIZE);
4657
4658         set_bit(0, vmx_vpid_bitmap); /* 0 is reserved for host */
4659
4660         r = kvm_init(&vmx_x86_ops, sizeof(struct vcpu_vmx),
4661                      __alignof__(struct vcpu_vmx), THIS_MODULE);
4662         if (r)
4663                 goto out3;
4664
4665         vmx_disable_intercept_for_msr(MSR_FS_BASE, false);
4666         vmx_disable_intercept_for_msr(MSR_GS_BASE, false);
4667         vmx_disable_intercept_for_msr(MSR_KERNEL_GS_BASE, true);
4668         vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_CS, false);
4669         vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_ESP, false);
4670         vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_EIP, false);
4671
4672         if (enable_ept) {
4673                 bypass_guest_pf = 0;
4674                 kvm_mmu_set_mask_ptes(0ull, 0ull, 0ull, 0ull,
4675                                 VMX_EPT_EXECUTABLE_MASK);
4676                 kvm_enable_tdp();
4677         } else
4678                 kvm_disable_tdp();
4679
4680         if (bypass_guest_pf)
4681                 kvm_mmu_set_nonpresent_ptes(~0xffeull, 0ull);
4682
4683         return 0;
4684
4685 out3:
4686         free_page((unsigned long)vmx_msr_bitmap_longmode);
4687 out2:
4688         free_page((unsigned long)vmx_msr_bitmap_legacy);
4689 out1:
4690         free_page((unsigned long)vmx_io_bitmap_b);
4691 out:
4692         free_page((unsigned long)vmx_io_bitmap_a);
4693         return r;
4694 }
4695
4696 static void __exit vmx_exit(void)
4697 {
4698         free_page((unsigned long)vmx_msr_bitmap_legacy);
4699         free_page((unsigned long)vmx_msr_bitmap_longmode);
4700         free_page((unsigned long)vmx_io_bitmap_b);
4701         free_page((unsigned long)vmx_io_bitmap_a);
4702
4703         kvm_exit();
4704 }
4705
4706 module_init(vmx_init)
4707 module_exit(vmx_exit)