arch/x86/kvm/vmx.c

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