arch/x86/kvm/vmx/vmx.h

   1 /* SPDX-License-Identifier: GPL-2.0 */
   2 #ifndef __KVM_X86_VMX_H
   3 #define __KVM_X86_VMX_H
   4
   5 #include <linux/kvm_host.h>
   6
   7 #include <asm/kvm.h>
   8 #include <asm/intel_pt.h>
   9 #include <asm/perf_event.h>
  10 #include <asm/posted_intr.h>
  11
  12 #include "capabilities.h"
  13 #include "../kvm_cache_regs.h"
  14 #include "vmcs.h"
  15 #include "vmx_ops.h"
  16 #include "../cpuid.h"
  17 #include "run_flags.h"
  18 #include "../mmu.h"
  19
  20 #define X2APIC_MSR(r) (APIC_BASE_MSR + ((r) >> 4))
  21
  22 #ifdef CONFIG_X86_64
  23 #define MAX_NR_USER_RETURN_MSRS 7
  24 #else
  25 #define MAX_NR_USER_RETURN_MSRS 4
  26 #endif
  27
  28 #define MAX_NR_LOADSTORE_MSRS   8
  29
  30 struct vmx_msrs {
  31         unsigned int            nr;
  32         struct vmx_msr_entry    val[MAX_NR_LOADSTORE_MSRS];
  33 };
  34
  35 struct vmx_uret_msr {
  36         bool load_into_hardware;
  37         u64 data;
  38         u64 mask;
  39 };
  40
  41 enum segment_cache_field {
  42         SEG_FIELD_SEL = 0,
  43         SEG_FIELD_BASE = 1,
  44         SEG_FIELD_LIMIT = 2,
  45         SEG_FIELD_AR = 3,
  46
  47         SEG_FIELD_NR = 4
  48 };
  49
  50 #define RTIT_ADDR_RANGE         4
  51
  52 struct pt_ctx {
  53         u64 ctl;
  54         u64 status;
  55         u64 output_base;
  56         u64 output_mask;
  57         u64 cr3_match;
  58         u64 addr_a[RTIT_ADDR_RANGE];
  59         u64 addr_b[RTIT_ADDR_RANGE];
  60 };
  61
  62 struct pt_desc {
  63         u64 ctl_bitmask;
  64         u32 num_address_ranges;
  65         u32 caps[PT_CPUID_REGS_NUM * PT_CPUID_LEAVES];
  66         struct pt_ctx host;
  67         struct pt_ctx guest;
  68 };
  69
  70 union vmx_exit_reason {
  71         struct {
  72                 u32     basic                   : 16;
  73                 u32     reserved16              : 1;
  74                 u32     reserved17              : 1;
  75                 u32     reserved18              : 1;
  76                 u32     reserved19              : 1;
  77                 u32     reserved20              : 1;
  78                 u32     reserved21              : 1;
  79                 u32     reserved22              : 1;
  80                 u32     reserved23              : 1;
  81                 u32     reserved24              : 1;
  82                 u32     reserved25              : 1;
  83                 u32     bus_lock_detected       : 1;
  84                 u32     enclave_mode            : 1;
  85                 u32     smi_pending_mtf         : 1;
  86                 u32     smi_from_vmx_root       : 1;
  87                 u32     reserved30              : 1;
  88                 u32     failed_vmentry          : 1;
  89         };
  90         u32 full;
  91 };
  92
  93 struct lbr_desc {
  94         /* Basic info about guest LBR records. */
  95         struct x86_pmu_lbr records;
  96
  97         /*
  98          * Emulate LBR feature via passthrough LBR registers when the
  99          * per-vcpu guest LBR event is scheduled on the current pcpu.
 100          *
 101          * The records may be inaccurate if the host reclaims the LBR.
 102          */
 103         struct perf_event *event;
 104
 105         /* True if LBRs are marked as not intercepted in the MSR bitmap */
 106         bool msr_passthrough;
 107 };
 108
 109 extern struct x86_pmu_lbr vmx_lbr_caps;
 110
 111 /*
 112  * The nested_vmx structure is part of vcpu_vmx, and holds information we need
 113  * for correct emulation of VMX (i.e., nested VMX) on this vcpu.
 114  */
 115 struct nested_vmx {
 116         /* Has the level1 guest done vmxon? */
 117         bool vmxon;
 118         gpa_t vmxon_ptr;
 119         bool pml_full;
 120
 121         /* The guest-physical address of the current VMCS L1 keeps for L2 */
 122         gpa_t current_vmptr;
 123         /*
 124          * Cache of the guest's VMCS, existing outside of guest memory.
 125          * Loaded from guest memory during VMPTRLD. Flushed to guest
 126          * memory during VMCLEAR and VMPTRLD.
 127          */
 128         struct vmcs12 *cached_vmcs12;
 129         /*
 130          * Cache of the guest's shadow VMCS, existing outside of guest
 131          * memory. Loaded from guest memory during VM entry. Flushed
 132          * to guest memory during VM exit.
 133          */
 134         struct vmcs12 *cached_shadow_vmcs12;
 135
 136         /*
 137          * GPA to HVA cache for accessing vmcs12->vmcs_link_pointer
 138          */
 139         struct gfn_to_hva_cache shadow_vmcs12_cache;
 140
 141         /*
 142          * GPA to HVA cache for VMCS12
 143          */
 144         struct gfn_to_hva_cache vmcs12_cache;
 145
 146         /*
 147          * Indicates if the shadow vmcs or enlightened vmcs must be updated
 148          * with the data held by struct vmcs12.
 149          */
 150         bool need_vmcs12_to_shadow_sync;
 151         bool dirty_vmcs12;
 152
 153         /*
 154          * Indicates whether MSR bitmap for L2 needs to be rebuilt due to
 155          * changes in MSR bitmap for L1 or switching to a different L2. Note,
 156          * this flag can only be used reliably in conjunction with a paravirt L1
 157          * which informs L0 whether any changes to MSR bitmap for L2 were done
 158          * on its side.
 159          */
 160         bool force_msr_bitmap_recalc;
 161
 162         /*
 163          * Indicates lazily loaded guest state has not yet been decached from
 164          * vmcs02.
 165          */
 166         bool need_sync_vmcs02_to_vmcs12_rare;
 167
 168         /*
 169          * vmcs02 has been initialized, i.e. state that is constant for
 170          * vmcs02 has been written to the backing VMCS.  Initialization
 171          * is delayed until L1 actually attempts to run a nested VM.
 172          */
 173         bool vmcs02_initialized;
 174
 175         bool change_vmcs01_virtual_apic_mode;
 176         bool reload_vmcs01_apic_access_page;
 177         bool update_vmcs01_cpu_dirty_logging;
 178         bool update_vmcs01_apicv_status;
 179
 180         /*
 181          * Enlightened VMCS has been enabled. It does not mean that L1 has to
 182          * use it. However, VMX features available to L1 will be limited based
 183          * on what the enlightened VMCS supports.
 184          */
 185         bool enlightened_vmcs_enabled;
 186
 187         /* L2 must run next, and mustn't decide to exit to L1. */
 188         bool nested_run_pending;
 189
 190         /* Pending MTF VM-exit into L1.  */
 191         bool mtf_pending;
 192
 193         struct loaded_vmcs vmcs02;
 194
 195         /*
 196          * Guest pages referred to in the vmcs02 with host-physical
 197          * pointers, so we must keep them pinned while L2 runs.
 198          */
 199         struct kvm_host_map apic_access_page_map;
 200         struct kvm_host_map virtual_apic_map;
 201         struct kvm_host_map pi_desc_map;
 202
 203         struct pi_desc *pi_desc;
 204         bool pi_pending;
 205         u16 posted_intr_nv;
 206
 207         struct hrtimer preemption_timer;
 208         u64 preemption_timer_deadline;
 209         bool has_preemption_timer_deadline;
 210         bool preemption_timer_expired;
 211
 212         /*
 213          * Used to snapshot MSRs that are conditionally loaded on VM-Enter in
 214          * order to propagate the guest's pre-VM-Enter value into vmcs02.  For
 215          * emulation of VMLAUNCH/VMRESUME, the snapshot will be of L1's value.
 216          * For KVM_SET_NESTED_STATE, the snapshot is of L2's value, _if_
 217          * userspace restores MSRs before nested state.  If userspace restores
 218          * MSRs after nested state, the snapshot holds garbage, but KVM can't
 219          * detect that, and the garbage value in vmcs02 will be overwritten by
 220          * MSR restoration in any case.
 221          */
 222         u64 pre_vmenter_debugctl;
 223         u64 pre_vmenter_bndcfgs;
 224
 225         /* to migrate it to L1 if L2 writes to L1's CR8 directly */
 226         int l1_tpr_threshold;
 227
 228         u16 vpid02;
 229         u16 last_vpid;
 230
 231         struct nested_vmx_msrs msrs;
 232
 233         /* SMM related state */
 234         struct {
 235                 /* in VMX operation on SMM entry? */
 236                 bool vmxon;
 237                 /* in guest mode on SMM entry? */
 238                 bool guest_mode;
 239         } smm;
 240
 241 #ifdef CONFIG_KVM_HYPERV
 242         gpa_t hv_evmcs_vmptr;
 243         struct kvm_host_map hv_evmcs_map;
 244         struct hv_enlightened_vmcs *hv_evmcs;
 245 #endif
 246 };
 247
 248 struct vcpu_vmx {
 249         struct kvm_vcpu       vcpu;
 250         u8                    fail;
 251         u8                    x2apic_msr_bitmap_mode;
 252
 253         /*
 254          * If true, host state has been stored in vmx->loaded_vmcs for
 255          * the CPU registers that only need to be switched when transitioning
 256          * to/from the kernel, and the registers have been loaded with guest
 257          * values.  If false, host state is loaded in the CPU registers
 258          * and vmx->loaded_vmcs->host_state is invalid.
 259          */
 260         bool                  guest_state_loaded;
 261
 262         unsigned long         exit_qualification;
 263         u32                   exit_intr_info;
 264         u32                   idt_vectoring_info;
 265         ulong                 rflags;
 266
 267         /*
 268          * User return MSRs are always emulated when enabled in the guest, but
 269          * only loaded into hardware when necessary, e.g. SYSCALL #UDs outside
 270          * of 64-bit mode or if EFER.SCE=1, thus the SYSCALL MSRs don't need to
 271          * be loaded into hardware if those conditions aren't met.
 272          */
 273         struct vmx_uret_msr   guest_uret_msrs[MAX_NR_USER_RETURN_MSRS];
 274         bool                  guest_uret_msrs_loaded;
 275 #ifdef CONFIG_X86_64
 276         u64                   msr_host_kernel_gs_base;
 277         u64                   msr_guest_kernel_gs_base;
 278 #endif
 279
 280         u64                   spec_ctrl;
 281         u32                   msr_ia32_umwait_control;
 282
 283         /*
 284          * loaded_vmcs points to the VMCS currently used in this vcpu. For a
 285          * non-nested (L1) guest, it always points to vmcs01. For a nested
 286          * guest (L2), it points to a different VMCS.
 287          */
 288         struct loaded_vmcs    vmcs01;
 289         struct loaded_vmcs   *loaded_vmcs;
 290
 291         struct msr_autoload {
 292                 struct vmx_msrs guest;
 293                 struct vmx_msrs host;
 294         } msr_autoload;
 295
 296         struct msr_autostore {
 297                 struct vmx_msrs guest;
 298         } msr_autostore;
 299
 300         struct {
 301                 int vm86_active;
 302                 ulong save_rflags;
 303                 struct kvm_segment segs[8];
 304         } rmode;
 305         struct {
 306                 u32 bitmask; /* 4 bits per segment (1 bit per field) */
 307                 struct kvm_save_segment {
 308                         u16 selector;
 309                         unsigned long base;
 310                         u32 limit;
 311                         u32 ar;
 312                 } seg[8];
 313         } segment_cache;
 314         int vpid;
 315         bool emulation_required;
 316
 317         union vmx_exit_reason exit_reason;
 318
 319         /* Posted interrupt descriptor */
 320         struct pi_desc pi_desc;
 321
 322         /* Used if this vCPU is waiting for PI notification wakeup. */
 323         struct list_head pi_wakeup_list;
 324
 325         /* Support for a guest hypervisor (nested VMX) */
 326         struct nested_vmx nested;
 327
 328         /* Dynamic PLE window. */
 329         unsigned int ple_window;
 330         bool ple_window_dirty;
 331
 332         /* Support for PML */
 333 #define PML_ENTITY_NUM          512
 334         struct page *pml_pg;
 335
 336         /* apic deadline value in host tsc */
 337         u64 hv_deadline_tsc;
 338
 339         unsigned long host_debugctlmsr;
 340
 341         /*
 342          * Only bits masked by msr_ia32_feature_control_valid_bits can be set in
 343          * msr_ia32_feature_control. FEAT_CTL_LOCKED is always included
 344          * in msr_ia32_feature_control_valid_bits.
 345          */
 346         u64 msr_ia32_feature_control;
 347         u64 msr_ia32_feature_control_valid_bits;
 348         /* SGX Launch Control public key hash */
 349         u64 msr_ia32_sgxlepubkeyhash[4];
 350         u64 msr_ia32_mcu_opt_ctrl;
 351         bool disable_fb_clear;
 352
 353         struct pt_desc pt_desc;
 354         struct lbr_desc lbr_desc;
 355
 356         /* Save desired MSR intercept (read: pass-through) state */
 357 #define MAX_POSSIBLE_PASSTHROUGH_MSRS   16
 358         struct {
 359                 DECLARE_BITMAP(read, MAX_POSSIBLE_PASSTHROUGH_MSRS);
 360                 DECLARE_BITMAP(write, MAX_POSSIBLE_PASSTHROUGH_MSRS);
 361         } shadow_msr_intercept;
 362
 363         /* ve_info must be page aligned. */
 364         struct vmx_ve_information *ve_info;
 365 };
 366
 367 struct kvm_vmx {
 368         struct kvm kvm;
 369
 370         unsigned int tss_addr;
 371         bool ept_identity_pagetable_done;
 372         gpa_t ept_identity_map_addr;
 373         /* Posted Interrupt Descriptor (PID) table for IPI virtualization */
 374         u64 *pid_table;
 375 };
 376
 377 void vmx_vcpu_load_vmcs(struct kvm_vcpu *vcpu, int cpu,
 378                         struct loaded_vmcs *buddy);
 379 int allocate_vpid(void);
 380 void free_vpid(int vpid);
 381 void vmx_set_constant_host_state(struct vcpu_vmx *vmx);
 382 void vmx_prepare_switch_to_guest(struct kvm_vcpu *vcpu);
 383 void vmx_set_host_fs_gs(struct vmcs_host_state *host, u16 fs_sel, u16 gs_sel,
 384                         unsigned long fs_base, unsigned long gs_base);
 385 int vmx_get_cpl(struct kvm_vcpu *vcpu);
 386 int vmx_get_cpl_no_cache(struct kvm_vcpu *vcpu);
 387 bool vmx_emulation_required(struct kvm_vcpu *vcpu);
 388 unsigned long vmx_get_rflags(struct kvm_vcpu *vcpu);
 389 void vmx_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags);
 390 u32 vmx_get_interrupt_shadow(struct kvm_vcpu *vcpu);
 391 void vmx_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask);
 392 int vmx_set_efer(struct kvm_vcpu *vcpu, u64 efer);
 393 void vmx_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0);
 394 void vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4);
 395 void set_cr4_guest_host_mask(struct vcpu_vmx *vmx);
 396 void ept_save_pdptrs(struct kvm_vcpu *vcpu);
 397 void vmx_get_segment(struct kvm_vcpu *vcpu, struct kvm_segment *var, int seg);
 398 void __vmx_set_segment(struct kvm_vcpu *vcpu, struct kvm_segment *var, int seg);
 399 u64 construct_eptp(struct kvm_vcpu *vcpu, hpa_t root_hpa, int root_level);
 400
 401 bool vmx_guest_inject_ac(struct kvm_vcpu *vcpu);
 402 void vmx_update_exception_bitmap(struct kvm_vcpu *vcpu);
 403 bool vmx_nmi_blocked(struct kvm_vcpu *vcpu);
 404 bool __vmx_interrupt_blocked(struct kvm_vcpu *vcpu);
 405 bool vmx_interrupt_blocked(struct kvm_vcpu *vcpu);
 406 bool vmx_get_nmi_mask(struct kvm_vcpu *vcpu);
 407 void vmx_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked);
 408 void vmx_set_virtual_apic_mode(struct kvm_vcpu *vcpu);
 409 struct vmx_uret_msr *vmx_find_uret_msr(struct vcpu_vmx *vmx, u32 msr);
 410 void pt_update_intercept_for_msr(struct kvm_vcpu *vcpu);
 411 void vmx_update_host_rsp(struct vcpu_vmx *vmx, unsigned long host_rsp);
 412 void vmx_spec_ctrl_restore_host(struct vcpu_vmx *vmx, unsigned int flags);
 413 unsigned int __vmx_vcpu_run_flags(struct vcpu_vmx *vmx);
 414 bool __vmx_vcpu_run(struct vcpu_vmx *vmx, unsigned long *regs,
 415                     unsigned int flags);
 416 int vmx_find_loadstore_msr_slot(struct vmx_msrs *m, u32 msr);
 417 void vmx_ept_load_pdptrs(struct kvm_vcpu *vcpu);
 418
 419 void vmx_disable_intercept_for_msr(struct kvm_vcpu *vcpu, u32 msr, int type);
 420 void vmx_enable_intercept_for_msr(struct kvm_vcpu *vcpu, u32 msr, int type);
 421
 422 u64 vmx_get_l2_tsc_offset(struct kvm_vcpu *vcpu);
 423 u64 vmx_get_l2_tsc_multiplier(struct kvm_vcpu *vcpu);
 424
 425 gva_t vmx_get_untagged_addr(struct kvm_vcpu *vcpu, gva_t gva, unsigned int flags);
 426
 427 static inline void vmx_set_intercept_for_msr(struct kvm_vcpu *vcpu, u32 msr,
 428                                              int type, bool value)
 429 {
 430         if (value)
 431                 vmx_enable_intercept_for_msr(vcpu, msr, type);
 432         else
 433                 vmx_disable_intercept_for_msr(vcpu, msr, type);
 434 }
 435
 436 void vmx_update_cpu_dirty_logging(struct kvm_vcpu *vcpu);
 437
 438 /*
 439  * Note, early Intel manuals have the write-low and read-high bitmap offsets
 440  * the wrong way round.  The bitmaps control MSRs 0x00000000-0x00001fff and
 441  * 0xc0000000-0xc0001fff.  The former (low) uses bytes 0-0x3ff for reads and
 442  * 0x800-0xbff for writes.  The latter (high) uses 0x400-0x7ff for reads and
 443  * 0xc00-0xfff for writes.  MSRs not covered by either of the ranges always
 444  * VM-Exit.
 445  */
 446 #define __BUILD_VMX_MSR_BITMAP_HELPER(rtype, action, bitop, access, base)      \
 447 static inline rtype vmx_##action##_msr_bitmap_##access(unsigned long *bitmap,  \
 448                                                        u32 msr)                \
 449 {                                                                              \
 450         int f = sizeof(unsigned long);                                         \
 451                                                                                \
 452         if (msr <= 0x1fff)                                                     \
 453                 return bitop##_bit(msr, bitmap + base / f);                    \
 454         else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff))                   \
 455                 return bitop##_bit(msr & 0x1fff, bitmap + (base + 0x400) / f); \
 456         return (rtype)true;                                                    \
 457 }
 458 #define BUILD_VMX_MSR_BITMAP_HELPERS(ret_type, action, bitop)                  \
 459         __BUILD_VMX_MSR_BITMAP_HELPER(ret_type, action, bitop, read,  0x0)     \
 460         __BUILD_VMX_MSR_BITMAP_HELPER(ret_type, action, bitop, write, 0x800)
 461
 462 BUILD_VMX_MSR_BITMAP_HELPERS(bool, test, test)
 463 BUILD_VMX_MSR_BITMAP_HELPERS(void, clear, __clear)
 464 BUILD_VMX_MSR_BITMAP_HELPERS(void, set, __set)
 465
 466 static inline u8 vmx_get_rvi(void)
 467 {
 468         return vmcs_read16(GUEST_INTR_STATUS) & 0xff;
 469 }
 470
 471 #define __KVM_REQUIRED_VMX_VM_ENTRY_CONTROLS                            \
 472         (VM_ENTRY_LOAD_DEBUG_CONTROLS)
 473 #ifdef CONFIG_X86_64
 474         #define KVM_REQUIRED_VMX_VM_ENTRY_CONTROLS                      \
 475                 (__KVM_REQUIRED_VMX_VM_ENTRY_CONTROLS |                 \
 476                  VM_ENTRY_IA32E_MODE)
 477 #else
 478         #define KVM_REQUIRED_VMX_VM_ENTRY_CONTROLS                      \
 479                 __KVM_REQUIRED_VMX_VM_ENTRY_CONTROLS
 480 #endif
 481 #define KVM_OPTIONAL_VMX_VM_ENTRY_CONTROLS                              \
 482         (VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL |                          \
 483          VM_ENTRY_LOAD_IA32_PAT |                                       \
 484          VM_ENTRY_LOAD_IA32_EFER |                                      \
 485          VM_ENTRY_LOAD_BNDCFGS |                                        \
 486          VM_ENTRY_PT_CONCEAL_PIP |                                      \
 487          VM_ENTRY_LOAD_IA32_RTIT_CTL)
 488
 489 #define __KVM_REQUIRED_VMX_VM_EXIT_CONTROLS                             \
 490         (VM_EXIT_SAVE_DEBUG_CONTROLS |                                  \
 491          VM_EXIT_ACK_INTR_ON_EXIT)
 492 #ifdef CONFIG_X86_64
 493         #define KVM_REQUIRED_VMX_VM_EXIT_CONTROLS                       \
 494                 (__KVM_REQUIRED_VMX_VM_EXIT_CONTROLS |                  \
 495                  VM_EXIT_HOST_ADDR_SPACE_SIZE)
 496 #else
 497         #define KVM_REQUIRED_VMX_VM_EXIT_CONTROLS                       \
 498                 __KVM_REQUIRED_VMX_VM_EXIT_CONTROLS
 499 #endif
 500 #define KVM_OPTIONAL_VMX_VM_EXIT_CONTROLS                               \
 501               (VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL |                     \
 502                VM_EXIT_SAVE_IA32_PAT |                                  \
 503                VM_EXIT_LOAD_IA32_PAT |                                  \
 504                VM_EXIT_SAVE_IA32_EFER |                                 \
 505                VM_EXIT_SAVE_VMX_PREEMPTION_TIMER |                      \
 506                VM_EXIT_LOAD_IA32_EFER |                                 \
 507                VM_EXIT_CLEAR_BNDCFGS |                                  \
 508                VM_EXIT_PT_CONCEAL_PIP |                                 \
 509                VM_EXIT_CLEAR_IA32_RTIT_CTL)
 510
 511 #define KVM_REQUIRED_VMX_PIN_BASED_VM_EXEC_CONTROL                      \
 512         (PIN_BASED_EXT_INTR_MASK |                                      \
 513          PIN_BASED_NMI_EXITING)
 514 #define KVM_OPTIONAL_VMX_PIN_BASED_VM_EXEC_CONTROL                      \
 515         (PIN_BASED_VIRTUAL_NMIS |                                       \
 516          PIN_BASED_POSTED_INTR |                                        \
 517          PIN_BASED_VMX_PREEMPTION_TIMER)
 518
 519 #define __KVM_REQUIRED_VMX_CPU_BASED_VM_EXEC_CONTROL                    \
 520         (CPU_BASED_HLT_EXITING |                                        \
 521          CPU_BASED_CR3_LOAD_EXITING |                                   \
 522          CPU_BASED_CR3_STORE_EXITING |                                  \
 523          CPU_BASED_UNCOND_IO_EXITING |                                  \
 524          CPU_BASED_MOV_DR_EXITING |                                     \
 525          CPU_BASED_USE_TSC_OFFSETTING |                                 \
 526          CPU_BASED_MWAIT_EXITING |                                      \
 527          CPU_BASED_MONITOR_EXITING |                                    \
 528          CPU_BASED_INVLPG_EXITING |                                     \
 529          CPU_BASED_RDPMC_EXITING |                                      \
 530          CPU_BASED_INTR_WINDOW_EXITING)
 531
 532 #ifdef CONFIG_X86_64
 533         #define KVM_REQUIRED_VMX_CPU_BASED_VM_EXEC_CONTROL              \
 534                 (__KVM_REQUIRED_VMX_CPU_BASED_VM_EXEC_CONTROL |         \
 535                  CPU_BASED_CR8_LOAD_EXITING |                           \
 536                  CPU_BASED_CR8_STORE_EXITING)
 537 #else
 538         #define KVM_REQUIRED_VMX_CPU_BASED_VM_EXEC_CONTROL              \
 539                 __KVM_REQUIRED_VMX_CPU_BASED_VM_EXEC_CONTROL
 540 #endif
 541
 542 #define KVM_OPTIONAL_VMX_CPU_BASED_VM_EXEC_CONTROL                      \
 543         (CPU_BASED_RDTSC_EXITING |                                      \
 544          CPU_BASED_TPR_SHADOW |                                         \
 545          CPU_BASED_USE_IO_BITMAPS |                                     \
 546          CPU_BASED_MONITOR_TRAP_FLAG |                                  \
 547          CPU_BASED_USE_MSR_BITMAPS |                                    \
 548          CPU_BASED_NMI_WINDOW_EXITING |                                 \
 549          CPU_BASED_PAUSE_EXITING |                                      \
 550          CPU_BASED_ACTIVATE_SECONDARY_CONTROLS |                        \
 551          CPU_BASED_ACTIVATE_TERTIARY_CONTROLS)
 552
 553 #define KVM_REQUIRED_VMX_SECONDARY_VM_EXEC_CONTROL 0
 554 #define KVM_OPTIONAL_VMX_SECONDARY_VM_EXEC_CONTROL                      \
 555         (SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |                      \
 556          SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |                        \
 557          SECONDARY_EXEC_WBINVD_EXITING |                                \
 558          SECONDARY_EXEC_ENABLE_VPID |                                   \
 559          SECONDARY_EXEC_ENABLE_EPT |                                    \
 560          SECONDARY_EXEC_UNRESTRICTED_GUEST |                            \
 561          SECONDARY_EXEC_PAUSE_LOOP_EXITING |                            \
 562          SECONDARY_EXEC_DESC |                                          \
 563          SECONDARY_EXEC_ENABLE_RDTSCP |                                 \
 564          SECONDARY_EXEC_ENABLE_INVPCID |                                \
 565          SECONDARY_EXEC_APIC_REGISTER_VIRT |                            \
 566          SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |                         \
 567          SECONDARY_EXEC_SHADOW_VMCS |                                   \
 568          SECONDARY_EXEC_ENABLE_XSAVES |                                 \
 569          SECONDARY_EXEC_RDSEED_EXITING |                                \
 570          SECONDARY_EXEC_RDRAND_EXITING |                                \
 571          SECONDARY_EXEC_ENABLE_PML |                                    \
 572          SECONDARY_EXEC_TSC_SCALING |                                   \
 573          SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE |                         \
 574          SECONDARY_EXEC_PT_USE_GPA |                                    \
 575          SECONDARY_EXEC_PT_CONCEAL_VMX |                                \
 576          SECONDARY_EXEC_ENABLE_VMFUNC |                                 \
 577          SECONDARY_EXEC_BUS_LOCK_DETECTION |                            \
 578          SECONDARY_EXEC_NOTIFY_VM_EXITING |                             \
 579          SECONDARY_EXEC_ENCLS_EXITING |                                 \
 580          SECONDARY_EXEC_EPT_VIOLATION_VE)
 581
 582 #define KVM_REQUIRED_VMX_TERTIARY_VM_EXEC_CONTROL 0
 583 #define KVM_OPTIONAL_VMX_TERTIARY_VM_EXEC_CONTROL                       \
 584         (TERTIARY_EXEC_IPI_VIRT)
 585
 586 #define BUILD_CONTROLS_SHADOW(lname, uname, bits)                                               \
 587 static inline void lname##_controls_set(struct vcpu_vmx *vmx, u##bits val)                      \
 588 {                                                                                               \
 589         if (vmx->loaded_vmcs->controls_shadow.lname != val) {                                   \
 590                 vmcs_write##bits(uname, val);                                                   \
 591                 vmx->loaded_vmcs->controls_shadow.lname = val;                                  \
 592         }                                                                                       \
 593 }                                                                                               \
 594 static inline u##bits __##lname##_controls_get(struct loaded_vmcs *vmcs)                        \
 595 {                                                                                               \
 596         return vmcs->controls_shadow.lname;                                                     \
 597 }                                                                                               \
 598 static inline u##bits lname##_controls_get(struct vcpu_vmx *vmx)                                \
 599 {                                                                                               \
 600         return __##lname##_controls_get(vmx->loaded_vmcs);                                      \
 601 }                                                                                               \
 602 static __always_inline void lname##_controls_setbit(struct vcpu_vmx *vmx, u##bits val)          \
 603 {                                                                                               \
 604         BUILD_BUG_ON(!(val & (KVM_REQUIRED_VMX_##uname | KVM_OPTIONAL_VMX_##uname)));           \
 605         lname##_controls_set(vmx, lname##_controls_get(vmx) | val);                             \
 606 }                                                                                               \
 607 static __always_inline void lname##_controls_clearbit(struct vcpu_vmx *vmx, u##bits val)        \
 608 {                                                                                               \
 609         BUILD_BUG_ON(!(val & (KVM_REQUIRED_VMX_##uname | KVM_OPTIONAL_VMX_##uname)));           \
 610         lname##_controls_set(vmx, lname##_controls_get(vmx) & ~val);                            \
 611 }
 612 BUILD_CONTROLS_SHADOW(vm_entry, VM_ENTRY_CONTROLS, 32)
 613 BUILD_CONTROLS_SHADOW(vm_exit, VM_EXIT_CONTROLS, 32)
 614 BUILD_CONTROLS_SHADOW(pin, PIN_BASED_VM_EXEC_CONTROL, 32)
 615 BUILD_CONTROLS_SHADOW(exec, CPU_BASED_VM_EXEC_CONTROL, 32)
 616 BUILD_CONTROLS_SHADOW(secondary_exec, SECONDARY_VM_EXEC_CONTROL, 32)
 617 BUILD_CONTROLS_SHADOW(tertiary_exec, TERTIARY_VM_EXEC_CONTROL, 64)
 618
 619 /*
 620  * VMX_REGS_LAZY_LOAD_SET - The set of registers that will be updated in the
 621  * cache on demand.  Other registers not listed here are synced to
 622  * the cache immediately after VM-Exit.
 623  */
 624 #define VMX_REGS_LAZY_LOAD_SET  ((1 << VCPU_REGS_RIP) |         \
 625                                 (1 << VCPU_REGS_RSP) |          \
 626                                 (1 << VCPU_EXREG_RFLAGS) |      \
 627                                 (1 << VCPU_EXREG_PDPTR) |       \
 628                                 (1 << VCPU_EXREG_SEGMENTS) |    \
 629                                 (1 << VCPU_EXREG_CR0) |         \
 630                                 (1 << VCPU_EXREG_CR3) |         \
 631                                 (1 << VCPU_EXREG_CR4) |         \
 632                                 (1 << VCPU_EXREG_EXIT_INFO_1) | \
 633                                 (1 << VCPU_EXREG_EXIT_INFO_2))
 634
 635 static inline unsigned long vmx_l1_guest_owned_cr0_bits(void)
 636 {
 637         unsigned long bits = KVM_POSSIBLE_CR0_GUEST_BITS;
 638
 639         /*
 640          * CR0.WP needs to be intercepted when KVM is shadowing legacy paging
 641          * in order to construct shadow PTEs with the correct protections.
 642          * Note!  CR0.WP technically can be passed through to the guest if
 643          * paging is disabled, but checking CR0.PG would generate a cyclical
 644          * dependency of sorts due to forcing the caller to ensure CR0 holds
 645          * the correct value prior to determining which CR0 bits can be owned
 646          * by L1.  Keep it simple and limit the optimization to EPT.
 647          */
 648         if (!enable_ept)
 649                 bits &= ~X86_CR0_WP;
 650         return bits;
 651 }
 652
 653 static __always_inline struct kvm_vmx *to_kvm_vmx(struct kvm *kvm)
 654 {
 655         return container_of(kvm, struct kvm_vmx, kvm);
 656 }
 657
 658 static __always_inline struct vcpu_vmx *to_vmx(struct kvm_vcpu *vcpu)
 659 {
 660         return container_of(vcpu, struct vcpu_vmx, vcpu);
 661 }
 662
 663 static inline struct lbr_desc *vcpu_to_lbr_desc(struct kvm_vcpu *vcpu)
 664 {
 665         return &to_vmx(vcpu)->lbr_desc;
 666 }
 667
 668 static inline struct x86_pmu_lbr *vcpu_to_lbr_records(struct kvm_vcpu *vcpu)
 669 {
 670         return &vcpu_to_lbr_desc(vcpu)->records;
 671 }
 672
 673 static inline bool intel_pmu_lbr_is_enabled(struct kvm_vcpu *vcpu)
 674 {
 675         return !!vcpu_to_lbr_records(vcpu)->nr;
 676 }
 677
 678 void intel_pmu_cross_mapped_check(struct kvm_pmu *pmu);
 679 int intel_pmu_create_guest_lbr_event(struct kvm_vcpu *vcpu);
 680 void vmx_passthrough_lbr_msrs(struct kvm_vcpu *vcpu);
 681
 682 static __always_inline unsigned long vmx_get_exit_qual(struct kvm_vcpu *vcpu)
 683 {
 684         struct vcpu_vmx *vmx = to_vmx(vcpu);
 685
 686         if (!kvm_register_test_and_mark_available(vcpu, VCPU_EXREG_EXIT_INFO_1))
 687                 vmx->exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
 688
 689         return vmx->exit_qualification;
 690 }
 691
 692 static __always_inline u32 vmx_get_intr_info(struct kvm_vcpu *vcpu)
 693 {
 694         struct vcpu_vmx *vmx = to_vmx(vcpu);
 695
 696         if (!kvm_register_test_and_mark_available(vcpu, VCPU_EXREG_EXIT_INFO_2))
 697                 vmx->exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
 698
 699         return vmx->exit_intr_info;
 700 }
 701
 702 struct vmcs *alloc_vmcs_cpu(bool shadow, int cpu, gfp_t flags);
 703 void free_vmcs(struct vmcs *vmcs);
 704 int alloc_loaded_vmcs(struct loaded_vmcs *loaded_vmcs);
 705 void free_loaded_vmcs(struct loaded_vmcs *loaded_vmcs);
 706 void loaded_vmcs_clear(struct loaded_vmcs *loaded_vmcs);
 707
 708 static inline struct vmcs *alloc_vmcs(bool shadow)
 709 {
 710         return alloc_vmcs_cpu(shadow, raw_smp_processor_id(),
 711                               GFP_KERNEL_ACCOUNT);
 712 }
 713
 714 static inline bool vmx_has_waitpkg(struct vcpu_vmx *vmx)
 715 {
 716         return secondary_exec_controls_get(vmx) &
 717                 SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE;
 718 }
 719
 720 static inline bool vmx_need_pf_intercept(struct kvm_vcpu *vcpu)
 721 {
 722         if (!enable_ept)
 723                 return true;
 724
 725         return allow_smaller_maxphyaddr &&
 726                cpuid_maxphyaddr(vcpu) < kvm_host.maxphyaddr;
 727 }
 728
 729 static inline bool is_unrestricted_guest(struct kvm_vcpu *vcpu)
 730 {
 731         return enable_unrestricted_guest && (!is_guest_mode(vcpu) ||
 732             (secondary_exec_controls_get(to_vmx(vcpu)) &
 733             SECONDARY_EXEC_UNRESTRICTED_GUEST));
 734 }
 735
 736 bool __vmx_guest_state_valid(struct kvm_vcpu *vcpu);
 737 static inline bool vmx_guest_state_valid(struct kvm_vcpu *vcpu)
 738 {
 739         return is_unrestricted_guest(vcpu) || __vmx_guest_state_valid(vcpu);
 740 }
 741
 742 void dump_vmcs(struct kvm_vcpu *vcpu);
 743
 744 static inline int vmx_get_instr_info_reg2(u32 vmx_instr_info)
 745 {
 746         return (vmx_instr_info >> 28) & 0xf;
 747 }
 748
 749 static inline bool vmx_can_use_ipiv(struct kvm_vcpu *vcpu)
 750 {
 751         return  lapic_in_kernel(vcpu) && enable_ipiv;
 752 }
 753
 754 static inline void vmx_segment_cache_clear(struct vcpu_vmx *vmx)
 755 {
 756         vmx->segment_cache.bitmask = 0;
 757 }
 758
 759 #endif /* __KVM_X86_VMX_H */