1 /* SPDX-License-Identifier: GPL-2.0 */ 2 #ifndef ARCH_X86_KVM_X86_H 3 #define ARCH_X86_KVM_X86_H 4 5 #include <linux/kvm_host.h> 6 #include <asm/fpu/xstate.h> 7 #include <asm/mce.h> 8 #include <asm/pvclock.h> 9 #include "msrs.h" 10 #include "mmu.h" 11 #include "regs.h" 12 #include "kvm_emulate.h" 13 #include "cpuid.h" 14 15 #define KVM_MAX_MCE_BANKS 32 16 17 int kvm_x86_vendor_init(struct kvm_x86_init_ops *ops); 18 void kvm_x86_vendor_exit(void); 19 20 void kvm_spurious_fault(void); 21 22 #define SIZE_OF_MEMSLOTS_HASHTABLE \ 23 (sizeof(((struct kvm_memslots *)0)->id_hash) * 2 * KVM_MAX_NR_ADDRESS_SPACES) 24 25 /* Sanity check the size of the memslot hash tables. */ 26 static_assert(SIZE_OF_MEMSLOTS_HASHTABLE == 27 (1024 * (1 + IS_ENABLED(CONFIG_X86_64)) * (1 + IS_ENABLED(CONFIG_KVM_SMM)))); 28 29 /* 30 * Assert that "struct kvm_{svm,vmx,tdx}" is an order-0 or order-1 allocation. 31 * Spilling over to an order-2 allocation isn't fundamentally problematic, but 32 * isn't expected to happen in the foreseeable future (O(years)). Assert that 33 * the size is an order-0 allocation when ignoring the memslot hash tables, to 34 * help detect and debug unexpected size increases. 35 */ 36 #define KVM_SANITY_CHECK_VM_STRUCT_SIZE(x) \ 37 do { \ 38 BUILD_BUG_ON(get_order(sizeof(struct x) - SIZE_OF_MEMSLOTS_HASHTABLE) && \ 39 !IS_ENABLED(CONFIG_DEBUG_KERNEL) && !IS_ENABLED(CONFIG_KASAN)); \ 40 BUILD_BUG_ON(get_order(sizeof(struct x)) > 1 && \ 41 !IS_ENABLED(CONFIG_DEBUG_KERNEL) && !IS_ENABLED(CONFIG_KASAN)); \ 42 } while (0) 43 44 #define KVM_NESTED_VMENTER_CONSISTENCY_CHECK(consistency_check) \ 45 ({ \ 46 bool failed = (consistency_check); \ 47 if (failed) \ 48 trace_kvm_nested_vmenter_failed(#consistency_check, 0); \ 49 failed; \ 50 }) 51 52 #define KVM_DEFAULT_PLE_GAP 128 53 #define KVM_VMX_DEFAULT_PLE_WINDOW 4096 54 #define KVM_DEFAULT_PLE_WINDOW_GROW 2 55 #define KVM_DEFAULT_PLE_WINDOW_SHRINK 0 56 #define KVM_VMX_DEFAULT_PLE_WINDOW_MAX UINT_MAX 57 #define KVM_SVM_DEFAULT_PLE_WINDOW_MAX USHRT_MAX 58 #define KVM_SVM_DEFAULT_PLE_WINDOW 3000 59 60 static inline unsigned int __grow_ple_window(unsigned int val, 61 unsigned int base, unsigned int modifier, unsigned int max) 62 { 63 u64 ret = val; 64 65 if (modifier < 1) 66 return base; 67 68 if (modifier < base) 69 ret *= modifier; 70 else 71 ret += modifier; 72 73 return min(ret, (u64)max); 74 } 75 76 static inline unsigned int __shrink_ple_window(unsigned int val, 77 unsigned int base, unsigned int modifier, unsigned int min) 78 { 79 if (modifier < 1) 80 return base; 81 82 if (modifier < base) 83 val /= modifier; 84 else 85 val -= modifier; 86 87 return max(val, min); 88 } 89 90 void kvm_service_local_tlb_flush_requests(struct kvm_vcpu *vcpu); 91 int kvm_check_nested_events(struct kvm_vcpu *vcpu); 92 93 /* Forcibly leave the nested mode in cases like a vCPU reset */ 94 static inline void kvm_leave_nested(struct kvm_vcpu *vcpu) 95 { 96 kvm_nested_call(leave_nested)(vcpu); 97 } 98 99 /* 100 * If IBRS is advertised to the vCPU, KVM must flush the indirect branch 101 * predictors when transitioning from L2 to L1, as L1 expects hardware (KVM in 102 * this case) to provide separate predictor modes. Bare metal isolates the host 103 * from the guest, but doesn't isolate different guests from one another (in 104 * this case L1 and L2). The exception is if bare metal supports same mode IBRS, 105 * which offers protection within the same mode, and hence protects L1 from L2. 106 */ 107 static inline void kvm_nested_vmexit_handle_ibrs(struct kvm_vcpu *vcpu) 108 { 109 if (cpu_feature_enabled(X86_FEATURE_AMD_IBRS_SAME_MODE)) 110 return; 111 112 if (guest_cpu_cap_has(vcpu, X86_FEATURE_SPEC_CTRL) || 113 guest_cpu_cap_has(vcpu, X86_FEATURE_AMD_IBRS)) 114 indirect_branch_prediction_barrier(); 115 } 116 117 /* 118 * Disallow modifying CPUID and feature MSRs, which affect the core virtual CPU 119 * model exposed to the guest and virtualized by KVM, if the vCPU has already 120 * run or is in guest mode (L2). In both cases, KVM has already consumed the 121 * current virtual CPU model, and doesn't support "unwinding" to react to the 122 * new model. 123 * 124 * Note, the only way is_guest_mode() can be true with 'last_vmentry_cpu == -1' 125 * is if userspace sets CPUID and feature MSRs (to enable VMX/SVM), then sets 126 * nested state, and then attempts to set CPUID and/or feature MSRs *again*. 127 */ 128 static inline bool kvm_can_set_cpuid_and_feature_msrs(struct kvm_vcpu *vcpu) 129 { 130 return vcpu->arch.last_vmentry_cpu == -1 && !is_guest_mode(vcpu); 131 } 132 133 /* 134 * WARN if a nested VM-Enter is pending completion, and userspace hasn't gained 135 * control since the nested VM-Enter was initiated (in which case, userspace 136 * may have modified vCPU state to induce an architecturally invalid VM-Exit). 137 */ 138 static inline void kvm_warn_on_nested_run_pending(struct kvm_vcpu *vcpu) 139 { 140 WARN_ON_ONCE(vcpu->arch.nested_run_pending == KVM_NESTED_RUN_PENDING); 141 } 142 143 static inline void kvm_set_mp_state(struct kvm_vcpu *vcpu, int mp_state) 144 { 145 vcpu->arch.mp_state = mp_state; 146 if (mp_state == KVM_MP_STATE_RUNNABLE) 147 vcpu->arch.pv.pv_unhalted = false; 148 } 149 150 static inline bool kvm_is_exception_pending(struct kvm_vcpu *vcpu) 151 { 152 return vcpu->arch.exception.pending || 153 vcpu->arch.exception_vmexit.pending || 154 kvm_test_request(KVM_REQ_TRIPLE_FAULT, vcpu); 155 } 156 157 static inline void kvm_clear_exception_queue(struct kvm_vcpu *vcpu) 158 { 159 vcpu->arch.exception.pending = false; 160 vcpu->arch.exception.injected = false; 161 vcpu->arch.exception_vmexit.pending = false; 162 } 163 164 static inline void kvm_queue_interrupt(struct kvm_vcpu *vcpu, u8 vector, 165 bool soft) 166 { 167 vcpu->arch.interrupt.injected = true; 168 vcpu->arch.interrupt.soft = soft; 169 vcpu->arch.interrupt.nr = vector; 170 } 171 172 static inline void kvm_clear_interrupt_queue(struct kvm_vcpu *vcpu) 173 { 174 vcpu->arch.interrupt.injected = false; 175 } 176 177 static inline bool kvm_event_needs_reinjection(struct kvm_vcpu *vcpu) 178 { 179 return vcpu->arch.exception.injected || vcpu->arch.interrupt.injected || 180 vcpu->arch.nmi_injected; 181 } 182 183 static inline bool kvm_exception_is_soft(unsigned int nr) 184 { 185 return (nr == BP_VECTOR) || (nr == OF_VECTOR); 186 } 187 188 static inline bool x86_exception_has_error_code(unsigned int vector) 189 { 190 static u32 exception_has_error_code = BIT(DF_VECTOR) | BIT(TS_VECTOR) | 191 BIT(NP_VECTOR) | BIT(SS_VECTOR) | BIT(GP_VECTOR) | 192 BIT(PF_VECTOR) | BIT(AC_VECTOR); 193 194 return (1U << vector) & exception_has_error_code; 195 } 196 197 static inline u8 vcpu_virt_addr_bits(struct kvm_vcpu *vcpu) 198 { 199 return kvm_is_cr4_bit_set(vcpu, X86_CR4_LA57) ? 57 : 48; 200 } 201 202 static inline u8 max_host_virt_addr_bits(void) 203 { 204 return kvm_cpu_cap_has(X86_FEATURE_LA57) ? 57 : 48; 205 } 206 207 /* 208 * x86 MSRs which contain linear addresses, x86 hidden segment bases, and 209 * IDT/GDT bases have static canonicality checks, the size of which depends 210 * only on the CPU's support for 5-level paging, rather than on the state of 211 * CR4.LA57. This applies to both WRMSR and to other instructions that set 212 * their values, e.g. SGDT. 213 * 214 * KVM passes through most of these MSRS and also doesn't intercept the 215 * instructions that set the hidden segment bases. 216 * 217 * Because of this, to be consistent with hardware, even if the guest doesn't 218 * have LA57 enabled in its CPUID, perform canonicality checks based on *host* 219 * support for 5 level paging. 220 * 221 * Finally, instructions which are related to MMU invalidation of a given 222 * linear address, also have a similar static canonical check on address. 223 * This allows for example to invalidate 5-level addresses of a guest from a 224 * host which uses 4-level paging. 225 */ 226 static inline bool is_noncanonical_address(u64 la, struct kvm_vcpu *vcpu, 227 unsigned int flags) 228 { 229 if (flags & (X86EMUL_F_INVLPG | X86EMUL_F_MSR | X86EMUL_F_DT_LOAD)) 230 return !__is_canonical_address(la, max_host_virt_addr_bits()); 231 else 232 return !__is_canonical_address(la, vcpu_virt_addr_bits(vcpu)); 233 } 234 235 static inline bool is_noncanonical_msr_address(u64 la, struct kvm_vcpu *vcpu) 236 { 237 return is_noncanonical_address(la, vcpu, X86EMUL_F_MSR); 238 } 239 240 static inline bool is_noncanonical_base_address(u64 la, struct kvm_vcpu *vcpu) 241 { 242 return is_noncanonical_address(la, vcpu, X86EMUL_F_DT_LOAD); 243 } 244 245 static inline bool is_noncanonical_invlpg_address(u64 la, struct kvm_vcpu *vcpu) 246 { 247 return is_noncanonical_address(la, vcpu, X86EMUL_F_INVLPG); 248 } 249 250 static inline void vcpu_cache_mmio_info(struct kvm_vcpu *vcpu, 251 gva_t gva, gfn_t gfn, unsigned access) 252 { 253 u64 gen = kvm_memslots(vcpu->kvm)->generation; 254 255 if (unlikely(gen & KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS)) 256 return; 257 258 /* 259 * If this is a shadow nested page table, the "GVA" is 260 * actually a nGPA. 261 */ 262 vcpu->arch.mmio_gva = mmu_is_nested(vcpu) ? 0 : gva & PAGE_MASK; 263 vcpu->arch.mmio_access = access; 264 vcpu->arch.mmio_gfn = gfn; 265 vcpu->arch.mmio_gen = gen; 266 } 267 268 static inline bool vcpu_match_mmio_gen(struct kvm_vcpu *vcpu) 269 { 270 return vcpu->arch.mmio_gen == kvm_memslots(vcpu->kvm)->generation; 271 } 272 273 /* 274 * Clear the mmio cache info for the given gva. If gva is MMIO_GVA_ANY, we 275 * clear all mmio cache info. 276 */ 277 #define MMIO_GVA_ANY (~(gva_t)0) 278 279 static inline void vcpu_clear_mmio_info(struct kvm_vcpu *vcpu, gva_t gva) 280 { 281 if (gva != MMIO_GVA_ANY && vcpu->arch.mmio_gva != (gva & PAGE_MASK)) 282 return; 283 284 vcpu->arch.mmio_gva = 0; 285 } 286 287 static inline bool vcpu_match_mmio_gva(struct kvm_vcpu *vcpu, unsigned long gva) 288 { 289 if (vcpu_match_mmio_gen(vcpu) && vcpu->arch.mmio_gva && 290 vcpu->arch.mmio_gva == (gva & PAGE_MASK)) 291 return true; 292 293 return false; 294 } 295 296 static inline bool vcpu_match_mmio_gpa(struct kvm_vcpu *vcpu, gpa_t gpa) 297 { 298 if (vcpu_match_mmio_gen(vcpu) && vcpu->arch.mmio_gfn && 299 vcpu->arch.mmio_gfn == gpa >> PAGE_SHIFT) 300 return true; 301 302 return false; 303 } 304 305 static inline bool kvm_check_has_quirk(struct kvm *kvm, u64 quirk) 306 { 307 return !(READ_ONCE(kvm->arch.disabled_quirks) & quirk); 308 } 309 310 static __always_inline void kvm_request_l1tf_flush_l1d(void) 311 { 312 #if IS_ENABLED(CONFIG_CPU_MITIGATIONS) && IS_ENABLED(CONFIG_KVM_INTEL) 313 /* 314 * Use a raw write to set the per-CPU flag, as KVM will ensure a flush 315 * even if preemption is currently enabled.. If the current vCPU task 316 * is migrated to a different CPU (or userspace runs the vCPU on a 317 * different task) before the next VM-Entry, then kvm_arch_vcpu_load() 318 * will request a flush on the new CPU. 319 */ 320 raw_cpu_write(irq_stat.kvm_cpu_l1tf_flush_l1d, 1); 321 #endif 322 } 323 324 void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event); 325 326 void kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip); 327 328 u64 get_kvmclock_ns(struct kvm *kvm); 329 uint64_t kvm_get_wall_clock_epoch(struct kvm *kvm); 330 bool kvm_get_monotonic_and_clockread(s64 *kernel_ns, u64 *tsc_timestamp); 331 int kvm_guest_time_update(struct kvm_vcpu *v); 332 333 void kvm_synchronize_tsc(struct kvm_vcpu *vcpu, u64 *user_value); 334 u64 kvm_scale_tsc(u64 tsc, u64 ratio); 335 u64 kvm_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc); 336 u64 kvm_calc_nested_tsc_offset(u64 l1_offset, u64 l2_offset, u64 l2_multiplier); 337 u64 kvm_calc_nested_tsc_multiplier(u64 l1_multiplier, u64 l2_multiplier); 338 u64 kvm_compute_l1_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc); 339 void kvm_vcpu_write_tsc_offset(struct kvm_vcpu *vcpu, u64 l1_offset); 340 341 static inline void adjust_tsc_offset_guest(struct kvm_vcpu *vcpu, 342 s64 adjustment) 343 { 344 kvm_vcpu_write_tsc_offset(vcpu, vcpu->arch.l1_tsc_offset + adjustment); 345 } 346 347 static inline void adjust_tsc_offset_host(struct kvm_vcpu *vcpu, s64 adjustment) 348 { 349 if (vcpu->arch.l1_tsc_scaling_ratio != kvm_caps.default_tsc_scaling_ratio) 350 WARN_ON(adjustment < 0); 351 adjustment = kvm_scale_tsc((u64) adjustment, 352 vcpu->arch.l1_tsc_scaling_ratio); 353 adjust_tsc_offset_guest(vcpu, adjustment); 354 } 355 356 int kvm_read_guest_virt(struct kvm_vcpu *vcpu, 357 gva_t addr, void *val, unsigned int bytes, 358 struct x86_exception *exception); 359 360 int kvm_write_guest_virt_system(struct kvm_vcpu *vcpu, 361 gva_t addr, void *val, unsigned int bytes, 362 struct x86_exception *exception); 363 364 int handle_ud(struct kvm_vcpu *vcpu); 365 366 void kvm_deliver_exception_payload(struct kvm_vcpu *vcpu, 367 struct kvm_queued_exception *ex); 368 void kvm_handle_exception_payload_quirk(struct kvm_vcpu *vcpu); 369 370 void kvm_fixup_and_inject_pf_error(struct kvm_vcpu *vcpu, gva_t gva, u16 error_code); 371 int x86_decode_emulated_instruction(struct kvm_vcpu *vcpu, int emulation_type, 372 void *insn, int insn_len); 373 int x86_emulate_instruction(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa, 374 int emulation_type, void *insn, int insn_len); 375 /* 376 * EMULTYPE_NO_DECODE - Set when re-emulating an instruction (after completing 377 * userspace I/O) to indicate that the emulation context 378 * should be reused as is, i.e. skip initialization of 379 * emulation context, instruction fetch and decode. 380 * 381 * EMULTYPE_TRAP_UD - Set when emulating an intercepted #UD from hardware. 382 * Indicates that only select instructions (tagged with 383 * EmulateOnUD) should be emulated (to minimize the emulator 384 * attack surface). See also EMULTYPE_TRAP_UD_FORCED. 385 * 386 * EMULTYPE_SKIP - Set when emulating solely to skip an instruction, i.e. to 387 * decode the instruction length. For use *only* by 388 * kvm_x86_ops.skip_emulated_instruction() implementations if 389 * EMULTYPE_COMPLETE_USER_EXIT is not set. 390 * 391 * EMULTYPE_ALLOW_RETRY_PF - Set when the emulator should resume the guest to 392 * retry native execution under certain conditions, 393 * Can only be set in conjunction with EMULTYPE_PF. 394 * 395 * EMULTYPE_TRAP_UD_FORCED - Set when emulating an intercepted #UD that was 396 * triggered by KVM's magic "force emulation" prefix, 397 * which is opt in via module param (off by default). 398 * Bypasses EmulateOnUD restriction despite emulating 399 * due to an intercepted #UD (see EMULTYPE_TRAP_UD). 400 * Used to test the full emulator from userspace. 401 * 402 * EMULTYPE_VMWARE_GP - Set when emulating an intercepted #GP for VMware 403 * backdoor emulation, which is opt in via module param. 404 * VMware backdoor emulation handles select instructions 405 * and reinjects the #GP for all other cases. 406 * 407 * EMULTYPE_PF - Set when an intercepted #PF triggers the emulation, in which case 408 * the CR2/GPA value pass on the stack is valid. 409 * 410 * EMULTYPE_COMPLETE_USER_EXIT - Set when the emulator should update interruptibility 411 * state and inject single-step #DBs after skipping 412 * an instruction (after completing userspace I/O). 413 * 414 * EMULTYPE_WRITE_PF_TO_SP - Set when emulating an intercepted page fault that 415 * is attempting to write a gfn that contains one or 416 * more of the PTEs used to translate the write itself, 417 * and the owning page table is being shadowed by KVM. 418 * If emulation of the faulting instruction fails and 419 * this flag is set, KVM will exit to userspace instead 420 * of retrying emulation as KVM cannot make forward 421 * progress. 422 * 423 * If emulation fails for a write to guest page tables, 424 * KVM unprotects (zaps) the shadow page for the target 425 * gfn and resumes the guest to retry the non-emulatable 426 * instruction (on hardware). Unprotecting the gfn 427 * doesn't allow forward progress for a self-changing 428 * access because doing so also zaps the translation for 429 * the gfn, i.e. retrying the instruction will hit a 430 * !PRESENT fault, which results in a new shadow page 431 * and sends KVM back to square one. 432 * 433 * EMULTYPE_SKIP_SOFT_INT - Set in combination with EMULTYPE_SKIP to only skip 434 * an instruction if it could generate a given software 435 * interrupt, which must be encoded via 436 * EMULTYPE_SET_SOFT_INT_VECTOR(). 437 */ 438 #define EMULTYPE_NO_DECODE (1 << 0) 439 #define EMULTYPE_TRAP_UD (1 << 1) 440 #define EMULTYPE_SKIP (1 << 2) 441 #define EMULTYPE_ALLOW_RETRY_PF (1 << 3) 442 #define EMULTYPE_TRAP_UD_FORCED (1 << 4) 443 #define EMULTYPE_VMWARE_GP (1 << 5) 444 #define EMULTYPE_PF (1 << 6) 445 #define EMULTYPE_COMPLETE_USER_EXIT (1 << 7) 446 #define EMULTYPE_WRITE_PF_TO_SP (1 << 8) 447 #define EMULTYPE_SKIP_SOFT_INT (1 << 9) 448 449 #define EMULTYPE_SET_SOFT_INT_VECTOR(v) ((u32)((v) & 0xff) << 16) 450 #define EMULTYPE_GET_SOFT_INT_VECTOR(e) (((e) >> 16) & 0xff) 451 452 static inline bool kvm_can_emulate_event_vectoring(int emul_type) 453 { 454 return !(emul_type & EMULTYPE_PF); 455 } 456 457 int kvm_emulate_instruction(struct kvm_vcpu *vcpu, int emulation_type); 458 int kvm_emulate_instruction_from_buffer(struct kvm_vcpu *vcpu, 459 void *insn, int insn_len); 460 void __kvm_prepare_emulation_failure_exit(struct kvm_vcpu *vcpu, 461 u64 *data, u8 ndata); 462 void kvm_prepare_emulation_failure_exit(struct kvm_vcpu *vcpu); 463 464 void kvm_prepare_event_vectoring_exit(struct kvm_vcpu *vcpu, gpa_t gpa); 465 void kvm_prepare_unexpected_reason_exit(struct kvm_vcpu *vcpu, u64 exit_reason); 466 467 fastpath_t handle_fastpath_hlt(struct kvm_vcpu *vcpu); 468 fastpath_t handle_fastpath_invd(struct kvm_vcpu *vcpu); 469 470 int kvm_emulate_as_nop(struct kvm_vcpu *vcpu); 471 int kvm_emulate_invd(struct kvm_vcpu *vcpu); 472 int kvm_emulate_mwait(struct kvm_vcpu *vcpu); 473 int kvm_handle_invalid_op(struct kvm_vcpu *vcpu); 474 int kvm_emulate_monitor(struct kvm_vcpu *vcpu); 475 476 int kvm_fast_pio(struct kvm_vcpu *vcpu, int size, unsigned short port, int in); 477 int kvm_emulate_cpuid(struct kvm_vcpu *vcpu); 478 int kvm_emulate_halt(struct kvm_vcpu *vcpu); 479 int kvm_emulate_halt_noskip(struct kvm_vcpu *vcpu); 480 int kvm_emulate_ap_reset_hold(struct kvm_vcpu *vcpu); 481 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu); 482 483 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector); 484 485 enum kvm_task_switch_reason { 486 TASK_SWITCH_CALL = 0, 487 TASK_SWITCH_IRET = 1, 488 TASK_SWITCH_JMP = 2, 489 TASK_SWITCH_GATE = 3, 490 }; 491 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index, 492 int reason, bool has_error_code, u32 error_code); 493 494 int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr); 495 int kvm_emulate_xsetbv(struct kvm_vcpu *vcpu); 496 int kvm_emulate_rdpmc(struct kvm_vcpu *vcpu); 497 498 int kvm_skip_emulated_instruction(struct kvm_vcpu *vcpu); 499 int kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err); 500 501 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr); 502 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code); 503 void kvm_queue_exception_p(struct kvm_vcpu *vcpu, unsigned nr, unsigned long payload); 504 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned int nr, 505 bool has_error_code, u32 error_code); 506 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault, 507 bool from_hardware); 508 void __kvm_inject_emulated_page_fault(struct kvm_vcpu *vcpu, 509 struct x86_exception *fault, 510 bool from_hardware); 511 512 static inline void kvm_inject_emulated_page_fault(struct kvm_vcpu *vcpu, 513 struct x86_exception *fault) 514 { 515 __kvm_inject_emulated_page_fault(vcpu, fault, false); 516 } 517 518 bool kvm_require_dr(struct kvm_vcpu *vcpu, int dr); 519 520 static inline void kvm_inject_gp(struct kvm_vcpu *vcpu, u32 error_code) 521 { 522 kvm_queue_exception_e(vcpu, GP_VECTOR, error_code); 523 } 524 525 void kvm_inject_nmi(struct kvm_vcpu *vcpu); 526 int kvm_get_nr_pending_nmis(struct kvm_vcpu *vcpu); 527 528 void __user *__x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, 529 u32 size); 530 int memslot_rmap_alloc(struct kvm_memory_slot *slot, unsigned long npages); 531 532 bool kvm_vcpu_is_reset_bsp(struct kvm_vcpu *vcpu); 533 bool kvm_vcpu_is_bsp(struct kvm_vcpu *vcpu); 534 535 enum kvm_apicv_inhibit { 536 537 /********************************************************************/ 538 /* INHIBITs that are relevant to both Intel's APICv and AMD's AVIC. */ 539 /********************************************************************/ 540 541 /* 542 * APIC acceleration is disabled by a module parameter 543 * and/or not supported in hardware. 544 */ 545 APICV_INHIBIT_REASON_DISABLED, 546 547 /* 548 * APIC acceleration is inhibited because AutoEOI feature is 549 * being used by a HyperV guest. 550 */ 551 APICV_INHIBIT_REASON_HYPERV, 552 553 /* 554 * APIC acceleration is inhibited because the userspace didn't yet 555 * enable the kernel/split irqchip. 556 */ 557 APICV_INHIBIT_REASON_ABSENT, 558 559 /* APIC acceleration is inhibited because KVM_GUESTDBG_BLOCKIRQ 560 * (out of band, debug measure of blocking all interrupts on this vCPU) 561 * was enabled, to avoid AVIC/APICv bypassing it. 562 */ 563 APICV_INHIBIT_REASON_BLOCKIRQ, 564 565 /* 566 * APICv is disabled because not all vCPUs have a 1:1 mapping between 567 * APIC ID and vCPU, _and_ KVM is not applying its x2APIC hotplug hack. 568 */ 569 APICV_INHIBIT_REASON_PHYSICAL_ID_ALIASED, 570 571 /* 572 * For simplicity, the APIC acceleration is inhibited 573 * first time either APIC ID or APIC base are changed by the guest 574 * from their reset values. 575 */ 576 APICV_INHIBIT_REASON_APIC_ID_MODIFIED, 577 APICV_INHIBIT_REASON_APIC_BASE_MODIFIED, 578 579 /******************************************************/ 580 /* INHIBITs that are relevant only to the AMD's AVIC. */ 581 /******************************************************/ 582 583 /* 584 * AVIC is inhibited on a vCPU because it runs a nested guest. 585 * 586 * This is needed because unlike APICv, the peers of this vCPU 587 * cannot use the doorbell mechanism to signal interrupts via AVIC when 588 * a vCPU runs nested. 589 */ 590 APICV_INHIBIT_REASON_NESTED, 591 592 /* 593 * On SVM, the wait for the IRQ window is implemented with pending vIRQ, 594 * which cannot be injected when the AVIC is enabled, thus AVIC 595 * is inhibited while KVM waits for IRQ window. 596 */ 597 APICV_INHIBIT_REASON_IRQWIN, 598 599 /* 600 * PIT (i8254) 're-inject' mode, relies on EOI intercept, 601 * which AVIC doesn't support for edge triggered interrupts. 602 */ 603 APICV_INHIBIT_REASON_PIT_REINJ, 604 605 /* 606 * AVIC is disabled because SEV doesn't support it. 607 */ 608 APICV_INHIBIT_REASON_SEV, 609 610 /* 611 * AVIC is disabled because not all vCPUs with a valid LDR have a 1:1 612 * mapping between logical ID and vCPU. 613 */ 614 APICV_INHIBIT_REASON_LOGICAL_ID_ALIASED, 615 616 /* 617 * AVIC is disabled because the vCPU's APIC ID is beyond the max 618 * supported by AVIC/x2AVIC, i.e. the vCPU is unaddressable. 619 */ 620 APICV_INHIBIT_REASON_PHYSICAL_ID_TOO_BIG, 621 622 NR_APICV_INHIBIT_REASONS, 623 }; 624 625 #define __APICV_INHIBIT_REASON(reason) \ 626 { BIT(APICV_INHIBIT_REASON_##reason), #reason } 627 628 #define APICV_INHIBIT_REASONS \ 629 __APICV_INHIBIT_REASON(DISABLED), \ 630 __APICV_INHIBIT_REASON(HYPERV), \ 631 __APICV_INHIBIT_REASON(ABSENT), \ 632 __APICV_INHIBIT_REASON(BLOCKIRQ), \ 633 __APICV_INHIBIT_REASON(PHYSICAL_ID_ALIASED), \ 634 __APICV_INHIBIT_REASON(APIC_ID_MODIFIED), \ 635 __APICV_INHIBIT_REASON(APIC_BASE_MODIFIED), \ 636 __APICV_INHIBIT_REASON(NESTED), \ 637 __APICV_INHIBIT_REASON(IRQWIN), \ 638 __APICV_INHIBIT_REASON(PIT_REINJ), \ 639 __APICV_INHIBIT_REASON(SEV), \ 640 __APICV_INHIBIT_REASON(LOGICAL_ID_ALIASED), \ 641 __APICV_INHIBIT_REASON(PHYSICAL_ID_TOO_BIG) 642 643 bool kvm_apicv_activated(struct kvm *kvm); 644 bool kvm_vcpu_apicv_activated(struct kvm_vcpu *vcpu); 645 void __kvm_vcpu_update_apicv(struct kvm_vcpu *vcpu); 646 void __kvm_set_or_clear_apicv_inhibit(struct kvm *kvm, 647 enum kvm_apicv_inhibit reason, bool set); 648 void kvm_set_or_clear_apicv_inhibit(struct kvm *kvm, 649 enum kvm_apicv_inhibit reason, bool set); 650 651 static inline void kvm_set_apicv_inhibit(struct kvm *kvm, 652 enum kvm_apicv_inhibit reason) 653 { 654 kvm_set_or_clear_apicv_inhibit(kvm, reason, true); 655 } 656 657 static inline void kvm_clear_apicv_inhibit(struct kvm *kvm, 658 enum kvm_apicv_inhibit reason) 659 { 660 kvm_set_or_clear_apicv_inhibit(kvm, reason, false); 661 } 662 663 void kvm_inc_or_dec_irq_window_inhibit(struct kvm *kvm, bool inc); 664 665 static inline void kvm_inc_apicv_irq_window_req(struct kvm *kvm) 666 { 667 kvm_inc_or_dec_irq_window_inhibit(kvm, true); 668 } 669 670 static inline void kvm_dec_apicv_irq_window_req(struct kvm *kvm) 671 { 672 kvm_inc_or_dec_irq_window_inhibit(kvm, false); 673 } 674 675 void kvm_make_scan_ioapic_request(struct kvm *kvm); 676 void kvm_make_scan_ioapic_request_mask(struct kvm *kvm, 677 unsigned long *vcpu_bitmap); 678 679 void kvm_setup_xss_caps(void); 680 681 /* 682 * Get a filtered version of KVM's supported XCR0 that strips out dynamic 683 * features for which the current process doesn't (yet) have permission to use. 684 * This is intended to be used only when enumerating support to userspace, 685 * e.g. in KVM_GET_SUPPORTED_CPUID and KVM_CAP_XSAVE2, it does NOT need to be 686 * used to check/restrict guest behavior as KVM rejects KVM_SET_CPUID{2} if 687 * userspace attempts to enable unpermitted features. 688 */ 689 static inline u64 kvm_get_filtered_xcr0(void) 690 { 691 u64 permitted_xcr0 = kvm_caps.supported_xcr0; 692 693 BUILD_BUG_ON(XFEATURE_MASK_USER_DYNAMIC != XFEATURE_MASK_XTILE_DATA); 694 695 if (permitted_xcr0 & XFEATURE_MASK_USER_DYNAMIC) { 696 permitted_xcr0 &= xstate_get_guest_group_perm(); 697 698 /* 699 * Treat XTILE_CFG as unsupported if the current process isn't 700 * allowed to use XTILE_DATA, as attempting to set XTILE_CFG in 701 * XCR0 without setting XTILE_DATA is architecturally illegal. 702 */ 703 if (!(permitted_xcr0 & XFEATURE_MASK_XTILE_DATA)) 704 permitted_xcr0 &= ~XFEATURE_MASK_XTILE_CFG; 705 } 706 return permitted_xcr0; 707 } 708 709 static inline bool kvm_mpx_supported(void) 710 { 711 return (kvm_caps.supported_xcr0 & (XFEATURE_MASK_BNDREGS | XFEATURE_MASK_BNDCSR)) 712 == (XFEATURE_MASK_BNDREGS | XFEATURE_MASK_BNDCSR); 713 } 714 715 extern unsigned int min_timer_period_us; 716 717 extern bool enable_vmware_backdoor; 718 719 extern int pi_inject_timer; 720 721 static inline u64 nsec_to_cycles(struct kvm_vcpu *vcpu, u64 nsec) 722 { 723 return pvclock_scale_delta(nsec, vcpu->arch.virtual_tsc_mult, 724 vcpu->arch.virtual_tsc_shift); 725 } 726 727 /* Same "calling convention" as do_div: 728 * - divide (n << 32) by base 729 * - put result in n 730 * - return remainder 731 */ 732 #define do_shl32_div32(n, base) \ 733 ({ \ 734 u32 __quot, __rem; \ 735 asm("divl %2" : "=a" (__quot), "=d" (__rem) \ 736 : "rm" (base), "0" (0), "1" ((u32) n)); \ 737 n = __quot; \ 738 __rem; \ 739 }) 740 741 static inline void kvm_disable_exits(struct kvm *kvm, u64 mask) 742 { 743 kvm->arch.disabled_exits |= mask; 744 } 745 746 static inline bool kvm_mwait_in_guest(struct kvm *kvm) 747 { 748 return kvm->arch.disabled_exits & KVM_X86_DISABLE_EXITS_MWAIT; 749 } 750 751 static inline bool kvm_hlt_in_guest(struct kvm *kvm) 752 { 753 return kvm->arch.disabled_exits & KVM_X86_DISABLE_EXITS_HLT; 754 } 755 756 static inline bool kvm_pause_in_guest(struct kvm *kvm) 757 { 758 return kvm->arch.disabled_exits & KVM_X86_DISABLE_EXITS_PAUSE; 759 } 760 761 static inline bool kvm_cstate_in_guest(struct kvm *kvm) 762 { 763 return kvm->arch.disabled_exits & KVM_X86_DISABLE_EXITS_CSTATE; 764 } 765 766 static inline bool kvm_aperfmperf_in_guest(struct kvm *kvm) 767 { 768 return kvm->arch.disabled_exits & KVM_X86_DISABLE_EXITS_APERFMPERF; 769 } 770 771 static inline bool kvm_notify_vmexit_enabled(struct kvm *kvm) 772 { 773 return kvm->arch.notify_vmexit_flags & KVM_X86_NOTIFY_VMEXIT_ENABLED; 774 } 775 776 static __always_inline void kvm_before_interrupt(struct kvm_vcpu *vcpu, 777 enum kvm_intr_type intr) 778 { 779 WRITE_ONCE(vcpu->arch.handling_intr_from_guest, (u8)intr); 780 } 781 782 static __always_inline void kvm_after_interrupt(struct kvm_vcpu *vcpu) 783 { 784 WRITE_ONCE(vcpu->arch.handling_intr_from_guest, 0); 785 } 786 787 static inline bool kvm_handling_nmi_from_guest(struct kvm_vcpu *vcpu) 788 { 789 return vcpu->arch.handling_intr_from_guest == KVM_HANDLING_NMI; 790 } 791 792 static inline bool kvm_pat_valid(u64 data) 793 { 794 if (data & 0xF8F8F8F8F8F8F8F8ull) 795 return false; 796 /* 0, 1, 4, 5, 6, 7 are valid values. */ 797 return (data | ((data & 0x0202020202020202ull) << 1)) == data; 798 } 799 800 static inline bool __kvm_pv_async_pf_enabled(u64 data) 801 { 802 u64 mask = KVM_ASYNC_PF_ENABLED | KVM_ASYNC_PF_DELIVERY_AS_INT; 803 804 return (data & mask) == mask; 805 } 806 807 static inline bool kvm_pv_async_pf_enabled(struct kvm_vcpu *vcpu) 808 { 809 return __kvm_pv_async_pf_enabled(vcpu->arch.apf.msr_en_val); 810 } 811 812 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu) 813 { 814 int i; 815 for (i = 0; i < ASYNC_PF_PER_VCPU; i++) 816 vcpu->arch.apf.gfns[i] = ~0; 817 } 818 819 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn); 820 821 /* 822 * Trigger machine check on the host. We assume all the MSRs are already set up 823 * by the CPU and that we still run on the same CPU as the MCE occurred on. 824 * We pass a fake environment to the machine check handler because we want 825 * the guest to be always treated like user space, no matter what context 826 * it used internally. 827 */ 828 static inline void kvm_machine_check(void) 829 { 830 #if defined(CONFIG_X86_MCE) 831 struct pt_regs regs = { 832 .cs = 3, /* Fake ring 3 no matter what the guest ran on */ 833 .flags = X86_EFLAGS_IF, 834 }; 835 836 do_machine_check(®s); 837 #endif 838 } 839 840 int kvm_handle_memory_failure(struct kvm_vcpu *vcpu, int r, 841 struct x86_exception *e); 842 void kvm_invalidate_pcid(struct kvm_vcpu *vcpu, unsigned long pcid); 843 int kvm_handle_invpcid(struct kvm_vcpu *vcpu, unsigned long type, gva_t gva); 844 845 int kvm_sev_es_mmio(struct kvm_vcpu *vcpu, bool is_write, gpa_t gpa, 846 unsigned int bytes, void *data); 847 int kvm_sev_es_string_io(struct kvm_vcpu *vcpu, unsigned int size, 848 unsigned int port, void *data, unsigned int count, 849 int in); 850 851 static inline void __kvm_prepare_emulated_mmio_exit(struct kvm_vcpu *vcpu, 852 gpa_t gpa, unsigned int len, 853 const void *data, 854 bool is_write) 855 { 856 struct kvm_run *run = vcpu->run; 857 858 KVM_BUG_ON(len > 8, vcpu->kvm); 859 860 run->mmio.len = len; 861 run->mmio.is_write = is_write; 862 run->exit_reason = KVM_EXIT_MMIO; 863 run->mmio.phys_addr = gpa; 864 if (is_write) 865 memcpy(run->mmio.data, data, len); 866 } 867 868 static inline void kvm_prepare_emulated_mmio_exit(struct kvm_vcpu *vcpu, 869 struct kvm_mmio_fragment *frag) 870 { 871 WARN_ON_ONCE(!vcpu->mmio_needed || !vcpu->mmio_nr_fragments); 872 873 __kvm_prepare_emulated_mmio_exit(vcpu, frag->gpa, min(8u, frag->len), 874 frag->data, vcpu->mmio_is_write); 875 } 876 877 static inline bool kvm_is_valid_map_gpa_range_ret(u64 hypercall_ret) 878 { 879 return !hypercall_ret || hypercall_ret == EINVAL || 880 hypercall_ret == EAGAIN; 881 } 882 883 static inline bool user_exit_on_hypercall(struct kvm *kvm, unsigned long hc_nr) 884 { 885 return kvm->arch.hypercall_exit_enabled & BIT(hc_nr); 886 } 887 888 int ____kvm_emulate_hypercall(struct kvm_vcpu *vcpu, int cpl, 889 int (*complete_hypercall)(struct kvm_vcpu *)); 890 891 #define __kvm_emulate_hypercall(_vcpu, cpl, complete_hypercall) \ 892 ({ \ 893 int __ret; \ 894 __ret = ____kvm_emulate_hypercall(_vcpu, cpl, complete_hypercall); \ 895 \ 896 if (__ret > 0) \ 897 __ret = complete_hypercall(_vcpu); \ 898 __ret; \ 899 }) 900 901 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu); 902 903 #endif 904