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 "kvm_cache_regs.h" 10 #include "kvm_emulate.h" 11 12 struct kvm_caps { 13 /* control of guest tsc rate supported? */ 14 bool has_tsc_control; 15 /* maximum supported tsc_khz for guests */ 16 u32 max_guest_tsc_khz; 17 /* number of bits of the fractional part of the TSC scaling ratio */ 18 u8 tsc_scaling_ratio_frac_bits; 19 /* maximum allowed value of TSC scaling ratio */ 20 u64 max_tsc_scaling_ratio; 21 /* 1ull << kvm_caps.tsc_scaling_ratio_frac_bits */ 22 u64 default_tsc_scaling_ratio; 23 /* bus lock detection supported? */ 24 bool has_bus_lock_exit; 25 /* notify VM exit supported? */ 26 bool has_notify_vmexit; 27 28 u64 supported_mce_cap; 29 u64 supported_xcr0; 30 u64 supported_xss; 31 u64 supported_perf_cap; 32 }; 33 34 void kvm_spurious_fault(void); 35 36 #define KVM_NESTED_VMENTER_CONSISTENCY_CHECK(consistency_check) \ 37 ({ \ 38 bool failed = (consistency_check); \ 39 if (failed) \ 40 trace_kvm_nested_vmenter_failed(#consistency_check, 0); \ 41 failed; \ 42 }) 43 44 /* 45 * The first...last VMX feature MSRs that are emulated by KVM. This may or may 46 * not cover all known VMX MSRs, as KVM doesn't emulate an MSR until there's an 47 * associated feature that KVM supports for nested virtualization. 48 */ 49 #define KVM_FIRST_EMULATED_VMX_MSR MSR_IA32_VMX_BASIC 50 #define KVM_LAST_EMULATED_VMX_MSR MSR_IA32_VMX_VMFUNC 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 #define MSR_IA32_CR_PAT_DEFAULT 0x0007040600070406ULL 91 92 void kvm_service_local_tlb_flush_requests(struct kvm_vcpu *vcpu); 93 int kvm_check_nested_events(struct kvm_vcpu *vcpu); 94 95 static inline bool kvm_vcpu_has_run(struct kvm_vcpu *vcpu) 96 { 97 return vcpu->arch.last_vmentry_cpu != -1; 98 } 99 100 static inline bool kvm_is_exception_pending(struct kvm_vcpu *vcpu) 101 { 102 return vcpu->arch.exception.pending || 103 vcpu->arch.exception_vmexit.pending || 104 kvm_test_request(KVM_REQ_TRIPLE_FAULT, vcpu); 105 } 106 107 static inline void kvm_clear_exception_queue(struct kvm_vcpu *vcpu) 108 { 109 vcpu->arch.exception.pending = false; 110 vcpu->arch.exception.injected = false; 111 vcpu->arch.exception_vmexit.pending = false; 112 } 113 114 static inline void kvm_queue_interrupt(struct kvm_vcpu *vcpu, u8 vector, 115 bool soft) 116 { 117 vcpu->arch.interrupt.injected = true; 118 vcpu->arch.interrupt.soft = soft; 119 vcpu->arch.interrupt.nr = vector; 120 } 121 122 static inline void kvm_clear_interrupt_queue(struct kvm_vcpu *vcpu) 123 { 124 vcpu->arch.interrupt.injected = false; 125 } 126 127 static inline bool kvm_event_needs_reinjection(struct kvm_vcpu *vcpu) 128 { 129 return vcpu->arch.exception.injected || vcpu->arch.interrupt.injected || 130 vcpu->arch.nmi_injected; 131 } 132 133 static inline bool kvm_exception_is_soft(unsigned int nr) 134 { 135 return (nr == BP_VECTOR) || (nr == OF_VECTOR); 136 } 137 138 static inline bool is_protmode(struct kvm_vcpu *vcpu) 139 { 140 return kvm_is_cr0_bit_set(vcpu, X86_CR0_PE); 141 } 142 143 static inline bool is_long_mode(struct kvm_vcpu *vcpu) 144 { 145 #ifdef CONFIG_X86_64 146 return !!(vcpu->arch.efer & EFER_LMA); 147 #else 148 return false; 149 #endif 150 } 151 152 static inline bool is_64_bit_mode(struct kvm_vcpu *vcpu) 153 { 154 int cs_db, cs_l; 155 156 WARN_ON_ONCE(vcpu->arch.guest_state_protected); 157 158 if (!is_long_mode(vcpu)) 159 return false; 160 static_call(kvm_x86_get_cs_db_l_bits)(vcpu, &cs_db, &cs_l); 161 return cs_l; 162 } 163 164 static inline bool is_64_bit_hypercall(struct kvm_vcpu *vcpu) 165 { 166 /* 167 * If running with protected guest state, the CS register is not 168 * accessible. The hypercall register values will have had to been 169 * provided in 64-bit mode, so assume the guest is in 64-bit. 170 */ 171 return vcpu->arch.guest_state_protected || is_64_bit_mode(vcpu); 172 } 173 174 static inline bool x86_exception_has_error_code(unsigned int vector) 175 { 176 static u32 exception_has_error_code = BIT(DF_VECTOR) | BIT(TS_VECTOR) | 177 BIT(NP_VECTOR) | BIT(SS_VECTOR) | BIT(GP_VECTOR) | 178 BIT(PF_VECTOR) | BIT(AC_VECTOR); 179 180 return (1U << vector) & exception_has_error_code; 181 } 182 183 static inline bool mmu_is_nested(struct kvm_vcpu *vcpu) 184 { 185 return vcpu->arch.walk_mmu == &vcpu->arch.nested_mmu; 186 } 187 188 static inline bool is_pae(struct kvm_vcpu *vcpu) 189 { 190 return kvm_is_cr4_bit_set(vcpu, X86_CR4_PAE); 191 } 192 193 static inline bool is_pse(struct kvm_vcpu *vcpu) 194 { 195 return kvm_is_cr4_bit_set(vcpu, X86_CR4_PSE); 196 } 197 198 static inline bool is_paging(struct kvm_vcpu *vcpu) 199 { 200 return likely(kvm_is_cr0_bit_set(vcpu, X86_CR0_PG)); 201 } 202 203 static inline bool is_pae_paging(struct kvm_vcpu *vcpu) 204 { 205 return !is_long_mode(vcpu) && is_pae(vcpu) && is_paging(vcpu); 206 } 207 208 static inline u8 vcpu_virt_addr_bits(struct kvm_vcpu *vcpu) 209 { 210 return kvm_is_cr4_bit_set(vcpu, X86_CR4_LA57) ? 57 : 48; 211 } 212 213 static inline bool is_noncanonical_address(u64 la, struct kvm_vcpu *vcpu) 214 { 215 return !__is_canonical_address(la, vcpu_virt_addr_bits(vcpu)); 216 } 217 218 static inline void vcpu_cache_mmio_info(struct kvm_vcpu *vcpu, 219 gva_t gva, gfn_t gfn, unsigned access) 220 { 221 u64 gen = kvm_memslots(vcpu->kvm)->generation; 222 223 if (unlikely(gen & KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS)) 224 return; 225 226 /* 227 * If this is a shadow nested page table, the "GVA" is 228 * actually a nGPA. 229 */ 230 vcpu->arch.mmio_gva = mmu_is_nested(vcpu) ? 0 : gva & PAGE_MASK; 231 vcpu->arch.mmio_access = access; 232 vcpu->arch.mmio_gfn = gfn; 233 vcpu->arch.mmio_gen = gen; 234 } 235 236 static inline bool vcpu_match_mmio_gen(struct kvm_vcpu *vcpu) 237 { 238 return vcpu->arch.mmio_gen == kvm_memslots(vcpu->kvm)->generation; 239 } 240 241 /* 242 * Clear the mmio cache info for the given gva. If gva is MMIO_GVA_ANY, we 243 * clear all mmio cache info. 244 */ 245 #define MMIO_GVA_ANY (~(gva_t)0) 246 247 static inline void vcpu_clear_mmio_info(struct kvm_vcpu *vcpu, gva_t gva) 248 { 249 if (gva != MMIO_GVA_ANY && vcpu->arch.mmio_gva != (gva & PAGE_MASK)) 250 return; 251 252 vcpu->arch.mmio_gva = 0; 253 } 254 255 static inline bool vcpu_match_mmio_gva(struct kvm_vcpu *vcpu, unsigned long gva) 256 { 257 if (vcpu_match_mmio_gen(vcpu) && vcpu->arch.mmio_gva && 258 vcpu->arch.mmio_gva == (gva & PAGE_MASK)) 259 return true; 260 261 return false; 262 } 263 264 static inline bool vcpu_match_mmio_gpa(struct kvm_vcpu *vcpu, gpa_t gpa) 265 { 266 if (vcpu_match_mmio_gen(vcpu) && vcpu->arch.mmio_gfn && 267 vcpu->arch.mmio_gfn == gpa >> PAGE_SHIFT) 268 return true; 269 270 return false; 271 } 272 273 static inline unsigned long kvm_register_read(struct kvm_vcpu *vcpu, int reg) 274 { 275 unsigned long val = kvm_register_read_raw(vcpu, reg); 276 277 return is_64_bit_mode(vcpu) ? val : (u32)val; 278 } 279 280 static inline void kvm_register_write(struct kvm_vcpu *vcpu, 281 int reg, unsigned long val) 282 { 283 if (!is_64_bit_mode(vcpu)) 284 val = (u32)val; 285 return kvm_register_write_raw(vcpu, reg, val); 286 } 287 288 static inline bool kvm_check_has_quirk(struct kvm *kvm, u64 quirk) 289 { 290 return !(kvm->arch.disabled_quirks & quirk); 291 } 292 293 void kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip); 294 295 u64 get_kvmclock_ns(struct kvm *kvm); 296 uint64_t kvm_get_wall_clock_epoch(struct kvm *kvm); 297 298 int kvm_read_guest_virt(struct kvm_vcpu *vcpu, 299 gva_t addr, void *val, unsigned int bytes, 300 struct x86_exception *exception); 301 302 int kvm_write_guest_virt_system(struct kvm_vcpu *vcpu, 303 gva_t addr, void *val, unsigned int bytes, 304 struct x86_exception *exception); 305 306 int handle_ud(struct kvm_vcpu *vcpu); 307 308 void kvm_deliver_exception_payload(struct kvm_vcpu *vcpu, 309 struct kvm_queued_exception *ex); 310 311 void kvm_vcpu_mtrr_init(struct kvm_vcpu *vcpu); 312 u8 kvm_mtrr_get_guest_memory_type(struct kvm_vcpu *vcpu, gfn_t gfn); 313 int kvm_mtrr_set_msr(struct kvm_vcpu *vcpu, u32 msr, u64 data); 314 int kvm_mtrr_get_msr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata); 315 bool kvm_mtrr_check_gfn_range_consistency(struct kvm_vcpu *vcpu, gfn_t gfn, 316 int page_num); 317 bool kvm_vector_hashing_enabled(void); 318 void kvm_fixup_and_inject_pf_error(struct kvm_vcpu *vcpu, gva_t gva, u16 error_code); 319 int x86_decode_emulated_instruction(struct kvm_vcpu *vcpu, int emulation_type, 320 void *insn, int insn_len); 321 int x86_emulate_instruction(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa, 322 int emulation_type, void *insn, int insn_len); 323 fastpath_t handle_fastpath_set_msr_irqoff(struct kvm_vcpu *vcpu); 324 325 extern u64 host_xcr0; 326 extern u64 host_xss; 327 extern u64 host_arch_capabilities; 328 329 extern struct kvm_caps kvm_caps; 330 331 extern bool enable_pmu; 332 333 /* 334 * Get a filtered version of KVM's supported XCR0 that strips out dynamic 335 * features for which the current process doesn't (yet) have permission to use. 336 * This is intended to be used only when enumerating support to userspace, 337 * e.g. in KVM_GET_SUPPORTED_CPUID and KVM_CAP_XSAVE2, it does NOT need to be 338 * used to check/restrict guest behavior as KVM rejects KVM_SET_CPUID{2} if 339 * userspace attempts to enable unpermitted features. 340 */ 341 static inline u64 kvm_get_filtered_xcr0(void) 342 { 343 u64 permitted_xcr0 = kvm_caps.supported_xcr0; 344 345 BUILD_BUG_ON(XFEATURE_MASK_USER_DYNAMIC != XFEATURE_MASK_XTILE_DATA); 346 347 if (permitted_xcr0 & XFEATURE_MASK_USER_DYNAMIC) { 348 permitted_xcr0 &= xstate_get_guest_group_perm(); 349 350 /* 351 * Treat XTILE_CFG as unsupported if the current process isn't 352 * allowed to use XTILE_DATA, as attempting to set XTILE_CFG in 353 * XCR0 without setting XTILE_DATA is architecturally illegal. 354 */ 355 if (!(permitted_xcr0 & XFEATURE_MASK_XTILE_DATA)) 356 permitted_xcr0 &= ~XFEATURE_MASK_XTILE_CFG; 357 } 358 return permitted_xcr0; 359 } 360 361 static inline bool kvm_mpx_supported(void) 362 { 363 return (kvm_caps.supported_xcr0 & (XFEATURE_MASK_BNDREGS | XFEATURE_MASK_BNDCSR)) 364 == (XFEATURE_MASK_BNDREGS | XFEATURE_MASK_BNDCSR); 365 } 366 367 extern unsigned int min_timer_period_us; 368 369 extern bool enable_vmware_backdoor; 370 371 extern int pi_inject_timer; 372 373 extern bool report_ignored_msrs; 374 375 extern bool eager_page_split; 376 377 static inline void kvm_pr_unimpl_wrmsr(struct kvm_vcpu *vcpu, u32 msr, u64 data) 378 { 379 if (report_ignored_msrs) 380 vcpu_unimpl(vcpu, "Unhandled WRMSR(0x%x) = 0x%llx\n", msr, data); 381 } 382 383 static inline void kvm_pr_unimpl_rdmsr(struct kvm_vcpu *vcpu, u32 msr) 384 { 385 if (report_ignored_msrs) 386 vcpu_unimpl(vcpu, "Unhandled RDMSR(0x%x)\n", msr); 387 } 388 389 static inline u64 nsec_to_cycles(struct kvm_vcpu *vcpu, u64 nsec) 390 { 391 return pvclock_scale_delta(nsec, vcpu->arch.virtual_tsc_mult, 392 vcpu->arch.virtual_tsc_shift); 393 } 394 395 /* Same "calling convention" as do_div: 396 * - divide (n << 32) by base 397 * - put result in n 398 * - return remainder 399 */ 400 #define do_shl32_div32(n, base) \ 401 ({ \ 402 u32 __quot, __rem; \ 403 asm("divl %2" : "=a" (__quot), "=d" (__rem) \ 404 : "rm" (base), "0" (0), "1" ((u32) n)); \ 405 n = __quot; \ 406 __rem; \ 407 }) 408 409 static inline bool kvm_mwait_in_guest(struct kvm *kvm) 410 { 411 return kvm->arch.mwait_in_guest; 412 } 413 414 static inline bool kvm_hlt_in_guest(struct kvm *kvm) 415 { 416 return kvm->arch.hlt_in_guest; 417 } 418 419 static inline bool kvm_pause_in_guest(struct kvm *kvm) 420 { 421 return kvm->arch.pause_in_guest; 422 } 423 424 static inline bool kvm_cstate_in_guest(struct kvm *kvm) 425 { 426 return kvm->arch.cstate_in_guest; 427 } 428 429 static inline bool kvm_notify_vmexit_enabled(struct kvm *kvm) 430 { 431 return kvm->arch.notify_vmexit_flags & KVM_X86_NOTIFY_VMEXIT_ENABLED; 432 } 433 434 enum kvm_intr_type { 435 /* Values are arbitrary, but must be non-zero. */ 436 KVM_HANDLING_IRQ = 1, 437 KVM_HANDLING_NMI, 438 }; 439 440 static __always_inline void kvm_before_interrupt(struct kvm_vcpu *vcpu, 441 enum kvm_intr_type intr) 442 { 443 WRITE_ONCE(vcpu->arch.handling_intr_from_guest, (u8)intr); 444 } 445 446 static __always_inline void kvm_after_interrupt(struct kvm_vcpu *vcpu) 447 { 448 WRITE_ONCE(vcpu->arch.handling_intr_from_guest, 0); 449 } 450 451 static inline bool kvm_handling_nmi_from_guest(struct kvm_vcpu *vcpu) 452 { 453 return vcpu->arch.handling_intr_from_guest == KVM_HANDLING_NMI; 454 } 455 456 static inline bool kvm_pat_valid(u64 data) 457 { 458 if (data & 0xF8F8F8F8F8F8F8F8ull) 459 return false; 460 /* 0, 1, 4, 5, 6, 7 are valid values. */ 461 return (data | ((data & 0x0202020202020202ull) << 1)) == data; 462 } 463 464 static inline bool kvm_dr7_valid(u64 data) 465 { 466 /* Bits [63:32] are reserved */ 467 return !(data >> 32); 468 } 469 static inline bool kvm_dr6_valid(u64 data) 470 { 471 /* Bits [63:32] are reserved */ 472 return !(data >> 32); 473 } 474 475 /* 476 * Trigger machine check on the host. We assume all the MSRs are already set up 477 * by the CPU and that we still run on the same CPU as the MCE occurred on. 478 * We pass a fake environment to the machine check handler because we want 479 * the guest to be always treated like user space, no matter what context 480 * it used internally. 481 */ 482 static inline void kvm_machine_check(void) 483 { 484 #if defined(CONFIG_X86_MCE) 485 struct pt_regs regs = { 486 .cs = 3, /* Fake ring 3 no matter what the guest ran on */ 487 .flags = X86_EFLAGS_IF, 488 }; 489 490 do_machine_check(®s); 491 #endif 492 } 493 494 void kvm_load_guest_xsave_state(struct kvm_vcpu *vcpu); 495 void kvm_load_host_xsave_state(struct kvm_vcpu *vcpu); 496 int kvm_spec_ctrl_test_value(u64 value); 497 bool __kvm_is_valid_cr4(struct kvm_vcpu *vcpu, unsigned long cr4); 498 int kvm_handle_memory_failure(struct kvm_vcpu *vcpu, int r, 499 struct x86_exception *e); 500 int kvm_handle_invpcid(struct kvm_vcpu *vcpu, unsigned long type, gva_t gva); 501 bool kvm_msr_allowed(struct kvm_vcpu *vcpu, u32 index, u32 type); 502 503 /* 504 * Internal error codes that are used to indicate that MSR emulation encountered 505 * an error that should result in #GP in the guest, unless userspace 506 * handles it. 507 */ 508 #define KVM_MSR_RET_INVALID 2 /* in-kernel MSR emulation #GP condition */ 509 #define KVM_MSR_RET_FILTERED 3 /* #GP due to userspace MSR filter */ 510 511 #define __cr4_reserved_bits(__cpu_has, __c) \ 512 ({ \ 513 u64 __reserved_bits = CR4_RESERVED_BITS; \ 514 \ 515 if (!__cpu_has(__c, X86_FEATURE_XSAVE)) \ 516 __reserved_bits |= X86_CR4_OSXSAVE; \ 517 if (!__cpu_has(__c, X86_FEATURE_SMEP)) \ 518 __reserved_bits |= X86_CR4_SMEP; \ 519 if (!__cpu_has(__c, X86_FEATURE_SMAP)) \ 520 __reserved_bits |= X86_CR4_SMAP; \ 521 if (!__cpu_has(__c, X86_FEATURE_FSGSBASE)) \ 522 __reserved_bits |= X86_CR4_FSGSBASE; \ 523 if (!__cpu_has(__c, X86_FEATURE_PKU)) \ 524 __reserved_bits |= X86_CR4_PKE; \ 525 if (!__cpu_has(__c, X86_FEATURE_LA57)) \ 526 __reserved_bits |= X86_CR4_LA57; \ 527 if (!__cpu_has(__c, X86_FEATURE_UMIP)) \ 528 __reserved_bits |= X86_CR4_UMIP; \ 529 if (!__cpu_has(__c, X86_FEATURE_VMX)) \ 530 __reserved_bits |= X86_CR4_VMXE; \ 531 if (!__cpu_has(__c, X86_FEATURE_PCID)) \ 532 __reserved_bits |= X86_CR4_PCIDE; \ 533 __reserved_bits; \ 534 }) 535 536 int kvm_sev_es_mmio_write(struct kvm_vcpu *vcpu, gpa_t src, unsigned int bytes, 537 void *dst); 538 int kvm_sev_es_mmio_read(struct kvm_vcpu *vcpu, gpa_t src, unsigned int bytes, 539 void *dst); 540 int kvm_sev_es_string_io(struct kvm_vcpu *vcpu, unsigned int size, 541 unsigned int port, void *data, unsigned int count, 542 int in); 543 544 #endif 545