1 /* SPDX-License-Identifier: GPL-2.0-only */ 2 /* 3 * tools/testing/selftests/kvm/include/x86_64/processor.h 4 * 5 * Copyright (C) 2018, Google LLC. 6 */ 7 8 #ifndef SELFTEST_KVM_PROCESSOR_H 9 #define SELFTEST_KVM_PROCESSOR_H 10 11 #include <assert.h> 12 #include <stdint.h> 13 #include <syscall.h> 14 15 #include <asm/msr-index.h> 16 #include <asm/prctl.h> 17 18 #include <linux/kvm_para.h> 19 #include <linux/stringify.h> 20 21 #include "kvm_util.h" 22 #include "ucall_common.h" 23 24 extern bool host_cpu_is_intel; 25 extern bool host_cpu_is_amd; 26 extern uint64_t guest_tsc_khz; 27 28 #ifndef MAX_NR_CPUID_ENTRIES 29 #define MAX_NR_CPUID_ENTRIES 100 30 #endif 31 32 /* Forced emulation prefix, used to invoke the emulator unconditionally. */ 33 #define KVM_FEP "ud2; .byte 'k', 'v', 'm';" 34 35 #define NMI_VECTOR 0x02 36 37 #define X86_EFLAGS_FIXED (1u << 1) 38 39 #define X86_CR4_VME (1ul << 0) 40 #define X86_CR4_PVI (1ul << 1) 41 #define X86_CR4_TSD (1ul << 2) 42 #define X86_CR4_DE (1ul << 3) 43 #define X86_CR4_PSE (1ul << 4) 44 #define X86_CR4_PAE (1ul << 5) 45 #define X86_CR4_MCE (1ul << 6) 46 #define X86_CR4_PGE (1ul << 7) 47 #define X86_CR4_PCE (1ul << 8) 48 #define X86_CR4_OSFXSR (1ul << 9) 49 #define X86_CR4_OSXMMEXCPT (1ul << 10) 50 #define X86_CR4_UMIP (1ul << 11) 51 #define X86_CR4_LA57 (1ul << 12) 52 #define X86_CR4_VMXE (1ul << 13) 53 #define X86_CR4_SMXE (1ul << 14) 54 #define X86_CR4_FSGSBASE (1ul << 16) 55 #define X86_CR4_PCIDE (1ul << 17) 56 #define X86_CR4_OSXSAVE (1ul << 18) 57 #define X86_CR4_SMEP (1ul << 20) 58 #define X86_CR4_SMAP (1ul << 21) 59 #define X86_CR4_PKE (1ul << 22) 60 61 struct xstate_header { 62 u64 xstate_bv; 63 u64 xcomp_bv; 64 u64 reserved[6]; 65 } __attribute__((packed)); 66 67 struct xstate { 68 u8 i387[512]; 69 struct xstate_header header; 70 u8 extended_state_area[0]; 71 } __attribute__ ((packed, aligned (64))); 72 73 #define XFEATURE_MASK_FP BIT_ULL(0) 74 #define XFEATURE_MASK_SSE BIT_ULL(1) 75 #define XFEATURE_MASK_YMM BIT_ULL(2) 76 #define XFEATURE_MASK_BNDREGS BIT_ULL(3) 77 #define XFEATURE_MASK_BNDCSR BIT_ULL(4) 78 #define XFEATURE_MASK_OPMASK BIT_ULL(5) 79 #define XFEATURE_MASK_ZMM_Hi256 BIT_ULL(6) 80 #define XFEATURE_MASK_Hi16_ZMM BIT_ULL(7) 81 #define XFEATURE_MASK_PT BIT_ULL(8) 82 #define XFEATURE_MASK_PKRU BIT_ULL(9) 83 #define XFEATURE_MASK_PASID BIT_ULL(10) 84 #define XFEATURE_MASK_CET_USER BIT_ULL(11) 85 #define XFEATURE_MASK_CET_KERNEL BIT_ULL(12) 86 #define XFEATURE_MASK_LBR BIT_ULL(15) 87 #define XFEATURE_MASK_XTILE_CFG BIT_ULL(17) 88 #define XFEATURE_MASK_XTILE_DATA BIT_ULL(18) 89 90 #define XFEATURE_MASK_AVX512 (XFEATURE_MASK_OPMASK | \ 91 XFEATURE_MASK_ZMM_Hi256 | \ 92 XFEATURE_MASK_Hi16_ZMM) 93 #define XFEATURE_MASK_XTILE (XFEATURE_MASK_XTILE_DATA | \ 94 XFEATURE_MASK_XTILE_CFG) 95 96 /* Note, these are ordered alphabetically to match kvm_cpuid_entry2. Eww. */ 97 enum cpuid_output_regs { 98 KVM_CPUID_EAX, 99 KVM_CPUID_EBX, 100 KVM_CPUID_ECX, 101 KVM_CPUID_EDX 102 }; 103 104 /* 105 * Pack the information into a 64-bit value so that each X86_FEATURE_XXX can be 106 * passed by value with no overhead. 107 */ 108 struct kvm_x86_cpu_feature { 109 u32 function; 110 u16 index; 111 u8 reg; 112 u8 bit; 113 }; 114 #define KVM_X86_CPU_FEATURE(fn, idx, gpr, __bit) \ 115 ({ \ 116 struct kvm_x86_cpu_feature feature = { \ 117 .function = fn, \ 118 .index = idx, \ 119 .reg = KVM_CPUID_##gpr, \ 120 .bit = __bit, \ 121 }; \ 122 \ 123 kvm_static_assert((fn & 0xc0000000) == 0 || \ 124 (fn & 0xc0000000) == 0x40000000 || \ 125 (fn & 0xc0000000) == 0x80000000 || \ 126 (fn & 0xc0000000) == 0xc0000000); \ 127 kvm_static_assert(idx < BIT(sizeof(feature.index) * BITS_PER_BYTE)); \ 128 feature; \ 129 }) 130 131 /* 132 * Basic Leafs, a.k.a. Intel defined 133 */ 134 #define X86_FEATURE_MWAIT KVM_X86_CPU_FEATURE(0x1, 0, ECX, 3) 135 #define X86_FEATURE_VMX KVM_X86_CPU_FEATURE(0x1, 0, ECX, 5) 136 #define X86_FEATURE_SMX KVM_X86_CPU_FEATURE(0x1, 0, ECX, 6) 137 #define X86_FEATURE_PDCM KVM_X86_CPU_FEATURE(0x1, 0, ECX, 15) 138 #define X86_FEATURE_PCID KVM_X86_CPU_FEATURE(0x1, 0, ECX, 17) 139 #define X86_FEATURE_X2APIC KVM_X86_CPU_FEATURE(0x1, 0, ECX, 21) 140 #define X86_FEATURE_MOVBE KVM_X86_CPU_FEATURE(0x1, 0, ECX, 22) 141 #define X86_FEATURE_TSC_DEADLINE_TIMER KVM_X86_CPU_FEATURE(0x1, 0, ECX, 24) 142 #define X86_FEATURE_XSAVE KVM_X86_CPU_FEATURE(0x1, 0, ECX, 26) 143 #define X86_FEATURE_OSXSAVE KVM_X86_CPU_FEATURE(0x1, 0, ECX, 27) 144 #define X86_FEATURE_RDRAND KVM_X86_CPU_FEATURE(0x1, 0, ECX, 30) 145 #define X86_FEATURE_HYPERVISOR KVM_X86_CPU_FEATURE(0x1, 0, ECX, 31) 146 #define X86_FEATURE_PAE KVM_X86_CPU_FEATURE(0x1, 0, EDX, 6) 147 #define X86_FEATURE_MCE KVM_X86_CPU_FEATURE(0x1, 0, EDX, 7) 148 #define X86_FEATURE_APIC KVM_X86_CPU_FEATURE(0x1, 0, EDX, 9) 149 #define X86_FEATURE_CLFLUSH KVM_X86_CPU_FEATURE(0x1, 0, EDX, 19) 150 #define X86_FEATURE_XMM KVM_X86_CPU_FEATURE(0x1, 0, EDX, 25) 151 #define X86_FEATURE_XMM2 KVM_X86_CPU_FEATURE(0x1, 0, EDX, 26) 152 #define X86_FEATURE_FSGSBASE KVM_X86_CPU_FEATURE(0x7, 0, EBX, 0) 153 #define X86_FEATURE_TSC_ADJUST KVM_X86_CPU_FEATURE(0x7, 0, EBX, 1) 154 #define X86_FEATURE_SGX KVM_X86_CPU_FEATURE(0x7, 0, EBX, 2) 155 #define X86_FEATURE_HLE KVM_X86_CPU_FEATURE(0x7, 0, EBX, 4) 156 #define X86_FEATURE_SMEP KVM_X86_CPU_FEATURE(0x7, 0, EBX, 7) 157 #define X86_FEATURE_INVPCID KVM_X86_CPU_FEATURE(0x7, 0, EBX, 10) 158 #define X86_FEATURE_RTM KVM_X86_CPU_FEATURE(0x7, 0, EBX, 11) 159 #define X86_FEATURE_MPX KVM_X86_CPU_FEATURE(0x7, 0, EBX, 14) 160 #define X86_FEATURE_SMAP KVM_X86_CPU_FEATURE(0x7, 0, EBX, 20) 161 #define X86_FEATURE_PCOMMIT KVM_X86_CPU_FEATURE(0x7, 0, EBX, 22) 162 #define X86_FEATURE_CLFLUSHOPT KVM_X86_CPU_FEATURE(0x7, 0, EBX, 23) 163 #define X86_FEATURE_CLWB KVM_X86_CPU_FEATURE(0x7, 0, EBX, 24) 164 #define X86_FEATURE_UMIP KVM_X86_CPU_FEATURE(0x7, 0, ECX, 2) 165 #define X86_FEATURE_PKU KVM_X86_CPU_FEATURE(0x7, 0, ECX, 3) 166 #define X86_FEATURE_OSPKE KVM_X86_CPU_FEATURE(0x7, 0, ECX, 4) 167 #define X86_FEATURE_LA57 KVM_X86_CPU_FEATURE(0x7, 0, ECX, 16) 168 #define X86_FEATURE_RDPID KVM_X86_CPU_FEATURE(0x7, 0, ECX, 22) 169 #define X86_FEATURE_SGX_LC KVM_X86_CPU_FEATURE(0x7, 0, ECX, 30) 170 #define X86_FEATURE_SHSTK KVM_X86_CPU_FEATURE(0x7, 0, ECX, 7) 171 #define X86_FEATURE_IBT KVM_X86_CPU_FEATURE(0x7, 0, EDX, 20) 172 #define X86_FEATURE_AMX_TILE KVM_X86_CPU_FEATURE(0x7, 0, EDX, 24) 173 #define X86_FEATURE_SPEC_CTRL KVM_X86_CPU_FEATURE(0x7, 0, EDX, 26) 174 #define X86_FEATURE_ARCH_CAPABILITIES KVM_X86_CPU_FEATURE(0x7, 0, EDX, 29) 175 #define X86_FEATURE_PKS KVM_X86_CPU_FEATURE(0x7, 0, ECX, 31) 176 #define X86_FEATURE_XTILECFG KVM_X86_CPU_FEATURE(0xD, 0, EAX, 17) 177 #define X86_FEATURE_XTILEDATA KVM_X86_CPU_FEATURE(0xD, 0, EAX, 18) 178 #define X86_FEATURE_XSAVES KVM_X86_CPU_FEATURE(0xD, 1, EAX, 3) 179 #define X86_FEATURE_XFD KVM_X86_CPU_FEATURE(0xD, 1, EAX, 4) 180 #define X86_FEATURE_XTILEDATA_XFD KVM_X86_CPU_FEATURE(0xD, 18, ECX, 2) 181 182 /* 183 * Extended Leafs, a.k.a. AMD defined 184 */ 185 #define X86_FEATURE_SVM KVM_X86_CPU_FEATURE(0x80000001, 0, ECX, 2) 186 #define X86_FEATURE_NX KVM_X86_CPU_FEATURE(0x80000001, 0, EDX, 20) 187 #define X86_FEATURE_GBPAGES KVM_X86_CPU_FEATURE(0x80000001, 0, EDX, 26) 188 #define X86_FEATURE_RDTSCP KVM_X86_CPU_FEATURE(0x80000001, 0, EDX, 27) 189 #define X86_FEATURE_LM KVM_X86_CPU_FEATURE(0x80000001, 0, EDX, 29) 190 #define X86_FEATURE_INVTSC KVM_X86_CPU_FEATURE(0x80000007, 0, EDX, 8) 191 #define X86_FEATURE_RDPRU KVM_X86_CPU_FEATURE(0x80000008, 0, EBX, 4) 192 #define X86_FEATURE_AMD_IBPB KVM_X86_CPU_FEATURE(0x80000008, 0, EBX, 12) 193 #define X86_FEATURE_NPT KVM_X86_CPU_FEATURE(0x8000000A, 0, EDX, 0) 194 #define X86_FEATURE_LBRV KVM_X86_CPU_FEATURE(0x8000000A, 0, EDX, 1) 195 #define X86_FEATURE_NRIPS KVM_X86_CPU_FEATURE(0x8000000A, 0, EDX, 3) 196 #define X86_FEATURE_TSCRATEMSR KVM_X86_CPU_FEATURE(0x8000000A, 0, EDX, 4) 197 #define X86_FEATURE_PAUSEFILTER KVM_X86_CPU_FEATURE(0x8000000A, 0, EDX, 10) 198 #define X86_FEATURE_PFTHRESHOLD KVM_X86_CPU_FEATURE(0x8000000A, 0, EDX, 12) 199 #define X86_FEATURE_VGIF KVM_X86_CPU_FEATURE(0x8000000A, 0, EDX, 16) 200 #define X86_FEATURE_SEV KVM_X86_CPU_FEATURE(0x8000001F, 0, EAX, 1) 201 #define X86_FEATURE_SEV_ES KVM_X86_CPU_FEATURE(0x8000001F, 0, EAX, 3) 202 203 /* 204 * KVM defined paravirt features. 205 */ 206 #define X86_FEATURE_KVM_CLOCKSOURCE KVM_X86_CPU_FEATURE(0x40000001, 0, EAX, 0) 207 #define X86_FEATURE_KVM_NOP_IO_DELAY KVM_X86_CPU_FEATURE(0x40000001, 0, EAX, 1) 208 #define X86_FEATURE_KVM_MMU_OP KVM_X86_CPU_FEATURE(0x40000001, 0, EAX, 2) 209 #define X86_FEATURE_KVM_CLOCKSOURCE2 KVM_X86_CPU_FEATURE(0x40000001, 0, EAX, 3) 210 #define X86_FEATURE_KVM_ASYNC_PF KVM_X86_CPU_FEATURE(0x40000001, 0, EAX, 4) 211 #define X86_FEATURE_KVM_STEAL_TIME KVM_X86_CPU_FEATURE(0x40000001, 0, EAX, 5) 212 #define X86_FEATURE_KVM_PV_EOI KVM_X86_CPU_FEATURE(0x40000001, 0, EAX, 6) 213 #define X86_FEATURE_KVM_PV_UNHALT KVM_X86_CPU_FEATURE(0x40000001, 0, EAX, 7) 214 /* Bit 8 apparently isn't used?!?! */ 215 #define X86_FEATURE_KVM_PV_TLB_FLUSH KVM_X86_CPU_FEATURE(0x40000001, 0, EAX, 9) 216 #define X86_FEATURE_KVM_ASYNC_PF_VMEXIT KVM_X86_CPU_FEATURE(0x40000001, 0, EAX, 10) 217 #define X86_FEATURE_KVM_PV_SEND_IPI KVM_X86_CPU_FEATURE(0x40000001, 0, EAX, 11) 218 #define X86_FEATURE_KVM_POLL_CONTROL KVM_X86_CPU_FEATURE(0x40000001, 0, EAX, 12) 219 #define X86_FEATURE_KVM_PV_SCHED_YIELD KVM_X86_CPU_FEATURE(0x40000001, 0, EAX, 13) 220 #define X86_FEATURE_KVM_ASYNC_PF_INT KVM_X86_CPU_FEATURE(0x40000001, 0, EAX, 14) 221 #define X86_FEATURE_KVM_MSI_EXT_DEST_ID KVM_X86_CPU_FEATURE(0x40000001, 0, EAX, 15) 222 #define X86_FEATURE_KVM_HC_MAP_GPA_RANGE KVM_X86_CPU_FEATURE(0x40000001, 0, EAX, 16) 223 #define X86_FEATURE_KVM_MIGRATION_CONTROL KVM_X86_CPU_FEATURE(0x40000001, 0, EAX, 17) 224 225 /* 226 * Same idea as X86_FEATURE_XXX, but X86_PROPERTY_XXX retrieves a multi-bit 227 * value/property as opposed to a single-bit feature. Again, pack the info 228 * into a 64-bit value to pass by value with no overhead. 229 */ 230 struct kvm_x86_cpu_property { 231 u32 function; 232 u8 index; 233 u8 reg; 234 u8 lo_bit; 235 u8 hi_bit; 236 }; 237 #define KVM_X86_CPU_PROPERTY(fn, idx, gpr, low_bit, high_bit) \ 238 ({ \ 239 struct kvm_x86_cpu_property property = { \ 240 .function = fn, \ 241 .index = idx, \ 242 .reg = KVM_CPUID_##gpr, \ 243 .lo_bit = low_bit, \ 244 .hi_bit = high_bit, \ 245 }; \ 246 \ 247 kvm_static_assert(low_bit < high_bit); \ 248 kvm_static_assert((fn & 0xc0000000) == 0 || \ 249 (fn & 0xc0000000) == 0x40000000 || \ 250 (fn & 0xc0000000) == 0x80000000 || \ 251 (fn & 0xc0000000) == 0xc0000000); \ 252 kvm_static_assert(idx < BIT(sizeof(property.index) * BITS_PER_BYTE)); \ 253 property; \ 254 }) 255 256 #define X86_PROPERTY_MAX_BASIC_LEAF KVM_X86_CPU_PROPERTY(0, 0, EAX, 0, 31) 257 #define X86_PROPERTY_PMU_VERSION KVM_X86_CPU_PROPERTY(0xa, 0, EAX, 0, 7) 258 #define X86_PROPERTY_PMU_NR_GP_COUNTERS KVM_X86_CPU_PROPERTY(0xa, 0, EAX, 8, 15) 259 #define X86_PROPERTY_PMU_GP_COUNTERS_BIT_WIDTH KVM_X86_CPU_PROPERTY(0xa, 0, EAX, 16, 23) 260 #define X86_PROPERTY_PMU_EBX_BIT_VECTOR_LENGTH KVM_X86_CPU_PROPERTY(0xa, 0, EAX, 24, 31) 261 #define X86_PROPERTY_PMU_EVENTS_MASK KVM_X86_CPU_PROPERTY(0xa, 0, EBX, 0, 7) 262 #define X86_PROPERTY_PMU_FIXED_COUNTERS_BITMASK KVM_X86_CPU_PROPERTY(0xa, 0, ECX, 0, 31) 263 #define X86_PROPERTY_PMU_NR_FIXED_COUNTERS KVM_X86_CPU_PROPERTY(0xa, 0, EDX, 0, 4) 264 #define X86_PROPERTY_PMU_FIXED_COUNTERS_BIT_WIDTH KVM_X86_CPU_PROPERTY(0xa, 0, EDX, 5, 12) 265 266 #define X86_PROPERTY_SUPPORTED_XCR0_LO KVM_X86_CPU_PROPERTY(0xd, 0, EAX, 0, 31) 267 #define X86_PROPERTY_XSTATE_MAX_SIZE_XCR0 KVM_X86_CPU_PROPERTY(0xd, 0, EBX, 0, 31) 268 #define X86_PROPERTY_XSTATE_MAX_SIZE KVM_X86_CPU_PROPERTY(0xd, 0, ECX, 0, 31) 269 #define X86_PROPERTY_SUPPORTED_XCR0_HI KVM_X86_CPU_PROPERTY(0xd, 0, EDX, 0, 31) 270 271 #define X86_PROPERTY_XSTATE_TILE_SIZE KVM_X86_CPU_PROPERTY(0xd, 18, EAX, 0, 31) 272 #define X86_PROPERTY_XSTATE_TILE_OFFSET KVM_X86_CPU_PROPERTY(0xd, 18, EBX, 0, 31) 273 #define X86_PROPERTY_AMX_MAX_PALETTE_TABLES KVM_X86_CPU_PROPERTY(0x1d, 0, EAX, 0, 31) 274 #define X86_PROPERTY_AMX_TOTAL_TILE_BYTES KVM_X86_CPU_PROPERTY(0x1d, 1, EAX, 0, 15) 275 #define X86_PROPERTY_AMX_BYTES_PER_TILE KVM_X86_CPU_PROPERTY(0x1d, 1, EAX, 16, 31) 276 #define X86_PROPERTY_AMX_BYTES_PER_ROW KVM_X86_CPU_PROPERTY(0x1d, 1, EBX, 0, 15) 277 #define X86_PROPERTY_AMX_NR_TILE_REGS KVM_X86_CPU_PROPERTY(0x1d, 1, EBX, 16, 31) 278 #define X86_PROPERTY_AMX_MAX_ROWS KVM_X86_CPU_PROPERTY(0x1d, 1, ECX, 0, 15) 279 280 #define X86_PROPERTY_MAX_KVM_LEAF KVM_X86_CPU_PROPERTY(0x40000000, 0, EAX, 0, 31) 281 282 #define X86_PROPERTY_MAX_EXT_LEAF KVM_X86_CPU_PROPERTY(0x80000000, 0, EAX, 0, 31) 283 #define X86_PROPERTY_MAX_PHY_ADDR KVM_X86_CPU_PROPERTY(0x80000008, 0, EAX, 0, 7) 284 #define X86_PROPERTY_MAX_VIRT_ADDR KVM_X86_CPU_PROPERTY(0x80000008, 0, EAX, 8, 15) 285 #define X86_PROPERTY_GUEST_MAX_PHY_ADDR KVM_X86_CPU_PROPERTY(0x80000008, 0, EAX, 16, 23) 286 #define X86_PROPERTY_SEV_C_BIT KVM_X86_CPU_PROPERTY(0x8000001F, 0, EBX, 0, 5) 287 #define X86_PROPERTY_PHYS_ADDR_REDUCTION KVM_X86_CPU_PROPERTY(0x8000001F, 0, EBX, 6, 11) 288 289 #define X86_PROPERTY_MAX_CENTAUR_LEAF KVM_X86_CPU_PROPERTY(0xC0000000, 0, EAX, 0, 31) 290 291 /* 292 * Intel's architectural PMU events are bizarre. They have a "feature" bit 293 * that indicates the feature is _not_ supported, and a property that states 294 * the length of the bit mask of unsupported features. A feature is supported 295 * if the size of the bit mask is larger than the "unavailable" bit, and said 296 * bit is not set. Fixed counters also bizarre enumeration, but inverted from 297 * arch events for general purpose counters. Fixed counters are supported if a 298 * feature flag is set **OR** the total number of fixed counters is greater 299 * than index of the counter. 300 * 301 * Wrap the events for general purpose and fixed counters to simplify checking 302 * whether or not a given architectural event is supported. 303 */ 304 struct kvm_x86_pmu_feature { 305 struct kvm_x86_cpu_feature f; 306 }; 307 #define KVM_X86_PMU_FEATURE(__reg, __bit) \ 308 ({ \ 309 struct kvm_x86_pmu_feature feature = { \ 310 .f = KVM_X86_CPU_FEATURE(0xa, 0, __reg, __bit), \ 311 }; \ 312 \ 313 kvm_static_assert(KVM_CPUID_##__reg == KVM_CPUID_EBX || \ 314 KVM_CPUID_##__reg == KVM_CPUID_ECX); \ 315 feature; \ 316 }) 317 318 #define X86_PMU_FEATURE_CPU_CYCLES KVM_X86_PMU_FEATURE(EBX, 0) 319 #define X86_PMU_FEATURE_INSNS_RETIRED KVM_X86_PMU_FEATURE(EBX, 1) 320 #define X86_PMU_FEATURE_REFERENCE_CYCLES KVM_X86_PMU_FEATURE(EBX, 2) 321 #define X86_PMU_FEATURE_LLC_REFERENCES KVM_X86_PMU_FEATURE(EBX, 3) 322 #define X86_PMU_FEATURE_LLC_MISSES KVM_X86_PMU_FEATURE(EBX, 4) 323 #define X86_PMU_FEATURE_BRANCH_INSNS_RETIRED KVM_X86_PMU_FEATURE(EBX, 5) 324 #define X86_PMU_FEATURE_BRANCHES_MISPREDICTED KVM_X86_PMU_FEATURE(EBX, 6) 325 #define X86_PMU_FEATURE_TOPDOWN_SLOTS KVM_X86_PMU_FEATURE(EBX, 7) 326 327 #define X86_PMU_FEATURE_INSNS_RETIRED_FIXED KVM_X86_PMU_FEATURE(ECX, 0) 328 #define X86_PMU_FEATURE_CPU_CYCLES_FIXED KVM_X86_PMU_FEATURE(ECX, 1) 329 #define X86_PMU_FEATURE_REFERENCE_TSC_CYCLES_FIXED KVM_X86_PMU_FEATURE(ECX, 2) 330 #define X86_PMU_FEATURE_TOPDOWN_SLOTS_FIXED KVM_X86_PMU_FEATURE(ECX, 3) 331 332 static inline unsigned int x86_family(unsigned int eax) 333 { 334 unsigned int x86; 335 336 x86 = (eax >> 8) & 0xf; 337 338 if (x86 == 0xf) 339 x86 += (eax >> 20) & 0xff; 340 341 return x86; 342 } 343 344 static inline unsigned int x86_model(unsigned int eax) 345 { 346 return ((eax >> 12) & 0xf0) | ((eax >> 4) & 0x0f); 347 } 348 349 /* Page table bitfield declarations */ 350 #define PTE_PRESENT_MASK BIT_ULL(0) 351 #define PTE_WRITABLE_MASK BIT_ULL(1) 352 #define PTE_USER_MASK BIT_ULL(2) 353 #define PTE_ACCESSED_MASK BIT_ULL(5) 354 #define PTE_DIRTY_MASK BIT_ULL(6) 355 #define PTE_LARGE_MASK BIT_ULL(7) 356 #define PTE_GLOBAL_MASK BIT_ULL(8) 357 #define PTE_NX_MASK BIT_ULL(63) 358 359 #define PHYSICAL_PAGE_MASK GENMASK_ULL(51, 12) 360 361 #define PAGE_SHIFT 12 362 #define PAGE_SIZE (1ULL << PAGE_SHIFT) 363 #define PAGE_MASK (~(PAGE_SIZE-1) & PHYSICAL_PAGE_MASK) 364 365 #define HUGEPAGE_SHIFT(x) (PAGE_SHIFT + (((x) - 1) * 9)) 366 #define HUGEPAGE_SIZE(x) (1UL << HUGEPAGE_SHIFT(x)) 367 #define HUGEPAGE_MASK(x) (~(HUGEPAGE_SIZE(x) - 1) & PHYSICAL_PAGE_MASK) 368 369 #define PTE_GET_PA(pte) ((pte) & PHYSICAL_PAGE_MASK) 370 #define PTE_GET_PFN(pte) (PTE_GET_PA(pte) >> PAGE_SHIFT) 371 372 /* General Registers in 64-Bit Mode */ 373 struct gpr64_regs { 374 u64 rax; 375 u64 rcx; 376 u64 rdx; 377 u64 rbx; 378 u64 rsp; 379 u64 rbp; 380 u64 rsi; 381 u64 rdi; 382 u64 r8; 383 u64 r9; 384 u64 r10; 385 u64 r11; 386 u64 r12; 387 u64 r13; 388 u64 r14; 389 u64 r15; 390 }; 391 392 struct desc64 { 393 uint16_t limit0; 394 uint16_t base0; 395 unsigned base1:8, type:4, s:1, dpl:2, p:1; 396 unsigned limit1:4, avl:1, l:1, db:1, g:1, base2:8; 397 uint32_t base3; 398 uint32_t zero1; 399 } __attribute__((packed)); 400 401 struct desc_ptr { 402 uint16_t size; 403 uint64_t address; 404 } __attribute__((packed)); 405 406 struct kvm_x86_state { 407 struct kvm_xsave *xsave; 408 struct kvm_vcpu_events events; 409 struct kvm_mp_state mp_state; 410 struct kvm_regs regs; 411 struct kvm_xcrs xcrs; 412 struct kvm_sregs sregs; 413 struct kvm_debugregs debugregs; 414 union { 415 struct kvm_nested_state nested; 416 char nested_[16384]; 417 }; 418 struct kvm_msrs msrs; 419 }; 420 421 static inline uint64_t get_desc64_base(const struct desc64 *desc) 422 { 423 return ((uint64_t)desc->base3 << 32) | 424 (desc->base0 | ((desc->base1) << 16) | ((desc->base2) << 24)); 425 } 426 427 static inline uint64_t rdtsc(void) 428 { 429 uint32_t eax, edx; 430 uint64_t tsc_val; 431 /* 432 * The lfence is to wait (on Intel CPUs) until all previous 433 * instructions have been executed. If software requires RDTSC to be 434 * executed prior to execution of any subsequent instruction, it can 435 * execute LFENCE immediately after RDTSC 436 */ 437 __asm__ __volatile__("lfence; rdtsc; lfence" : "=a"(eax), "=d"(edx)); 438 tsc_val = ((uint64_t)edx) << 32 | eax; 439 return tsc_val; 440 } 441 442 static inline uint64_t rdtscp(uint32_t *aux) 443 { 444 uint32_t eax, edx; 445 446 __asm__ __volatile__("rdtscp" : "=a"(eax), "=d"(edx), "=c"(*aux)); 447 return ((uint64_t)edx) << 32 | eax; 448 } 449 450 static inline uint64_t rdmsr(uint32_t msr) 451 { 452 uint32_t a, d; 453 454 __asm__ __volatile__("rdmsr" : "=a"(a), "=d"(d) : "c"(msr) : "memory"); 455 456 return a | ((uint64_t) d << 32); 457 } 458 459 static inline void wrmsr(uint32_t msr, uint64_t value) 460 { 461 uint32_t a = value; 462 uint32_t d = value >> 32; 463 464 __asm__ __volatile__("wrmsr" :: "a"(a), "d"(d), "c"(msr) : "memory"); 465 } 466 467 468 static inline uint16_t inw(uint16_t port) 469 { 470 uint16_t tmp; 471 472 __asm__ __volatile__("in %%dx, %%ax" 473 : /* output */ "=a" (tmp) 474 : /* input */ "d" (port)); 475 476 return tmp; 477 } 478 479 static inline uint16_t get_es(void) 480 { 481 uint16_t es; 482 483 __asm__ __volatile__("mov %%es, %[es]" 484 : /* output */ [es]"=rm"(es)); 485 return es; 486 } 487 488 static inline uint16_t get_cs(void) 489 { 490 uint16_t cs; 491 492 __asm__ __volatile__("mov %%cs, %[cs]" 493 : /* output */ [cs]"=rm"(cs)); 494 return cs; 495 } 496 497 static inline uint16_t get_ss(void) 498 { 499 uint16_t ss; 500 501 __asm__ __volatile__("mov %%ss, %[ss]" 502 : /* output */ [ss]"=rm"(ss)); 503 return ss; 504 } 505 506 static inline uint16_t get_ds(void) 507 { 508 uint16_t ds; 509 510 __asm__ __volatile__("mov %%ds, %[ds]" 511 : /* output */ [ds]"=rm"(ds)); 512 return ds; 513 } 514 515 static inline uint16_t get_fs(void) 516 { 517 uint16_t fs; 518 519 __asm__ __volatile__("mov %%fs, %[fs]" 520 : /* output */ [fs]"=rm"(fs)); 521 return fs; 522 } 523 524 static inline uint16_t get_gs(void) 525 { 526 uint16_t gs; 527 528 __asm__ __volatile__("mov %%gs, %[gs]" 529 : /* output */ [gs]"=rm"(gs)); 530 return gs; 531 } 532 533 static inline uint16_t get_tr(void) 534 { 535 uint16_t tr; 536 537 __asm__ __volatile__("str %[tr]" 538 : /* output */ [tr]"=rm"(tr)); 539 return tr; 540 } 541 542 static inline uint64_t get_cr0(void) 543 { 544 uint64_t cr0; 545 546 __asm__ __volatile__("mov %%cr0, %[cr0]" 547 : /* output */ [cr0]"=r"(cr0)); 548 return cr0; 549 } 550 551 static inline uint64_t get_cr3(void) 552 { 553 uint64_t cr3; 554 555 __asm__ __volatile__("mov %%cr3, %[cr3]" 556 : /* output */ [cr3]"=r"(cr3)); 557 return cr3; 558 } 559 560 static inline uint64_t get_cr4(void) 561 { 562 uint64_t cr4; 563 564 __asm__ __volatile__("mov %%cr4, %[cr4]" 565 : /* output */ [cr4]"=r"(cr4)); 566 return cr4; 567 } 568 569 static inline void set_cr4(uint64_t val) 570 { 571 __asm__ __volatile__("mov %0, %%cr4" : : "r" (val) : "memory"); 572 } 573 574 static inline u64 xgetbv(u32 index) 575 { 576 u32 eax, edx; 577 578 __asm__ __volatile__("xgetbv;" 579 : "=a" (eax), "=d" (edx) 580 : "c" (index)); 581 return eax | ((u64)edx << 32); 582 } 583 584 static inline void xsetbv(u32 index, u64 value) 585 { 586 u32 eax = value; 587 u32 edx = value >> 32; 588 589 __asm__ __volatile__("xsetbv" :: "a" (eax), "d" (edx), "c" (index)); 590 } 591 592 static inline void wrpkru(u32 pkru) 593 { 594 /* Note, ECX and EDX are architecturally required to be '0'. */ 595 asm volatile(".byte 0x0f,0x01,0xef\n\t" 596 : : "a" (pkru), "c"(0), "d"(0)); 597 } 598 599 static inline struct desc_ptr get_gdt(void) 600 { 601 struct desc_ptr gdt; 602 __asm__ __volatile__("sgdt %[gdt]" 603 : /* output */ [gdt]"=m"(gdt)); 604 return gdt; 605 } 606 607 static inline struct desc_ptr get_idt(void) 608 { 609 struct desc_ptr idt; 610 __asm__ __volatile__("sidt %[idt]" 611 : /* output */ [idt]"=m"(idt)); 612 return idt; 613 } 614 615 static inline void outl(uint16_t port, uint32_t value) 616 { 617 __asm__ __volatile__("outl %%eax, %%dx" : : "d"(port), "a"(value)); 618 } 619 620 static inline void __cpuid(uint32_t function, uint32_t index, 621 uint32_t *eax, uint32_t *ebx, 622 uint32_t *ecx, uint32_t *edx) 623 { 624 *eax = function; 625 *ecx = index; 626 627 asm volatile("cpuid" 628 : "=a" (*eax), 629 "=b" (*ebx), 630 "=c" (*ecx), 631 "=d" (*edx) 632 : "0" (*eax), "2" (*ecx) 633 : "memory"); 634 } 635 636 static inline void cpuid(uint32_t function, 637 uint32_t *eax, uint32_t *ebx, 638 uint32_t *ecx, uint32_t *edx) 639 { 640 return __cpuid(function, 0, eax, ebx, ecx, edx); 641 } 642 643 static inline uint32_t this_cpu_fms(void) 644 { 645 uint32_t eax, ebx, ecx, edx; 646 647 cpuid(1, &eax, &ebx, &ecx, &edx); 648 return eax; 649 } 650 651 static inline uint32_t this_cpu_family(void) 652 { 653 return x86_family(this_cpu_fms()); 654 } 655 656 static inline uint32_t this_cpu_model(void) 657 { 658 return x86_model(this_cpu_fms()); 659 } 660 661 static inline bool this_cpu_vendor_string_is(const char *vendor) 662 { 663 const uint32_t *chunk = (const uint32_t *)vendor; 664 uint32_t eax, ebx, ecx, edx; 665 666 cpuid(0, &eax, &ebx, &ecx, &edx); 667 return (ebx == chunk[0] && edx == chunk[1] && ecx == chunk[2]); 668 } 669 670 static inline bool this_cpu_is_intel(void) 671 { 672 return this_cpu_vendor_string_is("GenuineIntel"); 673 } 674 675 /* 676 * Exclude early K5 samples with a vendor string of "AMDisbetter!" 677 */ 678 static inline bool this_cpu_is_amd(void) 679 { 680 return this_cpu_vendor_string_is("AuthenticAMD"); 681 } 682 683 static inline uint32_t __this_cpu_has(uint32_t function, uint32_t index, 684 uint8_t reg, uint8_t lo, uint8_t hi) 685 { 686 uint32_t gprs[4]; 687 688 __cpuid(function, index, 689 &gprs[KVM_CPUID_EAX], &gprs[KVM_CPUID_EBX], 690 &gprs[KVM_CPUID_ECX], &gprs[KVM_CPUID_EDX]); 691 692 return (gprs[reg] & GENMASK(hi, lo)) >> lo; 693 } 694 695 static inline bool this_cpu_has(struct kvm_x86_cpu_feature feature) 696 { 697 return __this_cpu_has(feature.function, feature.index, 698 feature.reg, feature.bit, feature.bit); 699 } 700 701 static inline uint32_t this_cpu_property(struct kvm_x86_cpu_property property) 702 { 703 return __this_cpu_has(property.function, property.index, 704 property.reg, property.lo_bit, property.hi_bit); 705 } 706 707 static __always_inline bool this_cpu_has_p(struct kvm_x86_cpu_property property) 708 { 709 uint32_t max_leaf; 710 711 switch (property.function & 0xc0000000) { 712 case 0: 713 max_leaf = this_cpu_property(X86_PROPERTY_MAX_BASIC_LEAF); 714 break; 715 case 0x40000000: 716 max_leaf = this_cpu_property(X86_PROPERTY_MAX_KVM_LEAF); 717 break; 718 case 0x80000000: 719 max_leaf = this_cpu_property(X86_PROPERTY_MAX_EXT_LEAF); 720 break; 721 case 0xc0000000: 722 max_leaf = this_cpu_property(X86_PROPERTY_MAX_CENTAUR_LEAF); 723 } 724 return max_leaf >= property.function; 725 } 726 727 static inline bool this_pmu_has(struct kvm_x86_pmu_feature feature) 728 { 729 uint32_t nr_bits; 730 731 if (feature.f.reg == KVM_CPUID_EBX) { 732 nr_bits = this_cpu_property(X86_PROPERTY_PMU_EBX_BIT_VECTOR_LENGTH); 733 return nr_bits > feature.f.bit && !this_cpu_has(feature.f); 734 } 735 736 GUEST_ASSERT(feature.f.reg == KVM_CPUID_ECX); 737 nr_bits = this_cpu_property(X86_PROPERTY_PMU_NR_FIXED_COUNTERS); 738 return nr_bits > feature.f.bit || this_cpu_has(feature.f); 739 } 740 741 static __always_inline uint64_t this_cpu_supported_xcr0(void) 742 { 743 if (!this_cpu_has_p(X86_PROPERTY_SUPPORTED_XCR0_LO)) 744 return 0; 745 746 return this_cpu_property(X86_PROPERTY_SUPPORTED_XCR0_LO) | 747 ((uint64_t)this_cpu_property(X86_PROPERTY_SUPPORTED_XCR0_HI) << 32); 748 } 749 750 typedef u32 __attribute__((vector_size(16))) sse128_t; 751 #define __sse128_u union { sse128_t vec; u64 as_u64[2]; u32 as_u32[4]; } 752 #define sse128_lo(x) ({ __sse128_u t; t.vec = x; t.as_u64[0]; }) 753 #define sse128_hi(x) ({ __sse128_u t; t.vec = x; t.as_u64[1]; }) 754 755 static inline void read_sse_reg(int reg, sse128_t *data) 756 { 757 switch (reg) { 758 case 0: 759 asm("movdqa %%xmm0, %0" : "=m"(*data)); 760 break; 761 case 1: 762 asm("movdqa %%xmm1, %0" : "=m"(*data)); 763 break; 764 case 2: 765 asm("movdqa %%xmm2, %0" : "=m"(*data)); 766 break; 767 case 3: 768 asm("movdqa %%xmm3, %0" : "=m"(*data)); 769 break; 770 case 4: 771 asm("movdqa %%xmm4, %0" : "=m"(*data)); 772 break; 773 case 5: 774 asm("movdqa %%xmm5, %0" : "=m"(*data)); 775 break; 776 case 6: 777 asm("movdqa %%xmm6, %0" : "=m"(*data)); 778 break; 779 case 7: 780 asm("movdqa %%xmm7, %0" : "=m"(*data)); 781 break; 782 default: 783 BUG(); 784 } 785 } 786 787 static inline void write_sse_reg(int reg, const sse128_t *data) 788 { 789 switch (reg) { 790 case 0: 791 asm("movdqa %0, %%xmm0" : : "m"(*data)); 792 break; 793 case 1: 794 asm("movdqa %0, %%xmm1" : : "m"(*data)); 795 break; 796 case 2: 797 asm("movdqa %0, %%xmm2" : : "m"(*data)); 798 break; 799 case 3: 800 asm("movdqa %0, %%xmm3" : : "m"(*data)); 801 break; 802 case 4: 803 asm("movdqa %0, %%xmm4" : : "m"(*data)); 804 break; 805 case 5: 806 asm("movdqa %0, %%xmm5" : : "m"(*data)); 807 break; 808 case 6: 809 asm("movdqa %0, %%xmm6" : : "m"(*data)); 810 break; 811 case 7: 812 asm("movdqa %0, %%xmm7" : : "m"(*data)); 813 break; 814 default: 815 BUG(); 816 } 817 } 818 819 static inline void cpu_relax(void) 820 { 821 asm volatile("rep; nop" ::: "memory"); 822 } 823 824 static inline void udelay(unsigned long usec) 825 { 826 uint64_t start, now, cycles; 827 828 GUEST_ASSERT(guest_tsc_khz); 829 cycles = guest_tsc_khz / 1000 * usec; 830 831 /* 832 * Deliberately don't PAUSE, a.k.a. cpu_relax(), so that the delay is 833 * as accurate as possible, e.g. doesn't trigger PAUSE-Loop VM-Exits. 834 */ 835 start = rdtsc(); 836 do { 837 now = rdtsc(); 838 } while (now - start < cycles); 839 } 840 841 #define ud2() \ 842 __asm__ __volatile__( \ 843 "ud2\n" \ 844 ) 845 846 #define hlt() \ 847 __asm__ __volatile__( \ 848 "hlt\n" \ 849 ) 850 851 struct kvm_x86_state *vcpu_save_state(struct kvm_vcpu *vcpu); 852 void vcpu_load_state(struct kvm_vcpu *vcpu, struct kvm_x86_state *state); 853 void kvm_x86_state_cleanup(struct kvm_x86_state *state); 854 855 const struct kvm_msr_list *kvm_get_msr_index_list(void); 856 const struct kvm_msr_list *kvm_get_feature_msr_index_list(void); 857 bool kvm_msr_is_in_save_restore_list(uint32_t msr_index); 858 uint64_t kvm_get_feature_msr(uint64_t msr_index); 859 860 static inline void vcpu_msrs_get(struct kvm_vcpu *vcpu, 861 struct kvm_msrs *msrs) 862 { 863 int r = __vcpu_ioctl(vcpu, KVM_GET_MSRS, msrs); 864 865 TEST_ASSERT(r == msrs->nmsrs, 866 "KVM_GET_MSRS failed, r: %i (failed on MSR %x)", 867 r, r < 0 || r >= msrs->nmsrs ? -1 : msrs->entries[r].index); 868 } 869 static inline void vcpu_msrs_set(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs) 870 { 871 int r = __vcpu_ioctl(vcpu, KVM_SET_MSRS, msrs); 872 873 TEST_ASSERT(r == msrs->nmsrs, 874 "KVM_SET_MSRS failed, r: %i (failed on MSR %x)", 875 r, r < 0 || r >= msrs->nmsrs ? -1 : msrs->entries[r].index); 876 } 877 static inline void vcpu_debugregs_get(struct kvm_vcpu *vcpu, 878 struct kvm_debugregs *debugregs) 879 { 880 vcpu_ioctl(vcpu, KVM_GET_DEBUGREGS, debugregs); 881 } 882 static inline void vcpu_debugregs_set(struct kvm_vcpu *vcpu, 883 struct kvm_debugregs *debugregs) 884 { 885 vcpu_ioctl(vcpu, KVM_SET_DEBUGREGS, debugregs); 886 } 887 static inline void vcpu_xsave_get(struct kvm_vcpu *vcpu, 888 struct kvm_xsave *xsave) 889 { 890 vcpu_ioctl(vcpu, KVM_GET_XSAVE, xsave); 891 } 892 static inline void vcpu_xsave2_get(struct kvm_vcpu *vcpu, 893 struct kvm_xsave *xsave) 894 { 895 vcpu_ioctl(vcpu, KVM_GET_XSAVE2, xsave); 896 } 897 static inline void vcpu_xsave_set(struct kvm_vcpu *vcpu, 898 struct kvm_xsave *xsave) 899 { 900 vcpu_ioctl(vcpu, KVM_SET_XSAVE, xsave); 901 } 902 static inline void vcpu_xcrs_get(struct kvm_vcpu *vcpu, 903 struct kvm_xcrs *xcrs) 904 { 905 vcpu_ioctl(vcpu, KVM_GET_XCRS, xcrs); 906 } 907 static inline void vcpu_xcrs_set(struct kvm_vcpu *vcpu, struct kvm_xcrs *xcrs) 908 { 909 vcpu_ioctl(vcpu, KVM_SET_XCRS, xcrs); 910 } 911 912 const struct kvm_cpuid_entry2 *get_cpuid_entry(const struct kvm_cpuid2 *cpuid, 913 uint32_t function, uint32_t index); 914 const struct kvm_cpuid2 *kvm_get_supported_cpuid(void); 915 916 static inline uint32_t kvm_cpu_fms(void) 917 { 918 return get_cpuid_entry(kvm_get_supported_cpuid(), 0x1, 0)->eax; 919 } 920 921 static inline uint32_t kvm_cpu_family(void) 922 { 923 return x86_family(kvm_cpu_fms()); 924 } 925 926 static inline uint32_t kvm_cpu_model(void) 927 { 928 return x86_model(kvm_cpu_fms()); 929 } 930 931 bool kvm_cpuid_has(const struct kvm_cpuid2 *cpuid, 932 struct kvm_x86_cpu_feature feature); 933 934 static inline bool kvm_cpu_has(struct kvm_x86_cpu_feature feature) 935 { 936 return kvm_cpuid_has(kvm_get_supported_cpuid(), feature); 937 } 938 939 uint32_t kvm_cpuid_property(const struct kvm_cpuid2 *cpuid, 940 struct kvm_x86_cpu_property property); 941 942 static inline uint32_t kvm_cpu_property(struct kvm_x86_cpu_property property) 943 { 944 return kvm_cpuid_property(kvm_get_supported_cpuid(), property); 945 } 946 947 static __always_inline bool kvm_cpu_has_p(struct kvm_x86_cpu_property property) 948 { 949 uint32_t max_leaf; 950 951 switch (property.function & 0xc0000000) { 952 case 0: 953 max_leaf = kvm_cpu_property(X86_PROPERTY_MAX_BASIC_LEAF); 954 break; 955 case 0x40000000: 956 max_leaf = kvm_cpu_property(X86_PROPERTY_MAX_KVM_LEAF); 957 break; 958 case 0x80000000: 959 max_leaf = kvm_cpu_property(X86_PROPERTY_MAX_EXT_LEAF); 960 break; 961 case 0xc0000000: 962 max_leaf = kvm_cpu_property(X86_PROPERTY_MAX_CENTAUR_LEAF); 963 } 964 return max_leaf >= property.function; 965 } 966 967 static inline bool kvm_pmu_has(struct kvm_x86_pmu_feature feature) 968 { 969 uint32_t nr_bits; 970 971 if (feature.f.reg == KVM_CPUID_EBX) { 972 nr_bits = kvm_cpu_property(X86_PROPERTY_PMU_EBX_BIT_VECTOR_LENGTH); 973 return nr_bits > feature.f.bit && !kvm_cpu_has(feature.f); 974 } 975 976 TEST_ASSERT_EQ(feature.f.reg, KVM_CPUID_ECX); 977 nr_bits = kvm_cpu_property(X86_PROPERTY_PMU_NR_FIXED_COUNTERS); 978 return nr_bits > feature.f.bit || kvm_cpu_has(feature.f); 979 } 980 981 static __always_inline uint64_t kvm_cpu_supported_xcr0(void) 982 { 983 if (!kvm_cpu_has_p(X86_PROPERTY_SUPPORTED_XCR0_LO)) 984 return 0; 985 986 return kvm_cpu_property(X86_PROPERTY_SUPPORTED_XCR0_LO) | 987 ((uint64_t)kvm_cpu_property(X86_PROPERTY_SUPPORTED_XCR0_HI) << 32); 988 } 989 990 static inline size_t kvm_cpuid2_size(int nr_entries) 991 { 992 return sizeof(struct kvm_cpuid2) + 993 sizeof(struct kvm_cpuid_entry2) * nr_entries; 994 } 995 996 /* 997 * Allocate a "struct kvm_cpuid2* instance, with the 0-length arrary of 998 * entries sized to hold @nr_entries. The caller is responsible for freeing 999 * the struct. 1000 */ 1001 static inline struct kvm_cpuid2 *allocate_kvm_cpuid2(int nr_entries) 1002 { 1003 struct kvm_cpuid2 *cpuid; 1004 1005 cpuid = malloc(kvm_cpuid2_size(nr_entries)); 1006 TEST_ASSERT(cpuid, "-ENOMEM when allocating kvm_cpuid2"); 1007 1008 cpuid->nent = nr_entries; 1009 1010 return cpuid; 1011 } 1012 1013 void vcpu_init_cpuid(struct kvm_vcpu *vcpu, const struct kvm_cpuid2 *cpuid); 1014 1015 static inline struct kvm_cpuid_entry2 *__vcpu_get_cpuid_entry(struct kvm_vcpu *vcpu, 1016 uint32_t function, 1017 uint32_t index) 1018 { 1019 return (struct kvm_cpuid_entry2 *)get_cpuid_entry(vcpu->cpuid, 1020 function, index); 1021 } 1022 1023 static inline struct kvm_cpuid_entry2 *vcpu_get_cpuid_entry(struct kvm_vcpu *vcpu, 1024 uint32_t function) 1025 { 1026 return __vcpu_get_cpuid_entry(vcpu, function, 0); 1027 } 1028 1029 static inline int __vcpu_set_cpuid(struct kvm_vcpu *vcpu) 1030 { 1031 int r; 1032 1033 TEST_ASSERT(vcpu->cpuid, "Must do vcpu_init_cpuid() first"); 1034 r = __vcpu_ioctl(vcpu, KVM_SET_CPUID2, vcpu->cpuid); 1035 if (r) 1036 return r; 1037 1038 /* On success, refresh the cache to pick up adjustments made by KVM. */ 1039 vcpu_ioctl(vcpu, KVM_GET_CPUID2, vcpu->cpuid); 1040 return 0; 1041 } 1042 1043 static inline void vcpu_set_cpuid(struct kvm_vcpu *vcpu) 1044 { 1045 TEST_ASSERT(vcpu->cpuid, "Must do vcpu_init_cpuid() first"); 1046 vcpu_ioctl(vcpu, KVM_SET_CPUID2, vcpu->cpuid); 1047 1048 /* Refresh the cache to pick up adjustments made by KVM. */ 1049 vcpu_ioctl(vcpu, KVM_GET_CPUID2, vcpu->cpuid); 1050 } 1051 1052 static inline void vcpu_get_cpuid(struct kvm_vcpu *vcpu) 1053 { 1054 vcpu_ioctl(vcpu, KVM_GET_CPUID2, vcpu->cpuid); 1055 } 1056 1057 void vcpu_set_cpuid_property(struct kvm_vcpu *vcpu, 1058 struct kvm_x86_cpu_property property, 1059 uint32_t value); 1060 void vcpu_set_cpuid_maxphyaddr(struct kvm_vcpu *vcpu, uint8_t maxphyaddr); 1061 1062 void vcpu_clear_cpuid_entry(struct kvm_vcpu *vcpu, uint32_t function); 1063 1064 static inline bool vcpu_cpuid_has(struct kvm_vcpu *vcpu, 1065 struct kvm_x86_cpu_feature feature) 1066 { 1067 struct kvm_cpuid_entry2 *entry; 1068 1069 entry = __vcpu_get_cpuid_entry(vcpu, feature.function, feature.index); 1070 return *((&entry->eax) + feature.reg) & BIT(feature.bit); 1071 } 1072 1073 void vcpu_set_or_clear_cpuid_feature(struct kvm_vcpu *vcpu, 1074 struct kvm_x86_cpu_feature feature, 1075 bool set); 1076 1077 static inline void vcpu_set_cpuid_feature(struct kvm_vcpu *vcpu, 1078 struct kvm_x86_cpu_feature feature) 1079 { 1080 vcpu_set_or_clear_cpuid_feature(vcpu, feature, true); 1081 1082 } 1083 1084 static inline void vcpu_clear_cpuid_feature(struct kvm_vcpu *vcpu, 1085 struct kvm_x86_cpu_feature feature) 1086 { 1087 vcpu_set_or_clear_cpuid_feature(vcpu, feature, false); 1088 } 1089 1090 uint64_t vcpu_get_msr(struct kvm_vcpu *vcpu, uint64_t msr_index); 1091 int _vcpu_set_msr(struct kvm_vcpu *vcpu, uint64_t msr_index, uint64_t msr_value); 1092 1093 /* 1094 * Assert on an MSR access(es) and pretty print the MSR name when possible. 1095 * Note, the caller provides the stringified name so that the name of macro is 1096 * printed, not the value the macro resolves to (due to macro expansion). 1097 */ 1098 #define TEST_ASSERT_MSR(cond, fmt, msr, str, args...) \ 1099 do { \ 1100 if (__builtin_constant_p(msr)) { \ 1101 TEST_ASSERT(cond, fmt, str, args); \ 1102 } else if (!(cond)) { \ 1103 char buf[16]; \ 1104 \ 1105 snprintf(buf, sizeof(buf), "MSR 0x%x", msr); \ 1106 TEST_ASSERT(cond, fmt, buf, args); \ 1107 } \ 1108 } while (0) 1109 1110 /* 1111 * Returns true if KVM should return the last written value when reading an MSR 1112 * from userspace, e.g. the MSR isn't a command MSR, doesn't emulate state that 1113 * is changing, etc. This is NOT an exhaustive list! The intent is to filter 1114 * out MSRs that are not durable _and_ that a selftest wants to write. 1115 */ 1116 static inline bool is_durable_msr(uint32_t msr) 1117 { 1118 return msr != MSR_IA32_TSC; 1119 } 1120 1121 #define vcpu_set_msr(vcpu, msr, val) \ 1122 do { \ 1123 uint64_t r, v = val; \ 1124 \ 1125 TEST_ASSERT_MSR(_vcpu_set_msr(vcpu, msr, v) == 1, \ 1126 "KVM_SET_MSRS failed on %s, value = 0x%lx", msr, #msr, v); \ 1127 if (!is_durable_msr(msr)) \ 1128 break; \ 1129 r = vcpu_get_msr(vcpu, msr); \ 1130 TEST_ASSERT_MSR(r == v, "Set %s to '0x%lx', got back '0x%lx'", msr, #msr, v, r);\ 1131 } while (0) 1132 1133 void kvm_get_cpu_address_width(unsigned int *pa_bits, unsigned int *va_bits); 1134 void kvm_init_vm_address_properties(struct kvm_vm *vm); 1135 bool vm_is_unrestricted_guest(struct kvm_vm *vm); 1136 1137 struct ex_regs { 1138 uint64_t rax, rcx, rdx, rbx; 1139 uint64_t rbp, rsi, rdi; 1140 uint64_t r8, r9, r10, r11; 1141 uint64_t r12, r13, r14, r15; 1142 uint64_t vector; 1143 uint64_t error_code; 1144 uint64_t rip; 1145 uint64_t cs; 1146 uint64_t rflags; 1147 }; 1148 1149 struct idt_entry { 1150 uint16_t offset0; 1151 uint16_t selector; 1152 uint16_t ist : 3; 1153 uint16_t : 5; 1154 uint16_t type : 4; 1155 uint16_t : 1; 1156 uint16_t dpl : 2; 1157 uint16_t p : 1; 1158 uint16_t offset1; 1159 uint32_t offset2; uint32_t reserved; 1160 }; 1161 1162 void vm_install_exception_handler(struct kvm_vm *vm, int vector, 1163 void (*handler)(struct ex_regs *)); 1164 1165 /* If a toddler were to say "abracadabra". */ 1166 #define KVM_EXCEPTION_MAGIC 0xabacadabaULL 1167 1168 /* 1169 * KVM selftest exception fixup uses registers to coordinate with the exception 1170 * handler, versus the kernel's in-memory tables and KVM-Unit-Tests's in-memory 1171 * per-CPU data. Using only registers avoids having to map memory into the 1172 * guest, doesn't require a valid, stable GS.base, and reduces the risk of 1173 * for recursive faults when accessing memory in the handler. The downside to 1174 * using registers is that it restricts what registers can be used by the actual 1175 * instruction. But, selftests are 64-bit only, making register* pressure a 1176 * minor concern. Use r9-r11 as they are volatile, i.e. don't need to be saved 1177 * by the callee, and except for r11 are not implicit parameters to any 1178 * instructions. Ideally, fixup would use r8-r10 and thus avoid implicit 1179 * parameters entirely, but Hyper-V's hypercall ABI uses r8 and testing Hyper-V 1180 * is higher priority than testing non-faulting SYSCALL/SYSRET. 1181 * 1182 * Note, the fixup handler deliberately does not handle #DE, i.e. the vector 1183 * is guaranteed to be non-zero on fault. 1184 * 1185 * REGISTER INPUTS: 1186 * r9 = MAGIC 1187 * r10 = RIP 1188 * r11 = new RIP on fault 1189 * 1190 * REGISTER OUTPUTS: 1191 * r9 = exception vector (non-zero) 1192 * r10 = error code 1193 */ 1194 #define __KVM_ASM_SAFE(insn, fep) \ 1195 "mov $" __stringify(KVM_EXCEPTION_MAGIC) ", %%r9\n\t" \ 1196 "lea 1f(%%rip), %%r10\n\t" \ 1197 "lea 2f(%%rip), %%r11\n\t" \ 1198 fep "1: " insn "\n\t" \ 1199 "xor %%r9, %%r9\n\t" \ 1200 "2:\n\t" \ 1201 "mov %%r9b, %[vector]\n\t" \ 1202 "mov %%r10, %[error_code]\n\t" 1203 1204 #define KVM_ASM_SAFE(insn) __KVM_ASM_SAFE(insn, "") 1205 #define KVM_ASM_SAFE_FEP(insn) __KVM_ASM_SAFE(insn, KVM_FEP) 1206 1207 #define KVM_ASM_SAFE_OUTPUTS(v, ec) [vector] "=qm"(v), [error_code] "=rm"(ec) 1208 #define KVM_ASM_SAFE_CLOBBERS "r9", "r10", "r11" 1209 1210 #define kvm_asm_safe(insn, inputs...) \ 1211 ({ \ 1212 uint64_t ign_error_code; \ 1213 uint8_t vector; \ 1214 \ 1215 asm volatile(KVM_ASM_SAFE(insn) \ 1216 : KVM_ASM_SAFE_OUTPUTS(vector, ign_error_code) \ 1217 : inputs \ 1218 : KVM_ASM_SAFE_CLOBBERS); \ 1219 vector; \ 1220 }) 1221 1222 #define kvm_asm_safe_ec(insn, error_code, inputs...) \ 1223 ({ \ 1224 uint8_t vector; \ 1225 \ 1226 asm volatile(KVM_ASM_SAFE(insn) \ 1227 : KVM_ASM_SAFE_OUTPUTS(vector, error_code) \ 1228 : inputs \ 1229 : KVM_ASM_SAFE_CLOBBERS); \ 1230 vector; \ 1231 }) 1232 1233 #define kvm_asm_safe_fep(insn, inputs...) \ 1234 ({ \ 1235 uint64_t ign_error_code; \ 1236 uint8_t vector; \ 1237 \ 1238 asm volatile(KVM_ASM_SAFE(insn) \ 1239 : KVM_ASM_SAFE_OUTPUTS(vector, ign_error_code) \ 1240 : inputs \ 1241 : KVM_ASM_SAFE_CLOBBERS); \ 1242 vector; \ 1243 }) 1244 1245 #define kvm_asm_safe_ec_fep(insn, error_code, inputs...) \ 1246 ({ \ 1247 uint8_t vector; \ 1248 \ 1249 asm volatile(KVM_ASM_SAFE_FEP(insn) \ 1250 : KVM_ASM_SAFE_OUTPUTS(vector, error_code) \ 1251 : inputs \ 1252 : KVM_ASM_SAFE_CLOBBERS); \ 1253 vector; \ 1254 }) 1255 1256 #define BUILD_READ_U64_SAFE_HELPER(insn, _fep, _FEP) \ 1257 static inline uint8_t insn##_safe ##_fep(uint32_t idx, uint64_t *val) \ 1258 { \ 1259 uint64_t error_code; \ 1260 uint8_t vector; \ 1261 uint32_t a, d; \ 1262 \ 1263 asm volatile(KVM_ASM_SAFE##_FEP(#insn) \ 1264 : "=a"(a), "=d"(d), \ 1265 KVM_ASM_SAFE_OUTPUTS(vector, error_code) \ 1266 : "c"(idx) \ 1267 : KVM_ASM_SAFE_CLOBBERS); \ 1268 \ 1269 *val = (uint64_t)a | ((uint64_t)d << 32); \ 1270 return vector; \ 1271 } 1272 1273 /* 1274 * Generate {insn}_safe() and {insn}_safe_fep() helpers for instructions that 1275 * use ECX as in input index, and EDX:EAX as a 64-bit output. 1276 */ 1277 #define BUILD_READ_U64_SAFE_HELPERS(insn) \ 1278 BUILD_READ_U64_SAFE_HELPER(insn, , ) \ 1279 BUILD_READ_U64_SAFE_HELPER(insn, _fep, _FEP) \ 1280 1281 BUILD_READ_U64_SAFE_HELPERS(rdmsr) 1282 BUILD_READ_U64_SAFE_HELPERS(rdpmc) 1283 BUILD_READ_U64_SAFE_HELPERS(xgetbv) 1284 1285 static inline uint8_t wrmsr_safe(uint32_t msr, uint64_t val) 1286 { 1287 return kvm_asm_safe("wrmsr", "a"(val & -1u), "d"(val >> 32), "c"(msr)); 1288 } 1289 1290 static inline uint8_t xsetbv_safe(uint32_t index, uint64_t value) 1291 { 1292 u32 eax = value; 1293 u32 edx = value >> 32; 1294 1295 return kvm_asm_safe("xsetbv", "a" (eax), "d" (edx), "c" (index)); 1296 } 1297 1298 bool kvm_is_tdp_enabled(void); 1299 1300 static inline bool kvm_is_pmu_enabled(void) 1301 { 1302 return get_kvm_param_bool("enable_pmu"); 1303 } 1304 1305 static inline bool kvm_is_forced_emulation_enabled(void) 1306 { 1307 return !!get_kvm_param_integer("force_emulation_prefix"); 1308 } 1309 1310 uint64_t *__vm_get_page_table_entry(struct kvm_vm *vm, uint64_t vaddr, 1311 int *level); 1312 uint64_t *vm_get_page_table_entry(struct kvm_vm *vm, uint64_t vaddr); 1313 1314 uint64_t kvm_hypercall(uint64_t nr, uint64_t a0, uint64_t a1, uint64_t a2, 1315 uint64_t a3); 1316 uint64_t __xen_hypercall(uint64_t nr, uint64_t a0, void *a1); 1317 void xen_hypercall(uint64_t nr, uint64_t a0, void *a1); 1318 1319 static inline uint64_t __kvm_hypercall_map_gpa_range(uint64_t gpa, 1320 uint64_t size, uint64_t flags) 1321 { 1322 return kvm_hypercall(KVM_HC_MAP_GPA_RANGE, gpa, size >> PAGE_SHIFT, flags, 0); 1323 } 1324 1325 static inline void kvm_hypercall_map_gpa_range(uint64_t gpa, uint64_t size, 1326 uint64_t flags) 1327 { 1328 uint64_t ret = __kvm_hypercall_map_gpa_range(gpa, size, flags); 1329 1330 GUEST_ASSERT(!ret); 1331 } 1332 1333 void __vm_xsave_require_permission(uint64_t xfeature, const char *name); 1334 1335 #define vm_xsave_require_permission(xfeature) \ 1336 __vm_xsave_require_permission(xfeature, #xfeature) 1337 1338 enum pg_level { 1339 PG_LEVEL_NONE, 1340 PG_LEVEL_4K, 1341 PG_LEVEL_2M, 1342 PG_LEVEL_1G, 1343 PG_LEVEL_512G, 1344 PG_LEVEL_NUM 1345 }; 1346 1347 #define PG_LEVEL_SHIFT(_level) ((_level - 1) * 9 + 12) 1348 #define PG_LEVEL_SIZE(_level) (1ull << PG_LEVEL_SHIFT(_level)) 1349 1350 #define PG_SIZE_4K PG_LEVEL_SIZE(PG_LEVEL_4K) 1351 #define PG_SIZE_2M PG_LEVEL_SIZE(PG_LEVEL_2M) 1352 #define PG_SIZE_1G PG_LEVEL_SIZE(PG_LEVEL_1G) 1353 1354 void __virt_pg_map(struct kvm_vm *vm, uint64_t vaddr, uint64_t paddr, int level); 1355 void virt_map_level(struct kvm_vm *vm, uint64_t vaddr, uint64_t paddr, 1356 uint64_t nr_bytes, int level); 1357 1358 /* 1359 * Basic CPU control in CR0 1360 */ 1361 #define X86_CR0_PE (1UL<<0) /* Protection Enable */ 1362 #define X86_CR0_MP (1UL<<1) /* Monitor Coprocessor */ 1363 #define X86_CR0_EM (1UL<<2) /* Emulation */ 1364 #define X86_CR0_TS (1UL<<3) /* Task Switched */ 1365 #define X86_CR0_ET (1UL<<4) /* Extension Type */ 1366 #define X86_CR0_NE (1UL<<5) /* Numeric Error */ 1367 #define X86_CR0_WP (1UL<<16) /* Write Protect */ 1368 #define X86_CR0_AM (1UL<<18) /* Alignment Mask */ 1369 #define X86_CR0_NW (1UL<<29) /* Not Write-through */ 1370 #define X86_CR0_CD (1UL<<30) /* Cache Disable */ 1371 #define X86_CR0_PG (1UL<<31) /* Paging */ 1372 1373 #define PFERR_PRESENT_BIT 0 1374 #define PFERR_WRITE_BIT 1 1375 #define PFERR_USER_BIT 2 1376 #define PFERR_RSVD_BIT 3 1377 #define PFERR_FETCH_BIT 4 1378 #define PFERR_PK_BIT 5 1379 #define PFERR_SGX_BIT 15 1380 #define PFERR_GUEST_FINAL_BIT 32 1381 #define PFERR_GUEST_PAGE_BIT 33 1382 #define PFERR_IMPLICIT_ACCESS_BIT 48 1383 1384 #define PFERR_PRESENT_MASK BIT(PFERR_PRESENT_BIT) 1385 #define PFERR_WRITE_MASK BIT(PFERR_WRITE_BIT) 1386 #define PFERR_USER_MASK BIT(PFERR_USER_BIT) 1387 #define PFERR_RSVD_MASK BIT(PFERR_RSVD_BIT) 1388 #define PFERR_FETCH_MASK BIT(PFERR_FETCH_BIT) 1389 #define PFERR_PK_MASK BIT(PFERR_PK_BIT) 1390 #define PFERR_SGX_MASK BIT(PFERR_SGX_BIT) 1391 #define PFERR_GUEST_FINAL_MASK BIT_ULL(PFERR_GUEST_FINAL_BIT) 1392 #define PFERR_GUEST_PAGE_MASK BIT_ULL(PFERR_GUEST_PAGE_BIT) 1393 #define PFERR_IMPLICIT_ACCESS BIT_ULL(PFERR_IMPLICIT_ACCESS_BIT) 1394 1395 bool sys_clocksource_is_based_on_tsc(void); 1396 1397 #endif /* SELFTEST_KVM_PROCESSOR_H */ 1398