Support for new CPU model SapphireRapids
cover Support for new CPU model SapphireRapids
This series aims to add a new CPU model SapphireRapids, and tries to address the problem stated in
https://lore.kernel.org/all/[email protected]/T/#mcf67dbd1ad37c65d7988c36a2b267be9afd2fb30,
so that named CPU model can define its own AMX values, and QEMU won't pass the wrong AMX values to KVM in future platforms if they have different values supported.
The original patch is
https://lore.kernel.org/all/[email protected]/T/#u.
Wang, Lei (6):
i386: Introduce FeatureWordInfo for AMX CPUID leaf 0x1D and 0x1E
i386: Remove unused parameter "uint32_t bit" in
feature_word_description()
i386: Introduce new struct "MultiBitFeatureInfo" for multi-bit
features
i386: Mask and report unavailable multi-bit feature values
i386: Initialize AMX CPUID leaves with corresponding env->features[]
leaves
i386: Add new CPU model SapphireRapids
target/i386/cpu-internal.h | 11 ++ target/i386/cpu.c | 314 ++++++++++++++++++++++++++++++++++--- target/i386/cpu.h | 18 +++ 3 files changed, 323 insertions(+), 20 deletions(-)
Signature · 1 line
2.34.1
PATCH 1/6 i386: Introduce FeatureWordInfo for AMX CPUID leaf 0x1D and 0x1E
CPUID leaf 0x1D and 0x1E enumerate tile and TMUL information for AMX.
Introduce FeatureWord FEAT_1D_1_EAX, FEAT_1D_1_EBX, FEAT_1D_1_ECX and FEAT_1E_0_EBX. Thus these features of AMX can be expanded when "-cpu host/max" and can be configured in named CPU model.
--- target/i386/cpu.c | 55 +++++++++++++++++++++++++++++++++++++++++++++++ target/i386/cpu.h | 12 +++++++++++ 2 files changed, 67 insertions(+)
target/i386/cpu.c +55 -0
@@ -1002,6 +1002,45 @@ FeatureWordInfo feature_word_info[FEATURE_WORDS] = { }, .tcg_features = ~0U, }, + [FEAT_1D_1_EAX] = { + .type = CPUID_FEATURE_WORD, + .cpuid = { + .eax = 0x1D, + .needs_ecx = true, .ecx = 1, + .reg = R_EAX, + }, + .migratable_flags =CPUID_AMX_PALETTE_1_TOTAL_TILE_BYTES_MASK | + CPUID_AMX_PALETTE_1_BYTES_PER_TILE_MASK, + }, + [FEAT_1D_1_EBX] = { + .type = CPUID_FEATURE_WORD, + .cpuid = { + .eax = 0x1D, + .needs_ecx = true, .ecx = 1, + .reg = R_EBX, + }, + .migratable_flags =CPUID_AMX_PALETTE_1_BYTES_PER_ROW_MASK | + CPUID_AMX_PALETTE_1_MAX_NAMES_MASK, + }, + [FEAT_1D_1_ECX] = { + .type = CPUID_FEATURE_WORD, + .cpuid = { + .eax = 0x1D, + .needs_ecx = true, .ecx = 1, + .reg = R_ECX, + }, + .migratable_flags = CPUID_AMX_PALETTE_1_MAX_ROWS_MASK, + }, + [FEAT_1E_0_EBX] = { + .type = CPUID_FEATURE_WORD, + .cpuid = { + .eax = 0x1E, + .needs_ecx = true, .ecx = 0, + .reg = R_EBX, + }, + .migratable_flags =CPUID_AMX_TMUL_MAX_K_MASK | + CPUID_AMX_TMUL_MAX_N_MASK, + },/*Below are MSR exposed features*/ [FEAT_ARCH_CAPABILITIES] = { .type = MSR_FEATURE_WORD,
@@ -1371,6 +1410,22 @@ static FeatureDep feature_dependencies[] = { .from = { FEAT_7_0_EBX, CPUID_7_0_EBX_INTEL_PT }, .to = { FEAT_14_0_ECX, ~0ull }, }, + { + .from = { FEAT_7_0_EDX, CPUID_7_0_EDX_AMX_TILE }, + .to = { FEAT_1D_1_EAX, ~0ull }, + }, + { + .from = { FEAT_7_0_EDX, CPUID_7_0_EDX_AMX_TILE }, + .to = { FEAT_1D_1_EBX, ~0ull }, + }, + { + .from = { FEAT_7_0_EDX, CPUID_7_0_EDX_AMX_TILE }, + .to = { FEAT_1D_1_ECX, ~0ull }, + }, + { + .from = { FEAT_7_0_EDX, CPUID_7_0_EDX_AMX_TILE }, + .to = { FEAT_1E_0_EBX, ~0ull }, + }, { .from = { FEAT_8000_0001_EDX, CPUID_EXT2_RDTSCP }, .to = { FEAT_VMX_SECONDARY_CTLS, VMX_SECONDARY_EXEC_RDTSCP },
target/i386/cpu.h +12 -0
@@ -583,6 +583,14 @@ typedef enum X86Seg { XSTATE_Hi16_ZMM_MASK | XSTATE_PKRU_MASK | \ XSTATE_XTILE_CFG_MASK | XSTATE_XTILE_DATA_MASK) +#define CPUID_AMX_PALETTE_1_TOTAL_TILE_BYTES_MASK 0xffffU +#define CPUID_AMX_PALETTE_1_BYTES_PER_TILE_MASK (0xffffU << 16) +#define CPUID_AMX_PALETTE_1_BYTES_PER_ROW_MASK 0xffffU +#define CPUID_AMX_PALETTE_1_MAX_NAMES_MASK (0xffffU << 16) +#define CPUID_AMX_PALETTE_1_MAX_ROWS_MASK 0xffffU +#define CPUID_AMX_TMUL_MAX_K_MASK 0xffU +#define CPUID_AMX_TMUL_MAX_N_MASK (0xffffU << 8) +/* CPUID feature words */ typedef enum FeatureWord { FEAT_1_EDX, /* CPUID[1].EDX */
@@ -603,6 +611,10 @@ typedef enum FeatureWord {FEAT_6_EAX ,/* CPUID[6].EAX */ FEAT_XSAVE_XCR0_LO ,/* CPUID[EAX=0xd,ECX=0].EAX */ FEAT_XSAVE_XCR0_HI ,/* CPUID[EAX=0xd,ECX=0].EDX */ +FEAT_1D_1_EAX ,/* CPUID[EAX=0x1d,ECX=1].EAX */ +FEAT_1D_1_EBX ,/* CPUID[EAX=0x1d,ECX=1].EBX */ +FEAT_1D_1_ECX ,/* CPUID[EAX=0x1d,ECX=1].ECX */ +FEAT_1E_0_EBX ,/* CPUID[EAX=0x1e,ECX=0].EBX */ FEAT_ARCH_CAPABILITIES ,FEAT_CORE_CAPABILITY , FEAT_PERF_CAPABILITIES,
Signature · 1 line
2.34.1
PATCH 2/6 i386: Remove unused parameter "uint32_t bit" in feature_word_description()
In reply to Wang, LeiParameter "uint32_t bit" is not used in function feature_word_description(), so remove it.
--- target/i386/cpu.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)
target/i386/cpu.c +2 -2
@@ -4290,7 +4290,7 @@ static const TypeInfo max_x86_cpu_type_info = { .class_init = max_x86_cpu_class_init, }; -static char *feature_word_description (FeatureWordInfo *f,uint32_t bit) +static char *feature_word_description (FeatureWordInfo *f) {assert (f->type ==CPUID_FEATURE_WORD || f->type ==MSR_FEATURE_WORD );
@@ -4329,6 +4329,7 @@ static void mark_unavailable_features(X86CPU *cpu, FeatureWord w, uint64_t mask,CPUX86State *env = &cpu->env;FeatureWordInfo *f = &feature_word_info[w];int i; + g_autofree char *feat_word_str = feature_word_description(f);if (!cpu->force_features) { env->features[w] &= ~mask;
@@ -4341,7 +4342,6 @@ static void mark_unavailable_features(X86CPU *cpu, FeatureWord w, uint64_t mask,for (i =0 ; i <64 ; ++i) {if ((1ULL << i) & mask) { - g_autofree char *feat_word_str = feature_word_description(f, i); warn_report("%s: %s%s%s [bit %d]", verbose_prefix, feat_word_str,
Signature · 1 line
2.34.1
PATCH 3/6 i386: Introduce new struct "MultiBitFeatureInfo" for multi-bit features
In reply to Wang, LeiSome features use multiple CPUID bits to form a value to be used, e.g., CPUID(0x1E,0):EBX[23:08] is regarded as the tmul_maxn value for AMX. Introduce a new struct "MultiBitFeatureInfo" to hold the information for those features and create a corresponding member in struct FeatureWordInfo, so that the infomation can be assigned for each item in feature_word_info array and used in the future.
--- target/i386/cpu-internal.h | 9 +++++++ target/i386/cpu.c | 54 ++++++++++++++++++++++++++++++++++++++ 2 files changed, 63 insertions(+)
target/i386/cpu-internal.h +9 -0
@@ -25,6 +25,13 @@ typedef enum FeatureWordType {MSR_FEATURE_WORD , } FeatureWordType; +typedef struct MultiBitFeatureInfo { +const char *feat_name; +uint64_t mask; +unsigned high_bit_position; +unsigned low_bit_position; +}MultiBitFeatureInfo ; + typedef struct FeatureWordInfo { FeatureWordType type; /* feature flags names are taken from "Intel Processor Identification and
@@ -51,6 +58,8 @@ typedef struct FeatureWordInfo {uint64_t migratable_flags;/* Feature flags known to be migratable */ /* Features that shouldn't be auto-enabled by "-cpu host" */ uint64_t no_autoenable_flags; +unsigned num_multi_bit_features; +MultiBitFeatureInfo *multi_bit_features; } FeatureWordInfo;extern FeatureWordInfo feature_word_info[];
target/i386/cpu.c +54 -0
78 lines not on this page · raw message
Signature · 1 line
2.34.1
PATCH 4/6 i386: Mask and report unavailable multi-bit feature values
In reply to Wang, LeiSome feature words, e.g., feature words in AMX-related CPUID leaf 0x1D and 0x1E are not bit-wise but multiple bits represents one value. Handle this situation when the values specified are not the same as which are reported by KVM. The handling includes:
- The responsibility of masking bits and giving warnings are delegated to the feature enabler. A framwork is also provided to enable this.
- To simplify the initialization, a default function is provided if the the function is not specified.
The reason why delegating this responsibility rather than just marking them as zeros when they are not same is because different multi-bit features may have different logic, which is case by case, for example:
1. CPUID.0x14_0x1:EBX[15:0]. Even though it's multi-bits field, it's a bitmap and each bit represents a separate capability.
2. CPUID.0x14_0x1:EAX[2:0] represents the number of configurable Address Ranges. 3 bits as a whole to represent a integer value. It means the maximum capability of HW. If KVM reports M, then M to 0 is legal value to configure (because KVM can emulate each value correctly).
3. CPUID.0x1D_0x1:EAX[31:16] represents palette 1 bytes_per_tile. 16 bits as a whole represent an integer value. It's not like case 2 and SW needs to configure the same value as reported. Because it's not possible for SW to configure to a different value and KVM cannot emulate it.
So marking them blindly as zeros is incorrect, and delegating this responsibility can let each multi-bit feature have its own way to mask bits.
--- target/i386/cpu-internal.h | 2 ++ target/i386/cpu.c | 39 ++++++++++++++++++++++++++++++++++---- target/i386/cpu.h | 2 ++ 3 files changed, 39 insertions(+), 4 deletions(-)
target/i386/cpu-internal.h +2 -0
@@ -30,6 +30,8 @@ typedef struct MultiBitFeatureInfo {uint64_t mask;unsigned high_bit_position;unsigned low_bit_position; +void (*mark_unavailable_bits)(X86CPU *cpu,FeatureWord w,int index, +const char *verbose_prefix); } MultiBitFeatureInfo; typedef struct FeatureWordInfo {
target/i386/cpu.c +35 -4
@@ -4377,6 +4377,26 @@ static bool x86_cpu_have_filtered_features(X86CPU *cpu)return false; } +void mark_unavailable_bits (X86CPU *cpu,FeatureWord w,int index, +const char *verbose_prefix) +{ +FeatureWordInfo *f = &feature_word_info[w]; + g_autofree char *feat_word_str = feature_word_description(f); +uint64_t host_feat =x86_cpu_get_supported_feature_word (w, false); +MultiBitFeatureInfo mf = f->multi_bit_features[index]; + +if ((cpu->env.features[w] ^ host_feat) & mf.mask) { +if (!cpu->force_features) { + cpu->env.features[w] &= ~mf.mask; + } + cpu->filtered_features[w] |= mf.mask; +if (verbose_prefix) +warn_report ("%s: %s.%s [%u:%u]" , verbose_prefix, feat_word_str, + mf.feat_name, mf.high_bit_position, + mf.low_bit_position); + } +} +static void mark_unavailable_features (X86CPU *cpu,FeatureWord w,uint64_t mask,const char *verbose_prefix) {
@@ -6424,10 +6444,21 @@ static void x86_cpu_filter_features(X86CPU *cpu, bool verbose) }for (w =0 ; w <FEATURE_WORDS ; w++) { -uint64_t host_feat = -x86_cpu_get_supported_feature_word (w, false); -uint64_t requested_features = env->features[w]; -uint64_t unavailable_features = requested_features & ~host_feat; +uint64_t host_feat =x86_cpu_get_supported_feature_word (w, false); +FeatureWordInfo f = feature_word_info[w]; +uint64_t unavailable_features = env->features[w] & ~host_feat; +int i; + +for (i =0 ; i < f.num_multi_bit_features; i++) { +MultiBitFeatureInfo mf = f.multi_bit_features[i]; +if (!mf.mark_unavailable_bits) { + mf.mark_unavailable_bits = mark_unavailable_bits; + } + mf.mark_unavailable_bits (cpu, w, i, prefix); + + unavailable_features &= ~mf.mask; + } +mark_unavailable_features (cpu, w, unavailable_features, prefix); }
target/i386/cpu.h +2 -0
@@ -2103,6 +2103,8 @@ void cpu_x86_cpuid(CPUX86State *env, uint32_t index, uint32_t count,void cpu_clear_apic_feature (CPUX86State *env);void host_cpuid (uint32_t function,uint32_t count,uint32_t *eax,uint32_t *ebx,uint32_t *ecx,uint32_t *edx); +void mark_unavailable_bits (X86CPU *cpu,FeatureWord w,int index, +const char *verbose_prefix);/* helper.c */ void x86_cpu_set_a20 (X86CPU *cpu,int a20_state);
Signature · 1 line
2.34.1
PATCH 5/6 i386: Initialize AMX CPUID leaves with corresponding env->features[] leaves
In reply to Wang, LeiThe AMX-related CPUID value, i.e., CPUID(0x1D,1):EAX, CPUID(0x1D,1):EBX, CPUID(0x1D,1):ECX and CPUID(0x1E,0):EBX are hard-coded to Sapphire Rapids without considering future platforms.
Replace these hard-coded values with env->features[], so QEMU can pass the right value to KVM.
--- target/i386/cpu.c | 27 +++++++++++++-------------- 1 file changed, 13 insertions(+), 14 deletions(-)
target/i386/cpu.c +13 -14
@@ -576,16 +576,16 @@ static CPUCacheInfo legacy_l3_cache = {#define INTEL_PT_PSB_BITMAP (0x003f << 16)/* Support 2K,4K,8K,16K,32K,64K */ /* CPUID Leaf 0x1D constants: */ -#define INTEL_AMX_TILE_MAX_SUBLEAF 0x1 -#define INTEL_AMX_TOTAL_TILE_BYTES 0x2000 -#define INTEL_AMX_BYTES_PER_TILE 0x400 -#define INTEL_AMX_BYTES_PER_ROW 0x40 -#define INTEL_AMX_TILE_MAX_NAMES 0x8 -#define INTEL_AMX_TILE_MAX_ROWS 0x10 +#define INTEL_SPR_AMX_TILE_MAX_SUBLEAF 0x1 +#define INTEL_SPR_AMX_TOTAL_TILE_BYTES 0x2000 +#define INTEL_SPR_AMX_BYTES_PER_TILE 0x400 +#define INTEL_SPR_AMX_BYTES_PER_ROW 0x40 +#define INTEL_SPR_AMX_TILE_MAX_NAMES 0x8 +#define INTEL_SPR_AMX_TILE_MAX_ROWS 0x10/* CPUID Leaf 0x1E constants: */ -#define INTEL_AMX_TMUL_MAX_K 0x10 -#define INTEL_AMX_TMUL_MAX_N 0x40 +#define INTEL_SPR_AMX_TMUL_MAX_K 0x10 +#define INTEL_SPR_AMX_TMUL_MAX_N 0x40void x86_cpu_vendor_words2str (char *dst,uint32_t vendor1,uint32_t vendor2,uint32_t vendor3)
@@ -5763,12 +5763,11 @@ void cpu_x86_cpuid(CPUX86State *env, uint32_t index, uint32_t count,if (count ==0 ) {/* Highest numbered palette subleaf */ - *eax =INTEL_AMX_TILE_MAX_SUBLEAF ; + *eax =INTEL_SPR_AMX_TILE_MAX_SUBLEAF ; }else if (count ==1 ) { - *eax =INTEL_AMX_TOTAL_TILE_BYTES | - (INTEL_AMX_BYTES_PER_TILE <<16 ); - *ebx =INTEL_AMX_BYTES_PER_ROW | (INTEL_AMX_TILE_MAX_NAMES <<16 ); - *ecx =INTEL_AMX_TILE_MAX_ROWS ; + *eax = env->features[FEAT_1D_1_EAX ]; + *ebx = env->features[FEAT_1D_1_EBX ]; + *ecx = env->features[FEAT_1D_1_ECX ]; }break ; }
@@ -5784,7 +5783,7 @@ void cpu_x86_cpuid(CPUX86State *env, uint32_t index, uint32_t count,if (count ==0 ) {/* Highest numbered palette subleaf */ - *ebx =INTEL_AMX_TMUL_MAX_K | (INTEL_AMX_TMUL_MAX_N <<8 ); + *ebx = env->features[FEAT_1E_0_EBX ]; }break ; }
Signature · 1 line
2.34.1
PATCH 6/6 i386: Add new CPU model SapphireRapids
In reply to Wang, LeiThe new CPU model mostly inherits features from Icelake-Server, while adding new features:
- AMX (Advance Matrix eXtensions)
- Bus Lock Debug Exception
and new instructions:
- AVX VNNI (Vector Neural Network Instruction):
- VPDPBUS: Multiply and Add Unsigned and Signed Bytes
- VPDPBUSDS: Multiply and Add Unsigned and Signed Bytes with Saturation
- VPDPWSSD: Multiply and Add Signed Word Integers
- VPDPWSSDS: Multiply and Add Signed Integers with Saturation
- FP16: Replicates existing AVX512 computational SP (FP32) instructions using FP16 instead of FP32 for ~2X performance gain
- SERIALIZE: Provide software with a simple way to force the processor to complete all modifications, faster, allowed in all privilege levels and not causing an unconditional VM exit
- TSX Suspend Load Address Tracking: Allows programmers to choose which memory accesses do not need to be tracked in the TSX read set
- AVX512_BF16: Vector Neural Network Instructions supporting BFLOAT16 inputs and conversion instructions from IEEE single precision
Features may be added in future versions:
- CET (virtualization support hasn't been merged)
Instructions may be added in future versions:
- fast zero-length MOVSB (KVM doesn't support yet)
- fast short STOSB (KVM doesn't support yet)
- fast short CMPSB, SCASB (KVM doesn't support yet)
--- target/i386/cpu.c | 135 ++++++++++++++++++++++++++++++++++++++++++++++ target/i386/cpu.h | 4 ++ 2 files changed, 139 insertions(+)
target/i386/cpu.c +135 -0
@@ -3576,6 +3576,141 @@ static const X86CPUDefinition builtin_x86_defs[] = { {/* end of list */ } } }, + { + .name = "SapphireRapids", + .level = 0x20, + .vendor = CPUID_VENDOR_INTEL, + .family = 6, + .model = 143, + .stepping = 4, +/* +* please keep the ascending order so that we can have a clear view of +* bit position of each feature. +*/ + .features[FEAT_1_EDX ] = +CPUID_FP87 |CPUID_VME |CPUID_DE |CPUID_PSE |CPUID_TSC | +CPUID_MSR |CPUID_PAE |CPUID_MCE |CPUID_CX8 |CPUID_APIC | +CPUID_SEP |CPUID_MTRR |CPUID_PGE |CPUID_MCA |CPUID_CMOV | +CPUID_PAT |CPUID_PSE36 |CPUID_CLFLUSH |CPUID_MMX |CPUID_FXSR | + CPUID_SSE | CPUID_SSE2, + .features[FEAT_1_ECX ] = +CPUID_EXT_SSE3 |CPUID_EXT_PCLMULQDQ |CPUID_EXT_SSSE3 | +CPUID_EXT_FMA |CPUID_EXT_CX16 |CPUID_EXT_PCID |CPUID_EXT_SSE41 | +CPUID_EXT_SSE42 |CPUID_EXT_X2APIC |CPUID_EXT_MOVBE | +CPUID_EXT_POPCNT |CPUID_EXT_TSC_DEADLINE_TIMER |CPUID_EXT_AES | + CPUID_EXT_XSAVE | CPUID_EXT_AVX | CPUID_EXT_F16C | CPUID_EXT_RDRAND, + .features[FEAT_8000_0001_EDX ] = +CPUID_EXT2_SYSCALL |CPUID_EXT2_NX |CPUID_EXT2_PDPE1GB | + CPUID_EXT2_RDTSCP | CPUID_EXT2_LM, + .features[FEAT_8000_0001_ECX ] = + CPUID_EXT3_LAHF_LM | CPUID_EXT3_ABM | CPUID_EXT3_3DNOWPREFETCH, + .features[FEAT_8000_0008_EBX ] = + CPUID_8000_0008_EBX_WBNOINVD, + .features[FEAT_7_0_EBX ] = +CPUID_7_0_EBX_FSGSBASE |CPUID_7_0_EBX_BMI1 |CPUID_7_0_EBX_HLE | +CPUID_7_0_EBX_AVX2 |CPUID_7_0_EBX_SMEP |CPUID_7_0_EBX_BMI2 | +CPUID_7_0_EBX_ERMS |CPUID_7_0_EBX_INVPCID |CPUID_7_0_EBX_RTM | +CPUID_7_0_EBX_AVX512F |CPUID_7_0_EBX_AVX512DQ | +CPUID_7_0_EBX_RDSEED |CPUID_7_0_EBX_ADX |CPUID_7_0_EBX_SMAP | +CPUID_7_0_EBX_AVX512IFMA |CPUID_7_0_EBX_CLFLUSHOPT | +CPUID_7_0_EBX_CLWB |CPUID_7_0_EBX_AVX512CD |CPUID_7_0_EBX_SHA_NI | + CPUID_7_0_EBX_AVX512BW | CPUID_7_0_EBX_AVX512VL, + .features[FEAT_7_0_ECX ] = +CPUID_7_0_ECX_AVX512_VBMI |CPUID_7_0_ECX_UMIP |CPUID_7_0_ECX_PKU | +CPUID_7_0_ECX_AVX512_VBMI2 |CPUID_7_0_ECX_GFNI | +CPUID_7_0_ECX_VAES |CPUID_7_0_ECX_VPCLMULQDQ | +CPUID_7_0_ECX_AVX512VNNI |CPUID_7_0_ECX_AVX512BITALG | +CPUID_7_0_ECX_AVX512_VPOPCNTDQ |CPUID_7_0_ECX_LA57 | + CPUID_7_0_ECX_RDPID | CPUID_7_0_ECX_BUS_LOCK_DETECT, + .features[FEAT_7_0_EDX ] = +CPUID_7_0_EDX_FSRM |CPUID_7_0_EDX_SERIALIZE | +CPUID_7_0_EDX_TSX_LDTRK |CPUID_7_0_EDX_AMX_BF16 | +CPUID_7_0_EDX_AVX512_FP16 |CPUID_7_0_EDX_AMX_TILE | +CPUID_7_0_EDX_AMX_INT8 |CPUID_7_0_EDX_SPEC_CTRL | + CPUID_7_0_EDX_ARCH_CAPABILITIES | CPUID_7_0_EDX_SPEC_CTRL_SSBD, + .features[FEAT_ARCH_CAPABILITIES ] = +MSR_ARCH_CAP_RDCL_NO |MSR_ARCH_CAP_IBRS_ALL | +MSR_ARCH_CAP_SKIP_L1DFL_VMENTRY |MSR_ARCH_CAP_MDS_NO | + MSR_ARCH_CAP_PSCHANGE_MC_NO | MSR_ARCH_CAP_TAA_NO, + .features[FEAT_XSAVE ] = +CPUID_XSAVE_XSAVEOPT |CPUID_XSAVE_XSAVEC | + CPUID_XSAVE_XGETBV1 | CPUID_XSAVE_XSAVES | CPUID_D_1_EAX_XFD, + .features[FEAT_6_EAX ] = + CPUID_6_EAX_ARAT, + .features[FEAT_7_1_EAX ] = + CPUID_7_1_EAX_AVX_VNNI | CPUID_7_1_EAX_AVX512_BF16, + .features[FEAT_1D_1_EAX ] =INTEL_SPR_AMX_TOTAL_TILE_BYTES | + (INTEL_SPR_AMX_BYTES_PER_TILE << 16), + .features[FEAT_1D_1_EBX ] =INTEL_SPR_AMX_BYTES_PER_ROW | + (INTEL_SPR_AMX_TILE_MAX_NAMES << 16), + .features[FEAT_1D_1_ECX] = INTEL_SPR_AMX_TILE_MAX_ROWS, + .features[FEAT_1E_0_EBX ] =INTEL_SPR_AMX_TMUL_MAX_K | + (INTEL_SPR_AMX_TMUL_MAX_N << 8), + .features[FEAT_VMX_BASIC ] = + MSR_VMX_BASIC_INS_OUTS | MSR_VMX_BASIC_TRUE_CTLS, + .features[FEAT_VMX_ENTRY_CTLS ] = +VMX_VM_ENTRY_LOAD_DEBUG_CONTROLS |VMX_VM_ENTRY_IA32E_MODE | +VMX_VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL | + VMX_VM_ENTRY_LOAD_IA32_PAT | VMX_VM_ENTRY_LOAD_IA32_EFER, + .features[FEAT_VMX_EPT_VPID_CAPS ] = +MSR_VMX_EPT_EXECONLY | +MSR_VMX_EPT_PAGE_WALK_LENGTH_4 |MSR_VMX_EPT_PAGE_WALK_LENGTH_5 | +MSR_VMX_EPT_WB |MSR_VMX_EPT_2MB |MSR_VMX_EPT_1GB | +MSR_VMX_EPT_INVEPT |MSR_VMX_EPT_AD_BITS | +MSR_VMX_EPT_INVEPT_SINGLE_CONTEXT |MSR_VMX_EPT_INVEPT_ALL_CONTEXT | +MSR_VMX_EPT_INVVPID |MSR_VMX_EPT_INVVPID_SINGLE_ADDR | +MSR_VMX_EPT_INVVPID_SINGLE_CONTEXT | +MSR_VMX_EPT_INVVPID_ALL_CONTEXT | + MSR_VMX_EPT_INVVPID_SINGLE_CONTEXT_NOGLOBALS, + .features[FEAT_VMX_EXIT_CTLS ] = +VMX_VM_EXIT_SAVE_DEBUG_CONTROLS | +VMX_VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL | +VMX_VM_EXIT_ACK_INTR_ON_EXIT |VMX_VM_EXIT_SAVE_IA32_PAT | +VMX_VM_EXIT_LOAD_IA32_PAT |VMX_VM_EXIT_SAVE_IA32_EFER | + VMX_VM_EXIT_LOAD_IA32_EFER | VMX_VM_EXIT_SAVE_VMX_PREEMPTION_TIMER, + .features[FEAT_VMX_MISC ] = +MSR_VMX_MISC_STORE_LMA |MSR_VMX_MISC_ACTIVITY_HLT | + MSR_VMX_MISC_VMWRITE_VMEXIT, + .features[FEAT_VMX_PINBASED_CTLS ] = +VMX_PIN_BASED_EXT_INTR_MASK |VMX_PIN_BASED_NMI_EXITING | +VMX_PIN_BASED_VIRTUAL_NMIS |VMX_PIN_BASED_VMX_PREEMPTION_TIMER | + VMX_PIN_BASED_POSTED_INTR, + .features[FEAT_VMX_PROCBASED_CTLS ] = +VMX_CPU_BASED_VIRTUAL_INTR_PENDING | +VMX_CPU_BASED_USE_TSC_OFFSETING |VMX_CPU_BASED_HLT_EXITING | +VMX_CPU_BASED_INVLPG_EXITING |VMX_CPU_BASED_MWAIT_EXITING | +VMX_CPU_BASED_RDPMC_EXITING |VMX_CPU_BASED_RDTSC_EXITING | +VMX_CPU_BASED_CR3_LOAD_EXITING |VMX_CPU_BASED_CR3_STORE_EXITING | +VMX_CPU_BASED_CR8_LOAD_EXITING |VMX_CPU_BASED_CR8_STORE_EXITING | +VMX_CPU_BASED_TPR_SHADOW |VMX_CPU_BASED_VIRTUAL_NMI_PENDING | +VMX_CPU_BASED_MOV_DR_EXITING |VMX_CPU_BASED_UNCOND_IO_EXITING | +VMX_CPU_BASED_USE_IO_BITMAPS |VMX_CPU_BASED_MONITOR_TRAP_FLAG | +VMX_CPU_BASED_USE_MSR_BITMAPS |VMX_CPU_BASED_MONITOR_EXITING | +VMX_CPU_BASED_PAUSE_EXITING | + VMX_CPU_BASED_ACTIVATE_SECONDARY_CONTROLS, + .features[FEAT_VMX_SECONDARY_CTLS ] = +VMX_SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES | +VMX_SECONDARY_EXEC_ENABLE_EPT |VMX_SECONDARY_EXEC_DESC | +VMX_SECONDARY_EXEC_RDTSCP | +VMX_SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE | +VMX_SECONDARY_EXEC_ENABLE_VPID |VMX_SECONDARY_EXEC_WBINVD_EXITING | +VMX_SECONDARY_EXEC_UNRESTRICTED_GUEST | +VMX_SECONDARY_EXEC_APIC_REGISTER_VIRT | +VMX_SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY | +VMX_SECONDARY_EXEC_RDRAND_EXITING | +VMX_SECONDARY_EXEC_ENABLE_INVPCID | +VMX_SECONDARY_EXEC_ENABLE_VMFUNC |VMX_SECONDARY_EXEC_SHADOW_VMCS | +VMX_SECONDARY_EXEC_RDSEED_EXITING |VMX_SECONDARY_EXEC_ENABLE_PML | + VMX_SECONDARY_EXEC_XSAVES, + .features[FEAT_VMX_VMFUNC ] = + MSR_VMX_VMFUNC_EPT_SWITCHING, + .xlevel = 0x80000008, + .model_id = "Intel Xeon Processor (SapphireRapids)", + .versions = (X86CPUVersionDefinition []) { + { .version =1 }, + {/* end of list */ }, + }, + }, { .name = "Denverton", .level = 21,
target/i386/cpu.h +4 -0
@@ -891,10 +891,14 @@ uint64_t x86_cpu_get_supported_feature_word(FeatureWord w,#define CPUID_7_0_EDX_TSX_LDTRK (1U << 16)/* Architectural LBRs */ #define CPUID_7_0_EDX_ARCH_LBR (1U << 19) +/* AMX_BF16 instruction */ +#define CPUID_7_0_EDX_AMX_BF16 (1U << 22)/* AVX512_FP16 instruction */ #define CPUID_7_0_EDX_AVX512_FP16 (1U << 23)/* AMX tile (two-dimensional register) */ #define CPUID_7_0_EDX_AMX_TILE (1U << 24) +/* AMX_INT8 instruction */ +#define CPUID_7_0_EDX_AMX_INT8 (1U << 25)/* Speculation Control */ #define CPUID_7_0_EDX_SPEC_CTRL (1U << 26)/* Single Thread Indirect Branch Predictors */
Signature · 1 line
2.34.1