aboutsummaryrefslogtreecommitdiff
diff options
context:
space:
mode:
authorYang Weijiang <weijiang.yang@intel.com>2022-02-15 14:52:54 -0500
committerPaolo Bonzini <pbonzini@redhat.com>2022-05-14 12:32:41 +0200
commit301e90675c3fed6cdc48682021a1ab42bc0e0d76 (patch)
tree6297ef89365a2456042a1d3feeb9092fc711ccf1
parent5a778a5f820fdd907b95e93560637a61f6ea3c71 (diff)
downloadqemu-301e90675c3fed6cdc48682021a1ab42bc0e0d76.zip
qemu-301e90675c3fed6cdc48682021a1ab42bc0e0d76.tar.gz
qemu-301e90675c3fed6cdc48682021a1ab42bc0e0d76.tar.bz2
target/i386: Enable support for XSAVES based features
There're some new features, including Arch LBR, depending on XSAVES/XRSTORS support, the new instructions will save/restore data based on feature bits enabled in XCR0 | XSS. This patch adds the basic support for related CPUID enumeration and meanwhile changes the name from FEAT_XSAVE_COMP_{LO|HI} to FEAT_XSAVE_XCR0_{LO|HI} to differentiate clearly the feature bits in XCR0 and those in XSS. Signed-off-by: Yang Weijiang <weijiang.yang@intel.com> Message-Id: <20220215195258.29149-5-weijiang.yang@intel.com> Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
-rw-r--r--target/i386/cpu.c104
-rw-r--r--target/i386/cpu.h14
2 files changed, 92 insertions, 26 deletions
diff --git a/target/i386/cpu.c b/target/i386/cpu.c
index 856a865..643536d 100644
--- a/target/i386/cpu.c
+++ b/target/i386/cpu.c
@@ -937,6 +937,34 @@ FeatureWordInfo feature_word_info[FEATURE_WORDS] = {
},
.tcg_features = TCG_XSAVE_FEATURES,
},
+ [FEAT_XSAVE_XSS_LO] = {
+ .type = CPUID_FEATURE_WORD,
+ .feat_names = {
+ NULL, NULL, NULL, NULL,
+ NULL, NULL, NULL, NULL,
+ NULL, NULL, NULL, NULL,
+ NULL, NULL, NULL, NULL,
+ NULL, NULL, NULL, NULL,
+ NULL, NULL, NULL, NULL,
+ NULL, NULL, NULL, NULL,
+ NULL, NULL, NULL, NULL,
+ },
+ .cpuid = {
+ .eax = 0xD,
+ .needs_ecx = true,
+ .ecx = 1,
+ .reg = R_ECX,
+ },
+ },
+ [FEAT_XSAVE_XSS_HI] = {
+ .type = CPUID_FEATURE_WORD,
+ .cpuid = {
+ .eax = 0xD,
+ .needs_ecx = true,
+ .ecx = 1,
+ .reg = R_EDX
+ },
+ },
[FEAT_6_EAX] = {
.type = CPUID_FEATURE_WORD,
.feat_names = {
@@ -952,7 +980,7 @@ FeatureWordInfo feature_word_info[FEATURE_WORDS] = {
.cpuid = { .eax = 6, .reg = R_EAX, },
.tcg_features = TCG_6_EAX_FEATURES,
},
- [FEAT_XSAVE_COMP_LO] = {
+ [FEAT_XSAVE_XCR0_LO] = {
.type = CPUID_FEATURE_WORD,
.cpuid = {
.eax = 0xD,
@@ -965,7 +993,7 @@ FeatureWordInfo feature_word_info[FEATURE_WORDS] = {
XSTATE_OPMASK_MASK | XSTATE_ZMM_Hi256_MASK | XSTATE_Hi16_ZMM_MASK |
XSTATE_PKRU_MASK,
},
- [FEAT_XSAVE_COMP_HI] = {
+ [FEAT_XSAVE_XCR0_HI] = {
.type = CPUID_FEATURE_WORD,
.cpuid = {
.eax = 0xD,
@@ -1382,6 +1410,9 @@ static const X86RegisterInfo32 x86_reg_info_32[CPU_NB_REGS32] = {
};
#undef REGISTER
+/* CPUID feature bits available in XSS */
+#define CPUID_XSTATE_XSS_MASK (0)
+
ExtSaveArea x86_ext_save_areas[XSAVE_STATE_AREA_COUNT] = {
[XSTATE_FP_BIT] = {
/* x87 FP state component is always enabled if XSAVE is supported */
@@ -1424,15 +1455,18 @@ ExtSaveArea x86_ext_save_areas[XSAVE_STATE_AREA_COUNT] = {
},
};
-static uint32_t xsave_area_size(uint64_t mask)
+static uint32_t xsave_area_size(uint64_t mask, bool compacted)
{
+ uint64_t ret = x86_ext_save_areas[0].size;
+ const ExtSaveArea *esa;
+ uint32_t offset = 0;
int i;
- uint64_t ret = 0;
- for (i = 0; i < ARRAY_SIZE(x86_ext_save_areas); i++) {
- const ExtSaveArea *esa = &x86_ext_save_areas[i];
+ for (i = 2; i < ARRAY_SIZE(x86_ext_save_areas); i++) {
+ esa = &x86_ext_save_areas[i];
if ((mask >> i) & 1) {
- ret = MAX(ret, esa->offset + esa->size);
+ offset = compacted ? ret : esa->offset;
+ ret = MAX(ret, offset + esa->size);
}
}
return ret;
@@ -1443,10 +1477,10 @@ static inline bool accel_uses_host_cpuid(void)
return kvm_enabled() || hvf_enabled();
}
-static inline uint64_t x86_cpu_xsave_components(X86CPU *cpu)
+static inline uint64_t x86_cpu_xsave_xcr0_components(X86CPU *cpu)
{
- return ((uint64_t)cpu->env.features[FEAT_XSAVE_COMP_HI]) << 32 |
- cpu->env.features[FEAT_XSAVE_COMP_LO];
+ return ((uint64_t)cpu->env.features[FEAT_XSAVE_XCR0_HI]) << 32 |
+ cpu->env.features[FEAT_XSAVE_XCR0_LO];
}
/* Return name of 32-bit register, from a R_* constant */
@@ -1458,6 +1492,12 @@ static const char *get_register_name_32(unsigned int reg)
return x86_reg_info_32[reg].name;
}
+static inline uint64_t x86_cpu_xsave_xss_components(X86CPU *cpu)
+{
+ return ((uint64_t)cpu->env.features[FEAT_XSAVE_XSS_HI]) << 32 |
+ cpu->env.features[FEAT_XSAVE_XSS_LO];
+}
+
/*
* Returns the set of feature flags that are supported and migratable by
* QEMU, for a given FeatureWord.
@@ -4511,8 +4551,8 @@ static const char *x86_cpu_feature_name(FeatureWord w, int bitnr)
/* XSAVE components are automatically enabled by other features,
* so return the original feature name instead
*/
- if (w == FEAT_XSAVE_COMP_LO || w == FEAT_XSAVE_COMP_HI) {
- int comp = (w == FEAT_XSAVE_COMP_HI) ? bitnr + 32 : bitnr;
+ if (w == FEAT_XSAVE_XCR0_LO || w == FEAT_XSAVE_XCR0_HI) {
+ int comp = (w == FEAT_XSAVE_XCR0_HI) ? bitnr + 32 : bitnr;
if (comp < ARRAY_SIZE(x86_ext_save_areas) &&
x86_ext_save_areas[comp].bits) {
@@ -5408,25 +5448,36 @@ void cpu_x86_cpuid(CPUX86State *env, uint32_t index, uint32_t count,
}
if (count == 0) {
- *ecx = xsave_area_size(x86_cpu_xsave_components(cpu));
- *eax = env->features[FEAT_XSAVE_COMP_LO];
- *edx = env->features[FEAT_XSAVE_COMP_HI];
+ *ecx = xsave_area_size(x86_cpu_xsave_xcr0_components(cpu), false);
+ *eax = env->features[FEAT_XSAVE_XCR0_LO];
+ *edx = env->features[FEAT_XSAVE_XCR0_HI];
/*
* The initial value of xcr0 and ebx == 0, On host without kvm
* commit 412a3c41(e.g., CentOS 6), the ebx's value always == 0
* even through guest update xcr0, this will crash some legacy guest
* (e.g., CentOS 6), So set ebx == ecx to workaroud it.
*/
- *ebx = kvm_enabled() ? *ecx : xsave_area_size(env->xcr0);
+ *ebx = kvm_enabled() ? *ecx : xsave_area_size(env->xcr0, false);
} else if (count == 1) {
+ uint64_t xstate = x86_cpu_xsave_xcr0_components(cpu) |
+ x86_cpu_xsave_xss_components(cpu);
+
*eax = env->features[FEAT_XSAVE];
+ *ebx = xsave_area_size(xstate, true);
+ *ecx = env->features[FEAT_XSAVE_XSS_LO];
+ *edx = env->features[FEAT_XSAVE_XSS_HI];
} else if (count < ARRAY_SIZE(x86_ext_save_areas)) {
- if ((x86_cpu_xsave_components(cpu) >> count) & 1) {
- const ExtSaveArea *esa = &x86_ext_save_areas[count];
+ const ExtSaveArea *esa = &x86_ext_save_areas[count];
+
+ if (x86_cpu_xsave_xcr0_components(cpu) & (1ULL << count)) {
*eax = esa->size;
*ebx = esa->offset;
*ecx = esa->ecx &
(ESA_FEATURE_ALIGN64_MASK | ESA_FEATURE_XFD_MASK);
+ } else if (x86_cpu_xsave_xss_components(cpu) & (1ULL << count)) {
+ *eax = esa->size;
+ *ebx = 0;
+ *ecx = 1;
}
}
break;
@@ -5477,8 +5528,8 @@ void cpu_x86_cpuid(CPUX86State *env, uint32_t index, uint32_t count,
} else {
*eax &= env->features[FEAT_SGX_12_1_EAX];
*ebx &= 0; /* ebx reserve */
- *ecx &= env->features[FEAT_XSAVE_COMP_LO];
- *edx &= env->features[FEAT_XSAVE_COMP_HI];
+ *ecx &= env->features[FEAT_XSAVE_XSS_LO];
+ *edx &= env->features[FEAT_XSAVE_XSS_HI];
/* FP and SSE are always allowed regardless of XSAVE/XCR0. */
*ecx |= XSTATE_FP_MASK | XSTATE_SSE_MASK;
@@ -5874,6 +5925,9 @@ static void x86_cpu_reset(DeviceState *dev)
}
for (i = 2; i < ARRAY_SIZE(x86_ext_save_areas); i++) {
const ExtSaveArea *esa = &x86_ext_save_areas[i];
+ if (!((1 << i) & CPUID_XSTATE_XCR0_MASK)) {
+ continue;
+ }
if (env->features[esa->feature] & esa->bits) {
xcr0 |= 1ull << i;
}
@@ -5988,8 +6042,8 @@ static void x86_cpu_enable_xsave_components(X86CPU *cpu)
static bool request_perm;
if (!(env->features[FEAT_1_ECX] & CPUID_EXT_XSAVE)) {
- env->features[FEAT_XSAVE_COMP_LO] = 0;
- env->features[FEAT_XSAVE_COMP_HI] = 0;
+ env->features[FEAT_XSAVE_XCR0_LO] = 0;
+ env->features[FEAT_XSAVE_XCR0_HI] = 0;
return;
}
@@ -6007,8 +6061,10 @@ static void x86_cpu_enable_xsave_components(X86CPU *cpu)
request_perm = true;
}
- env->features[FEAT_XSAVE_COMP_LO] = mask;
- env->features[FEAT_XSAVE_COMP_HI] = mask >> 32;
+ env->features[FEAT_XSAVE_XCR0_LO] = mask & CPUID_XSTATE_XCR0_MASK;
+ env->features[FEAT_XSAVE_XCR0_HI] = mask >> 32;
+ env->features[FEAT_XSAVE_XSS_LO] = mask & CPUID_XSTATE_XSS_MASK;
+ env->features[FEAT_XSAVE_XSS_HI] = mask >> 32;
}
/***** Steps involved on loading and filtering CPUID data
diff --git a/target/i386/cpu.h b/target/i386/cpu.h
index 6730df5..453b80e 100644
--- a/target/i386/cpu.h
+++ b/target/i386/cpu.h
@@ -568,6 +568,14 @@ typedef enum X86Seg {
#define ESA_FEATURE_XFD_MASK (1U << ESA_FEATURE_XFD_BIT)
+/* CPUID feature bits available in XCR0 */
+#define CPUID_XSTATE_XCR0_MASK (XSTATE_FP_MASK | XSTATE_SSE_MASK | \
+ XSTATE_YMM_MASK | XSTATE_BNDREGS_MASK | \
+ XSTATE_BNDCSR_MASK | XSTATE_OPMASK_MASK | \
+ XSTATE_ZMM_Hi256_MASK | \
+ XSTATE_Hi16_ZMM_MASK | XSTATE_PKRU_MASK | \
+ XSTATE_XTILE_CFG_MASK | XSTATE_XTILE_DATA_MASK)
+
/* CPUID feature words */
typedef enum FeatureWord {
FEAT_1_EDX, /* CPUID[1].EDX */
@@ -586,8 +594,8 @@ typedef enum FeatureWord {
FEAT_SVM, /* CPUID[8000_000A].EDX */
FEAT_XSAVE, /* CPUID[EAX=0xd,ECX=1].EAX */
FEAT_6_EAX, /* CPUID[6].EAX */
- FEAT_XSAVE_COMP_LO, /* CPUID[EAX=0xd,ECX=0].EAX */
- FEAT_XSAVE_COMP_HI, /* CPUID[EAX=0xd,ECX=0].EDX */
+ FEAT_XSAVE_XCR0_LO, /* CPUID[EAX=0xd,ECX=0].EAX */
+ FEAT_XSAVE_XCR0_HI, /* CPUID[EAX=0xd,ECX=0].EDX */
FEAT_ARCH_CAPABILITIES,
FEAT_CORE_CAPABILITY,
FEAT_PERF_CAPABILITIES,
@@ -604,6 +612,8 @@ typedef enum FeatureWord {
FEAT_SGX_12_0_EAX, /* CPUID[EAX=0x12,ECX=0].EAX (SGX) */
FEAT_SGX_12_0_EBX, /* CPUID[EAX=0x12,ECX=0].EBX (SGX MISCSELECT[31:0]) */
FEAT_SGX_12_1_EAX, /* CPUID[EAX=0x12,ECX=1].EAX (SGX ATTRIBUTES[31:0]) */
+ FEAT_XSAVE_XSS_LO, /* CPUID[EAX=0xd,ECX=1].ECX */
+ FEAT_XSAVE_XSS_HI, /* CPUID[EAX=0xd,ECX=1].EDX */
FEATURE_WORDS,
} FeatureWord;