x86/cpufeatures: Add macros for Intel's new fast rep string features
PATCH v3 1/2 x86/cpufeatures: Add macros for Intel's new fast rep string features
KVM_GET_SUPPORTED_CPUID should reflect these host CPUID bits. The bits are already cached in word 12. Give the bits X86_FEATURE names, so that they can be easily referenced. Hide these bits from /proc/cpuinfo, since the host kernel makes no use of them at present.
--- arch/x86/include/asm/cpufeatures.h | 3 +++ 1 file changed, 3 insertions(+)
arch/x86/include/asm/cpufeatures.h +3 -0
@@ -308,6 +308,9 @@/* Intel-defined CPU features, CPUID level 0x00000007:1 (EAX), word 12 */ #define X86_FEATURE_AVX_VNNI (12*32+ 4)/* AVX VNNI instructions */ #define X86_FEATURE_AVX512_BF16 (12*32+ 5)/* AVX512 BFLOAT16 instructions */ +#define X86_FEATURE_FZRM (12*32+10)/* "" Fast zero-length REP MOVSB */ +#define X86_FEATURE_FSRS (12*32+11)/* "" Fast short REP STOSB */ +#define X86_FEATURE_FSRC (12*32+12)/* "" Fast short REP {CMPSB,SCASB} */ /* AMD-defined CPU features, CPUID level 0x80000008 (EBX), word 13 */ #define X86_FEATURE_CLZERO (13*32+ 0)/* CLZERO instruction */
Signature · 1 line
2.37.2.789.g6183377224-goog
PATCH v3 2/2 KVM: x86: Expose CPUID.(EAX=7,ECX=1).EAX[12:10] to the guest
Fast zero-length REP MOVSB, fast short REP STOSB, and fast short REP {CMPSB,SCASB} are inherent features of the processor that cannot be hidden by the hypervisor. When these features are present on the host, enumerate them in KVM_GET_SUPPORTED_CPUID.
--- arch/x86/kvm/cpuid.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
arch/x86/kvm/cpuid.c +1 -1
@@ -636,7 +636,7 @@ void kvm_set_cpu_caps(void)kvm_cpu_cap_set (X86_FEATURE_SPEC_CTRL_SSBD );kvm_cpu_cap_mask (CPUID_7_1_EAX , -F (AVX_VNNI ) |F (AVX512_BF16 ) +F (AVX_VNNI ) |F (AVX512_BF16 ) |F (FZRM ) |F (FSRS ) |F (FSRC ) ); kvm_cpu_cap_mask(CPUID_D_1_EAX,
Signature · 1 line
2.37.2.789.g6183377224-goog
Re: [PATCH v3 2/2] KVM: x86: Expose CPUID.(EAX=7,ECX=1).EAX[12:10] to the guest
The shortlog is trying to win some kind of award.
And technically, KVM_GET_SUPPORTED_CPUID advertises features to userspace, not to the guest. How about this?
KVM: x86: Advertise fast REP string features inherent to the CPU
Quoted 6 lines · Jim Mattson
Fast zero-length REP MOVSB, fast short REP STOSB, and fast short REP {CMPSB,SCASB} are inherent features of the processor that cannot be hidden by the hypervisor. When these features are present on the host, enumerate them in
KVM_GET_SUPPORTED_CPUID.Signed-off-by Jim Mattson <[email protected]>
With a less cryptic shortlog,
H. Peter Anvin
reviewerQuoted 15 lines · Jim Mattson
KVM_GET_SUPPORTED_CPUIDshould reflect these host CPUID bits. The bits are already cached in word 12. Give the bitsX86_FEATUREnames, so that they can be easily referenced. Hide these bits from/proc/cpuinfo, since the host kernel makes no use of them at present.Signed-off-by Jim Mattson <[email protected]>--- arch/x86/include/asm/cpufeatures.h | 3 +++ 1 file changed, 3 insertions(+)arch/x86/include/asm/cpufeatures.h +0
-0
arch/x86/include/asm/cpufeatures.h line 316
Quoted 9 lines
arch/x86/include/asm/cpufeatures.h +3
-0@@ -308,6 +308,9 @@/* Intel-defined CPU features, CPUID level 0x00000007:1 (EAX), word 12 */ #define X86_FEATURE_AVX_VNNI (12*32+ 4)/* AVX VNNI instructions */ #define X86_FEATURE_AVX512_BF16 (12*32+ 5)/* AVX512 BFLOAT16 instructions */ +#define X86_FEATURE_FZRM (12*32+10)/* "" Fast zero-length REP MOVSB */ +#define X86_FEATURE_FSRS (12*32+11)/* "" Fast short REP STOSB */ +#define X86_FEATURE_FSRC (12*32+12)/* "" Fast short REP {CMPSB,SCASB} */ /* AMD-defined CPU features, CPUID level 0x80000008 (EBX), word 13 */ #define X86_FEATURE_CLZERO (13*32+ 0)/* CLZERO instruction */
Any reason why these bits are hidden from /proc/cpuinfo?
-hpa
Borislav Petkov
maintainerQuoted 1 line · H. Peter Anvin
Any reason why these bits are hidden from
/proc/cpuinfo?
Yes, we aim to hide such purely CPUID bits from /proc/cpuinfo because it becomes a dumping ground for "enablement" of new features. But
1. those features are not really used - most userspace like binutils and gcc, etc do their own detection. (Yes, yes, I'd like to have ubiquitous CPUID faulting).
2. /proc/cpuinfo is an ABI so we have to carry *all* those gazillion flags for no good reason
So we have tools/arch/x86/kcpuid/ which we control and we can extend with all the CPUID querying needs we have.
These kvm enablement things are kinda needed because guest userspace gets an emulated CPUID so in order to detect features on its own, it needs them. And kvm has tied features to x86's X86_FEATURE stuff and there are sometimes weird interactions with it too but that's another topic...
Oh and we still do add visible flags to /proc/cpuinfo but only when they're features which need and have received non-trivial kernel enablement like TDX or SNP or so.
Signature · 4 lines
Regards/Gruss,
Boris.
https://people.kernel.org/tglx/notes-about-netiquetteQuoted 7 lines · Jim Mattson
KVM_GET_SUPPORTED_CPUIDshould reflect these host CPUID bits. The bits are already cached in word 12. Give the bitsX86_FEATUREnames, so that they can be easily referenced. Hide these bits from/proc/cpuinfo, since the host kernel makes no use of them at present.Signed-off-by Jim Mattson <[email protected]>---
Quoted 6 lines · Jim Mattson
KVM_GET_SUPPORTED_CPUIDshould reflect these host CPUID bits. The bits are already cached in word 12. Give the bitsX86_FEATUREnames, so that they can be easily referenced. Hide these bits from/proc/cpuinfo, since the host kernel makes no use of them at present.
Applied to kvm-x86 misc, thanks!
[1/2] x86/cpufeatures: Add macros for Intel's new fast rep string features
https://github.com/kvm-x86/linux/commit/751b1e1ee8e9 [2/2] KVM: x86: Expose CPUID.(EAX=7,ECX=1).EAX[12:10] to the guest https://github.com/kvm-x86/linux/commit/bbbe171ac12f
--
https://github.com/kvm-x86/linux/tree/next
https://github.com/kvm-x86/linux/tree/fixes