Skip to contentSkip to conversation

x86: KVM: Expose CPUID to guest for new Intel platform instructions

PATCH In review Jiaxi Chen 5 participants 24 messages Oct 19, 2022 – Nov 3, 2022

Subsystems (3)
lore
Share & tools
lore

Jiaxi Chen

author
#0 lore raw

cover x86: KVM: Expose CPUID to guest for new Intel platform instructions

Latest Intel platform Granite Rapids/Sierra Forest has introduced below new instructions and CPUIDs:

- CMPccXADD CPUID.(EAX=7,ECX=1):EAX[bit 7]
- AMX-FP16 CPUID.(EAX=7,ECX=1):EAX[bit 21]
- AVX-IFMA CPUID.(EAX=7,ECX=1):EAX[bit 23]
- AVX-VNNI-INT8 CPUID.(EAX=7,ECX=1):EDX[bit 4]
- AVX-NE-CONVERT CPUID.(EAX=7,ECX=1):EDX[bit 5]
- PREFETCHITI CPUID.(EAX=7,ECX=1):EDX[bit 14]

Details can be found in recent Intel ISE (Instruction Set Extensions)[1].

This series enables related CPUID bit in the kernel feature bits and expose them to KVM guest OS.

[1] Intel ISE: https://cdrdv2.intel.com/v1/dl/getContent/671368

Chang S. Bae (1):
x86: KVM: Enable AMX-FP16 CPUID and expose it to guest

Jiaxi Chen (5):
x86: KVM: Enable CMPccXADD CPUID and expose it to guest
x86: KVM: Enable AVX-IFMA CPUID and expose it to guest
x86: KVM: Enable AVX-VNNI-INT8 CPUID and expose it to guest
x86: KVM: Enable AVX-NE-CONVERT CPUID and expose it to guest
x86: KVM: Enable PREFETCHIT0/1 CPUID and expose it to guest

 arch/x86/include/asm/cpufeature.h        |  7 +++++--
 arch/x86/include/asm/cpufeatures.h       | 10 +++++++++-
 arch/x86/include/asm/disabled-features.h |  3 ++-
 arch/x86/include/asm/required-features.h |  3 ++-
 arch/x86/kernel/cpu/common.c             |  1 +
 arch/x86/kvm/cpuid.c                     |  8 ++++++--
 arch/x86/kvm/reverse_cpuid.h             |  1 +
 7 files changed, 26 insertions(+), 7 deletions(-)

base-commit: e18d6152ff0f41b7f01f9817372022df04e0d354

Signature · 1 line
2.27.0
author
#1 lore raw

PATCH 1/6 x86: KVM: Enable CMPccXADD CPUID and expose it to guest

CMPccXADD is a new set of instructions in the latest Intel platform Sierra Forest. It includes a semaphore operation that can compare and add the operands if condition is met, which can improve database performance.

The bit definition:
CPUID.(EAX=7,ECX=1):EAX[bit 7]

This patch enables this CPUID in the kernel feature bits and expose it to guest OS.

---
 arch/x86/include/asm/cpufeatures.h | 1 +
 arch/x86/kvm/cpuid.c               | 2 +-
 2 files changed, 2 insertions(+), 1 deletion(-)
arch/x86/include/asm/cpufeatures.h +1 -0
index ef4775c6db01..445626cb5779 100644
@@ -308,6 +308,7 @@
 /* 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_CMPCCXADD           (12*32+ 7) /* CMPccXADD instructions */
 
 /* AMD-defined CPU features, CPUID level 0x80000008 (EBX), word 13 */
 #define X86_FEATURE_CLZERO		(13*32+ 0) /* CLZERO instruction */
arch/x86/kvm/cpuid.c +1 -1
index 7065462378e2..3f745f6fdc43 100644
@@ -657,7 +657,7 @@ 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(CMPCCXADD)
 	);
 
 	kvm_cpu_cap_mask(CPUID_D_1_EAX,
Signature · 1 line
2.27.0
#2 lore raw

Re: [PATCH 1/6] x86: KVM: Enable CMPccXADD CPUID and expose it to guest

For all the shortlogs, "expose it to guest" is technically wrong. Adding recognition in kvm/cpuid.c advertises KVM support to host userspace. Whether or not a feature is exposed to the guest is up to the userspace VMM.

Quoted 2 lines · Jiaxi Chen

CMPccXADD is a new set of instructions in the latest Intel platform Sierra Forest. It includes a semaphore operation that can compare and add the

In general, avoid pronouns in changelogs, it's not clear what "it" refers to here.

And for all of these changelogs, please explicitly state that there are no VMX controls for these instructions, assuming that's actually true. From a KVM perspective, that's far more interesting than the details of the instruction(s).

Quoted 7 lines

operands if condition is met, which can improve database performance.

The bit definition:
CPUID.(EAX=7,ECX=1):EAX[bit 7]

This patch enables this CPUID in the kernel feature bits and expose it to guest OS.

Same thing here, KVM doesn't decide whether or not to expose the feature to the guest.

Quoted 11 lines
---
 arch/x86/include/asm/cpufeatures.h | 1 +
 arch/x86/kvm/cpuid.c               | 2 +-
 2 files changed, 2 insertions(+), 1 deletion(-)
arch/x86/include/asm/cpufeatures.h +0 -0
index ef4775c6db01..445626cb5779 100644

arch/x86/include/asm/cpufeatures.h line 311

Quoted 4 lines
arch/x86/include/asm/cpufeatures.h +1 -0
@@ -308,6 +308,7 @@
 /* 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_CMPCCXADD           (12*32+ 7) /* CMPccXADD instructions */

Boris,

What do you think about moving CPUID_7_1_EAX to be a KVM-only leaf too? AFAICT, KVM passthrough is the only reason the existing features are defined.

author Applied
#3 lore raw

Re: [PATCH 1/6] x86: KVM: Enable CMPccXADD CPUID and expose it to guest

Quoted 3 lines · Sean Christopherson

For all the shortlogs, "expose it to guest" is technically wrong. Adding recognition in kvm/cpuid.c advertises KVM support to host userspace. Whether or not a feature is exposed to the guest is up to the userspace VMM.

Thanks for reminding. How about to change the subject to this:
x86: KVM: Advertise CMPccXADD CPUID to userspace

Quoted 7 lines · Jiaxi Chen
Quoted 2 lines

CMPccXADD is a new set of instructions in the latest Intel platform Sierra Forest. It includes a semaphore operation that can compare and add the

In general, avoid pronouns in changelogs, it's not clear what "it" refers to here.

Will change it to: 'This new instruction set' here and avoid use pronouns in the future commit message.

Quoted 4 lines

And for all of these changelogs, please explicitly state that there are no VMX controls for these instructions, assuming that's actually true. From a KVM perspective, that's far more interesting than the details of the instruction(s).

Yes, thanks for comments. Will change this patch series to:
This instruction has no other VMX control except for exposed to userspace.

Quoted 11 lines
Quoted 7 lines

operands if condition is met, which can improve database performance.

The bit definition:
CPUID.(EAX=7,ECX=1):EAX[bit 7]

This patch enables this CPUID in the kernel feature bits and expose it to guest OS.

Same thing here, KVM doesn't decide whether or not to expose the feature to the guest.

Applied.Thanks.

Quoted 20 lines
Quoted 11 lines
---
 arch/x86/include/asm/cpufeatures.h | 1 +
 arch/x86/kvm/cpuid.c               | 2 +-
 2 files changed, 2 insertions(+), 1 deletion(-)
arch/x86/include/asm/cpufeatures.h +0 -0
index ef4775c6db01..445626cb5779 100644

arch/x86/include/asm/cpufeatures.h line 311

Quoted 4 lines
arch/x86/include/asm/cpufeatures.h +1 -0
@@ -308,6 +308,7 @@
 /* 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_CMPCCXADD           (12*32+ 7) /* CMPccXADD instructions */

Boris,

What do you think about moving CPUID_7_1_EAX to be a KVM-only leaf too? AFAICT, KVM passthrough is the only reason the existing features are defined.

author
#4 lore raw

Re: [PATCH 1/6] x86: KVM: Enable CMPccXADD CPUID and expose it to guest

In reply to Sean Christopherson
Quoted 43 lines · Sean Christopherson

For all the shortlogs, "expose it to guest" is technically wrong. Adding recognition in kvm/cpuid.c advertises KVM support to host userspace. Whether or not a feature is exposed to the guest is up to the userspace VMM.

Quoted 2 lines · Jiaxi Chen

CMPccXADD is a new set of instructions in the latest Intel platform Sierra Forest. It includes a semaphore operation that can compare and add the

In general, avoid pronouns in changelogs, it's not clear what "it" refers to here.

And for all of these changelogs, please explicitly state that there are no VMX controls for these instructions, assuming that's actually true. From a KVM perspective, that's far more interesting than the details of the instruction(s).

Quoted 7 lines

operands if condition is met, which can improve database performance.

The bit definition:
CPUID.(EAX=7,ECX=1):EAX[bit 7]

This patch enables this CPUID in the kernel feature bits and expose it to guest OS.

Same thing here, KVM doesn't decide whether or not to expose the feature to the guest.

Quoted 11 lines
---
 arch/x86/include/asm/cpufeatures.h | 1 +
 arch/x86/kvm/cpuid.c               | 2 +-
 2 files changed, 2 insertions(+), 1 deletion(-)
arch/x86/include/asm/cpufeatures.h +0 -0
index ef4775c6db01..445626cb5779 100644

arch/x86/include/asm/cpufeatures.h line 311

Quoted 4 lines
arch/x86/include/asm/cpufeatures.h +1 -0
@@ -308,6 +308,7 @@
 /* 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_CMPCCXADD           (12*32+ 7) /* CMPccXADD instructions */

Boris,

Quoted 2 lines

What do you think about moving CPUID_7_1_EAX to be a KVM-only leaf too? AFAICT, KVM passthrough is the only reason the existing features are defined.

Boris,

Since CPUID_7_1_EAX has only 5 features now, it is a big waste, should we move it to KVM-only leaf as Sean suggested. What's your opinion about this?

Signature · 2 lines
Regards,
Jiaxi
maintainer
#5 lore raw

Re: [PATCH 1/6] x86: KVM: Enable CMPccXADD CPUID and expose it to guest

Quoted 2 lines · Jiaxi Chen
Quoted 2 lines

What do you think about moving CPUID_7_1_EAX to be a KVM-only leaf too? AFAICT, KVM passthrough is the only reason the existing features are defined.

Yap, looking at the patches which added those 2 feature flags upstream, they don't look like some particular use was the goal but rather to expose it to guests. Besides, AVX512 apps do their own CPUID detection.

Quoted 3 lines

Since CPUID_7_1_EAX has only 5 features now, it is a big waste, should we move it to KVM-only leaf as Sean suggested. What's your opinion about this?

Yes, pls do.

And when you do, make sure to undo what

b302e4b176d0 ("x86/cpufeatures: Enumerate the new AVX512 BFLOAT16 instructions")

added.

Thx.

Signature · 4 lines
Regards/Gruss,
    Boris.

https://people.kernel.org/tglx/notes-about-netiquette
author
#6 lore raw

Re: [PATCH 1/6] x86: KVM: Enable CMPccXADD CPUID and expose it to guest

Quoted 14 lines · Borislav Petkov
Quoted 2 lines · Jiaxi Chen
Quoted 2 lines

What do you think about moving CPUID_7_1_EAX to be a KVM-only leaf too? AFAICT, KVM passthrough is the only reason the existing features are defined.

Yap, looking at the patches which added those 2 feature flags upstream, they don't look like some particular use was the goal but rather to expose it to guests. Besides, AVX512 apps do their own CPUID detection.

Quoted 3 lines

Since CPUID_7_1_EAX has only 5 features now, it is a big waste, should we move it to KVM-only leaf as Sean suggested. What's your opinion about this?

Yes, pls do.

Quoted 8 lines

And when you do, make sure to undo what

b302e4b176d0 ("x86/cpufeatures: Enumerate the new AVX512 BFLOAT16 instructions")

added.

Thx.

Yes, will do this in v2. Thanks for reminding~

Signature · 2 lines
Regards,
Jiaxi
author
#7 lore raw

Re: [PATCH 1/6] x86: KVM: Enable CMPccXADD CPUID and expose it to guest

In reply to Borislav Petkov
Quoted 14 lines · Borislav Petkov
Quoted 2 lines · Jiaxi Chen
Quoted 2 lines

What do you think about moving CPUID_7_1_EAX to be a KVM-only leaf too? AFAICT, KVM passthrough is the only reason the existing features are defined.

Yap, looking at the patches which added those 2 feature flags upstream, they don't look like some particular use was the goal but rather to expose it to guests. Besides, AVX512 apps do their own CPUID detection.

Quoted 3 lines

Since CPUID_7_1_EAX has only 5 features now, it is a big waste, should we move it to KVM-only leaf as Sean suggested. What's your opinion about this?

Yes, pls do.

Quoted 8 lines

And when you do, make sure to undo what

b302e4b176d0 ("x86/cpufeatures: Enumerate the new AVX512 BFLOAT16 instructions")

added.

Thx.

Hi Sean and Boris,

Just realized moving CPUID_7_1_EAX to kvm-only leaf will not save space in enum cpuid_leafs[]. CPUID_7_1_EAX is indeed removed, but someone else, ie. CPUID_DUMMY needs to take the place, otherwise the cpuid_leafs array would be deranged. Therefore, the length of x86 cpuid leaves is not decreased.

Wonder if the intention of moving this leaf to kvm-only is for saving space in x86_capability[], or just because there's no other use case in the host kernel side and the cpuflags of this features can be removed.

Hope for your suggestions.

Signature · 2 lines
Regards,
Jiaxi
#8 lore raw

Re: [PATCH 1/6] x86: KVM: Enable CMPccXADD CPUID and expose it to guest

Quoted 25 lines · Jiaxi Chen
Quoted 14 lines · Borislav Petkov
Quoted 2 lines · Jiaxi Chen
Quoted 2 lines

What do you think about moving CPUID_7_1_EAX to be a KVM-only leaf too? AFAICT, KVM passthrough is the only reason the existing features are defined.

Yap, looking at the patches which added those 2 feature flags upstream, they don't look like some particular use was the goal but rather to expose it to guests. Besides, AVX512 apps do their own CPUID detection.

Quoted 3 lines

Since CPUID_7_1_EAX has only 5 features now, it is a big waste, should we move it to KVM-only leaf as Sean suggested. What's your opinion about this?

Yes, pls do.

Quoted 8 lines

And when you do, make sure to undo what

b302e4b176d0 ("x86/cpufeatures: Enumerate the new AVX512 BFLOAT16 instructions")

added.

Thx.

Quoted 7 lines

Hi Sean and Boris,

Just realized moving CPUID_7_1_EAX to kvm-only leaf will not save space in enum cpuid_leafs[]. CPUID_7_1_EAX is indeed removed, but someone else, ie. CPUID_DUMMY needs to take the place, otherwise the cpuid_leafs array would be deranged. Therefore, the length of x86 cpuid leaves is not decreased.

The order of "enum cpuid_leafs" is completely arbitrary.

After replacing CPUID_7_1_EAX with CPUID_DUMMY, replace CPUID_DUMMY with the last leaf, which is currently CPUID_8000_001F_EAX, and update the #defines accordingly. Alternatively, Boris may prefer skipping the intermediate CPUID_DUMMY step and just replace CPUID_7_1_EAX with CPUID_8000_001F_EAX straightaway.

author
#9 lore raw

Re: [PATCH 1/6] x86: KVM: Enable CMPccXADD CPUID and expose it to guest

Quoted 34 lines · Sean Christopherson
Quoted 25 lines · Jiaxi Chen
Quoted 14 lines · Borislav Petkov
Quoted 2 lines · Jiaxi Chen
Quoted 2 lines

What do you think about moving CPUID_7_1_EAX to be a KVM-only leaf too? AFAICT, KVM passthrough is the only reason the existing features are defined.

Yap, looking at the patches which added those 2 feature flags upstream, they don't look like some particular use was the goal but rather to expose it to guests. Besides, AVX512 apps do their own CPUID detection.

Quoted 3 lines

Since CPUID_7_1_EAX has only 5 features now, it is a big waste, should we move it to KVM-only leaf as Sean suggested. What's your opinion about this?

Yes, pls do.

Quoted 8 lines

And when you do, make sure to undo what

b302e4b176d0 ("x86/cpufeatures: Enumerate the new AVX512 BFLOAT16 instructions")

added.

Thx.

Quoted 7 lines

Hi Sean and Boris,

Just realized moving CPUID_7_1_EAX to kvm-only leaf will not save space in enum cpuid_leafs[]. CPUID_7_1_EAX is indeed removed, but someone else, ie. CPUID_DUMMY needs to take the place, otherwise the cpuid_leafs array would be deranged. Therefore, the length of x86 cpuid leaves is not decreased.

Quoted 6 lines

The order of "enum cpuid_leafs" is completely arbitrary.

After replacing CPUID_7_1_EAX with CPUID_DUMMY, replace CPUID_DUMMY with the last leaf, which is currently CPUID_8000_001F_EAX, and update the #defines accordingly. Alternatively, Boris may prefer skipping the intermediate CPUID_DUMMY step and just replace CPUID_7_1_EAX with CPUID_8000_001F_EAX straightaway.

Yes, thanks for Sean's kind suggestion. I think use CPUID_DUMMY as the transition leaf will make the code logic and commit message clearer. Will change it in v2.

Signature · 2 lines
Regards,
Jiaxi
author
#10 lore raw

PATCH 2/6 x86: KVM: Enable AMX-FP16 CPUID and expose it to guest

In reply to Jiaxi Chen

From: "Chang S. Bae" <[email protected]>

Latest Intel platform Granite Rapids has introduced a new instruction - AMX-FP16, which performs dot-products of two FP16 tiles and accumulates the results into a packed single precision tile. This instruction adds FP16 capability and also allows a FP16 GPU trained model to run faster without loss of accuracy or added SW overhead.

The bit definition:
CPUID.(EAX=7,ECX=1):EAX[bit 21]

This patch enables this CPUID in the kernel feature bits and expose it to guest OS.

---
 arch/x86/include/asm/cpufeatures.h | 1 +
 arch/x86/kvm/cpuid.c               | 2 +-
 2 files changed, 2 insertions(+), 1 deletion(-)
arch/x86/include/asm/cpufeatures.h +1 -0
index 445626cb5779..9313240e3cdd 100644
@@ -309,6 +309,7 @@
 #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_CMPCCXADD           (12*32+ 7) /* CMPccXADD instructions */
+#define X86_FEATURE_AMX_FP16		(12*32+21) /* AMX fp16 Support */
 
 /* AMD-defined CPU features, CPUID level 0x80000008 (EBX), word 13 */
 #define X86_FEATURE_CLZERO		(13*32+ 0) /* CLZERO instruction */
arch/x86/kvm/cpuid.c +1 -1
index 3f745f6fdc43..d983ddb974ba 100644
@@ -657,7 +657,7 @@ 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(CMPCCXADD)
+		F(AVX_VNNI) | F(AVX512_BF16) | F(CMPCCXADD) | F(AMX_FP16)
 	);
 
 	kvm_cpu_cap_mask(CPUID_D_1_EAX,
Signature · 1 line
2.27.0
#11 lore raw

Re: [PATCH 2/6] x86: KVM: Enable AMX-FP16 CPUID and expose it to guest

arch/x86/include/asm/cpufeatures.h line 312

Quoted 9 lines · Jiaxi Chen
arch/x86/include/asm/cpufeatures.h +1 -0
index 445626cb5779..9313240e3cdd 100644
@@ -309,6 +309,7 @@
 #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_CMPCCXADD           (12*32+ 7) /* CMPccXADD instructions */
+#define X86_FEATURE_AMX_FP16		(12*32+21) /* AMX fp16 Support */

Please zap these from /proc/cpuinfo by doing this:

#define X86_FEATURE_AMX_FP16 (12*32+21) /* "" AMX fp16 Support */

Quoted 7 lines
arch/x86/include/asm/cpufeatures.h +0 -0
 /* AMD-defined CPU features, CPUID level 0x80000008 (EBX), word 13 */
 #define X86_FEATURE_CLZERO		(13*32+ 0) /* CLZERO instruction */
arch/x86/kvm/cpuid.c +0 -0
index 3f745f6fdc43..d983ddb974ba 100644

arch/x86/kvm/cpuid.c void kvm_set_cpu_caps(void) · line 663

Quoted 8 lines
arch/x86/kvm/cpuid.c +1 -1
@@ -657,7 +657,7 @@ 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(CMPCCXADD)
+		F(AVX_VNNI) | F(AVX512_BF16) | F(CMPCCXADD) | F(AMX_FP16)
 	);
 
 	kvm_cpu_cap_mask(CPUID_D_1_EAX,

KVM folks, is the idea that every feature that is enumerated to a guest needs to be in one of these masks? Or is there something special about the features in these masks?

maintainer
#12 lore raw

Re: [PATCH 2/6] x86: KVM: Enable AMX-FP16 CPUID and expose it to guest

Quoted 11 lines · Dave Hansen

arch/x86/kvm/cpuid.c void kvm_set_cpu_caps(void) · line 663

Quoted 7 lines

kvm_cpu_cap_mask(CPUID_7_1_EAX,

arch/x86/kvm/cpuid.c +1 -1
-		F(AVX_VNNI) | F(AVX512_BF16) | F(CMPCCXADD)
+		F(AVX_VNNI) | F(AVX512_BF16) | F(CMPCCXADD) | F(AMX_FP16)
  	);
 
  	kvm_cpu_cap_mask(CPUID_D_1_EAX,

KVM folks, is the idea that every feature that is enumerated to a guest needs to be in one of these masks? Or is there something special about the features in these masks?

Yes, all features are vetted manually to see whether they require new MSRs and the like. Therefore, anything that userspace can set in the guest's CPUID must be in the list.

Paolo

#13 lore raw

Re: [PATCH 2/6] x86: KVM: Enable AMX-FP16 CPUID and expose it to guest

Quoted 7 lines · Paolo Bonzini
Quoted 6 lines · Dave Hansen
Quoted 5 lines

        kvm_cpu_cap_mask(CPUID_7_1_EAX,

-        F(AVX_VNNI) | F(AVX512_BF16) | F(CMPCCXADD)
+        F(AVX_VNNI) | F(AVX512_BF16) | F(CMPCCXADD) | F(AMX_FP16)

      );
        kvm_cpu_cap_mask(CPUID_D_1_EAX,

Quoted 7 lines
Quoted 3 lines

KVM folks, is the idea that every feature that is enumerated to a guest needs to be in one of these masks?  Or is there something special about the features in these masks?

Yes, all features are vetted manually to see whether they require new MSRs and the like.  Therefore, anything that userspace can set in the guest's CPUID must be in the list.

Makes sense.

Intel folks, when you add these bits, can you please include information about the "vetting" that you performed?

For example, it would be handy to say:

AMX_FP16 is just a new instruction that operates on existing AMX tile registers. It needs no additional enabling on top of the existing kernel AMX enabling.

author
#14 lore raw

Re: [PATCH 2/6] x86: KVM: Enable AMX-FP16 CPUID and expose it to guest

Quoted 18 lines · Dave Hansen
Quoted 7 lines · Paolo Bonzini
Quoted 6 lines · Dave Hansen
Quoted 5 lines

        kvm_cpu_cap_mask(CPUID_7_1_EAX,

-        F(AVX_VNNI) | F(AVX512_BF16) | F(CMPCCXADD)
+        F(AVX_VNNI) | F(AVX512_BF16) | F(CMPCCXADD) | F(AMX_FP16)

      );
        kvm_cpu_cap_mask(CPUID_D_1_EAX,

Quoted 7 lines
Quoted 3 lines

KVM folks, is the idea that every feature that is enumerated to a guest needs to be in one of these masks?  Or is there something special about the features in these masks?

Yes, all features are vetted manually to see whether they require new MSRs and the like.  Therefore, anything that userspace can set in the guest's CPUID must be in the list.

Makes sense.

Quoted 8 lines

Intel folks, when you add these bits, can you please include information about the "vetting" that you performed?

For example, it would be handy to say:

AMX_FP16 is just a new instruction that operates on existing AMX tile registers. It needs no additional enabling on top of the existing kernel AMX enabling.

Thanks for Dave's suggestion. It makes the commit message much clear and readable. Will modify it in v2.

Signature · 2 lines
Regards,
Jiaxi
author
#15 lore raw

PATCH 3/6 x86: KVM: Enable AVX-IFMA CPUID and expose it to guest

In reply to Jiaxi Chen

AVX-IFMA is a new instruction in the latest Intel platform Sierra Forest. AVX-IFMA packed multiplies unsigned 52-bit integers and add the low/high 52-bit products to Qword Accumulators.

The bit definition:
CPUID.(EAX=7,ECX=1):EAX[bit 23]

This patch enables this CPUID in the kernel feature bits and expose it to guest OS.

---
 arch/x86/include/asm/cpufeatures.h | 1 +
 arch/x86/kvm/cpuid.c               | 4 ++--
 2 files changed, 3 insertions(+), 2 deletions(-)
arch/x86/include/asm/cpufeatures.h +1 -0
index 9313240e3cdd..a682f646243f 100644
@@ -310,6 +310,7 @@
 #define X86_FEATURE_AVX512_BF16		(12*32+ 5) /* AVX512 BFLOAT16 instructions */
 #define X86_FEATURE_CMPCCXADD           (12*32+ 7) /* CMPccXADD instructions */
 #define X86_FEATURE_AMX_FP16		(12*32+21) /* AMX fp16 Support */
+#define X86_FEATURE_AVX_IFMA            (12*32+23) /* Support for VPMADD52[H,L]UQ */
 
 /* AMD-defined CPU features, CPUID level 0x80000008 (EBX), word 13 */
 #define X86_FEATURE_CLZERO		(13*32+ 0) /* CLZERO instruction */
arch/x86/kvm/cpuid.c +2 -2
index d983ddb974ba..837bcd1373e5 100644
@@ -657,8 +657,8 @@ 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(CMPCCXADD) | F(AMX_FP16)
-	);
+		F(AVX_VNNI) | F(AVX512_BF16) | F(CMPCCXADD) | F(AMX_FP16) |
+		F(AVX_IFMA));
 
 	kvm_cpu_cap_mask(CPUID_D_1_EAX,
 		F(XSAVEOPT) | F(XSAVEC) | F(XGETBV1) | F(XSAVES) | f_xfd
Signature · 1 line
2.27.0
author
#16 lore raw

PATCH 4/6 x86: KVM: Enable AVX-VNNI-INT8 CPUID and expose it to guest

In reply to Jiaxi Chen

AVX-VNNI-INT8 is a new set of instructions in the latest Intel platform Sierra Forest. It multiplies the individual bytes of two unsigned or unsigned source operands, then add and accumulate the results into the destination dword element size operand. This instruction allows for the platform to have superior AI capabilities.

The bit definition:
CPUID.(EAX=7,ECX=1):EDX[bit 4]

This patch enables this CPUID in the kernel feature bits and expose it to guest OS. Since the CPUID involves a bit of EDX (EAX=7,ECX=1) which has not been enumerated yet, this patch adds CPUID_7_1_EDX to CPUID subleaves. At the same time, word 20 is newly-defined in CPU features for CPUID level 0x00000007:1 (EDX).

---
 arch/x86/include/asm/cpufeature.h        | 7 +++++--
 arch/x86/include/asm/cpufeatures.h       | 5 ++++-
 arch/x86/include/asm/disabled-features.h | 3 ++-
 arch/x86/include/asm/required-features.h | 3 ++-
 arch/x86/kernel/cpu/common.c             | 1 +
 arch/x86/kvm/cpuid.c                     | 5 ++++-
 arch/x86/kvm/reverse_cpuid.h             | 1 +
 7 files changed, 19 insertions(+), 6 deletions(-)
arch/x86/include/asm/cpufeature.h +5 -2
index 1a85e1fb0922..d46b802930b0 100644
@@ -32,6 +32,7 @@ enum cpuid_leafs
 	CPUID_8000_0007_EBX,
 	CPUID_7_EDX,
 	CPUID_8000_001F_EAX,
+	CPUID_7_1_EDX,
 };
 
 #define X86_CAP_FMT_NUM "%d:%d"
@@ -94,8 +95,9 @@ extern const char * const x86_bug_flags[NBUGINTS*32];
 	   CHECK_BIT_IN_MASK_WORD(REQUIRED_MASK, 17, feature_bit) ||	\
 	   CHECK_BIT_IN_MASK_WORD(REQUIRED_MASK, 18, feature_bit) ||	\
 	   CHECK_BIT_IN_MASK_WORD(REQUIRED_MASK, 19, feature_bit) ||	\
+	   CHECK_BIT_IN_MASK_WORD(REQUIRED_MASK, 20, feature_bit) ||    \
 	   REQUIRED_MASK_CHECK					  ||	\
-	   BUILD_BUG_ON_ZERO(NCAPINTS != 20))
+	   BUILD_BUG_ON_ZERO(NCAPINTS != 21))
 
 #define DISABLED_MASK_BIT_SET(feature_bit)				\
 	 ( CHECK_BIT_IN_MASK_WORD(DISABLED_MASK,  0, feature_bit) ||	\
@@ -118,8 +120,9 @@ extern const char * const x86_bug_flags[NBUGINTS*32];
 	   CHECK_BIT_IN_MASK_WORD(DISABLED_MASK, 17, feature_bit) ||	\
 	   CHECK_BIT_IN_MASK_WORD(DISABLED_MASK, 18, feature_bit) ||	\
 	   CHECK_BIT_IN_MASK_WORD(DISABLED_MASK, 19, feature_bit) ||	\
+	   CHECK_BIT_IN_MASK_WORD(DISABLED_MASK, 20, feature_bit) ||    \
 	   DISABLED_MASK_CHECK					  ||	\
-	   BUILD_BUG_ON_ZERO(NCAPINTS != 20))
+	   BUILD_BUG_ON_ZERO(NCAPINTS != 21))
 
 #define cpu_has(c, bit)							\
 	(__builtin_constant_p(bit) && REQUIRED_MASK_BIT_SET(bit) ? 1 :	\
arch/x86/include/asm/cpufeatures.h +4 -1
index a682f646243f..b2aa761ea110 100644
@@ -13,7 +13,7 @@
 /*
  * Defines x86 CPU feature bits
  */
-#define NCAPINTS			20	   /* N 32-bit words worth of info */
+#define NCAPINTS			21	   /* N 32-bit words worth of info */
 #define NBUGINTS			1	   /* N 32-bit bug flags */
 
 /*
@@ -423,6 +423,9 @@
 #define X86_FEATURE_V_TSC_AUX		(19*32+ 9) /* "" Virtual TSC_AUX */
 #define X86_FEATURE_SME_COHERENT	(19*32+10) /* "" AMD hardware-enforced cache coherency */
 
+/* Intel-defined CPU features, CPUID level 0x00000007:1 (EDX), word 20 */
+#define X86_FEATURE_AVX_VNNI_INT8       (20*32+ 4) /* Support for VPDPB[SU,UU,SS]D[,S] */
+
 /*
  * BUG word(s)
  */
arch/x86/include/asm/disabled-features.h +2 -1
index 33d2cd04d254..2db6929ca868 100644
@@ -111,6 +111,7 @@
 #define DISABLED_MASK17	0
 #define DISABLED_MASK18	0
 #define DISABLED_MASK19	0
-#define DISABLED_MASK_CHECK BUILD_BUG_ON_ZERO(NCAPINTS != 20)
+#define DISABLED_MASK20 0
+#define DISABLED_MASK_CHECK BUILD_BUG_ON_ZERO(NCAPINTS != 21)
 
 #endif /* _ASM_X86_DISABLED_FEATURES_H */
arch/x86/include/asm/required-features.h +2 -1
index aff774775c67..3b5522b3f051 100644
@@ -98,6 +98,7 @@
 #define REQUIRED_MASK17	0
 #define REQUIRED_MASK18	0
 #define REQUIRED_MASK19	0
-#define REQUIRED_MASK_CHECK BUILD_BUG_ON_ZERO(NCAPINTS != 20)
+#define REQUIRED_MASK20 0
+#define REQUIRED_MASK_CHECK BUILD_BUG_ON_ZERO(NCAPINTS != 21)
 
 #endif /* _ASM_X86_REQUIRED_FEATURES_H */
arch/x86/kernel/cpu/common.c +1 -0
index 3e508f239098..7c659127ed89 100644
@@ -1031,6 +1031,7 @@ void get_cpu_cap(struct cpuinfo_x86 *c)
 		if (eax >= 1) {
 			cpuid_count(0x00000007, 1, &eax, &ebx, &ecx, &edx);
 			c->x86_capability[CPUID_7_1_EAX] = eax;
+			c->x86_capability[CPUID_7_1_EDX] = edx;
 		}
 	}
 
arch/x86/kvm/cpuid.c +4 -1
index 837bcd1373e5..b1b53a5c788a 100644
@@ -660,6 +660,9 @@ void kvm_set_cpu_caps(void)
 		F(AVX_VNNI) | F(AVX512_BF16) | F(CMPCCXADD) | F(AMX_FP16) |
 		F(AVX_IFMA));
 
+	kvm_cpu_cap_mask(CPUID_7_1_EDX,
+		F(AVX_VNNI_INT8));
+
 	kvm_cpu_cap_mask(CPUID_D_1_EAX,
 		F(XSAVEOPT) | F(XSAVEC) | F(XGETBV1) | F(XSAVES) | f_xfd
 	);
@@ -913,9 +916,9 @@ static inline int __do_cpuid_func(struct kvm_cpuid_array *array, u32 function)
 				goto out;
 
 			cpuid_entry_override(entry, CPUID_7_1_EAX);
+			cpuid_entry_override(entry, CPUID_7_1_EDX);
 			entry->ebx = 0;
 			entry->ecx = 0;
-			entry->edx = 0;
 		}
 		break;
 	case 0xa: { /* Architectural Performance Monitoring */
arch/x86/kvm/reverse_cpuid.h +1 -0
index a19d473d0184..fb4f4bffad5c 100644
@@ -48,6 +48,7 @@ static const struct cpuid_reg reverse_cpuid[] = {
 	[CPUID_7_1_EAX]       = {         7, 1, CPUID_EAX},
 	[CPUID_12_EAX]        = {0x00000012, 0, CPUID_EAX},
 	[CPUID_8000_001F_EAX] = {0x8000001f, 0, CPUID_EAX},
+	[CPUID_7_1_EDX]       = {         7, 1, CPUID_EDX},
 };
 
 /*
Signature · 1 line
2.27.0
maintainer
#17 lore raw

Re: [PATCH 4/6] x86: KVM: Enable AVX-VNNI-INT8 CPUID and expose it to guest

Quoted 6 lines · Jiaxi Chen

AVX-VNNI-INT8 is a new set of instructions in the latest Intel platform Sierra Forest. It multiplies the individual bytes of two unsigned or unsigned source operands, then add and accumulate the results into the destination dword element size operand. This instruction allows for the platform to have superior AI capabilities.

Quoted 7 lines

The bit definition:
CPUID.(EAX=7,ECX=1):EDX[bit 4]

This patch enables this CPUID in the kernel feature bits and expose it to guest OS. Since the CPUID involves a bit of EDX (EAX=7,ECX=1) which has not been enumerated yet, this patch adds CPUID_7_1_EDX to CPUID subleaves. At the same time, word 20 is newly-defined in CPU features for CPUID level

For all your commit messages:

Avoid having "This patch" or "This commit" in the commit message. It is tautologically useless.

Also, do

$ git grep 'This patch' Documentation/process

for more details.

For this particular one, use scattered.c instead of adding a new leaf.

Thx.

Signature · 4 lines
Regards/Gruss,
    Boris.

https://people.kernel.org/tglx/notes-about-netiquette
#18 lore raw

Re: [PATCH 4/6] x86: KVM: Enable AVX-VNNI-INT8 CPUID and expose it to guest

Quoted 11 lines · Borislav Petkov
Quoted 8 lines · Jiaxi Chen

AVX-VNNI-INT8 is a new set of instructions in the latest Intel platform Sierra Forest. It multiplies the individual bytes of two unsigned or unsigned source operands, then add and accumulate the results into the destination dword element size operand. This instruction allows for the platform to have superior AI capabilities.

The bit definition:
CPUID.(EAX=7,ECX=1):EDX[bit 4]

For this particular one, use scattered.c instead of adding a new leaf.

Unless the kernel wants to use X86_FEATURE_AVX_VNNI_INT8, which seems unlikely, there's no need to create a scattered entry. This can be handled in KVM by adding a KVM-only leaf entry (which will be needed no matter what), plus a #define for X86_FEATURE_AVX_VNNI_INT8 to direct it to the KVM entry.

E.g.

arch/x86/kvm/reverse_cpuid.h +12 -0
index a19d473d0184..25e7bfc61607 100644
@@ -13,6 +13,7 @@
  */
 enum kvm_only_cpuid_leafs {
        CPUID_12_EAX     = NCAPINTS,
+       CPUID_7_1_EDX,
        NR_KVM_CPU_CAPS,
 
        NKVMCAPINTS = NR_KVM_CPU_CAPS - NCAPINTS,
@@ -24,6 +25,16 @@ enum kvm_only_cpuid_leafs {
 #define KVM_X86_FEATURE_SGX1           KVM_X86_FEATURE(CPUID_12_EAX, 0)
 #define KVM_X86_FEATURE_SGX2           KVM_X86_FEATURE(CPUID_12_EAX, 1)
 
+#define KVM_X86_FEATURE_AVX_VNNI_INT8  KVM_X86_FEATURE(CPUID_7_1_EDX, 4)
+
+/*
+ * Alias X86_FEATURE_* to the KVM variant for features in KVM-only leafs that
+ * aren't scattered by cpufeatures.h so that X86_FEATURE_* can be used in KVM,
+ * e.g. to query guest CPUID.  As a bonus, no translation is needed for these
+ * features in __feature_translate().
+ */
+#define X86_FEATURE_AVX_VNNI_INT8      KVM_X86_FEATURE_AVX_VNNI_INT8
+
 struct cpuid_reg {
        u32 function;
        u32 index;
@@ -48,6 +59,7 @@ static const struct cpuid_reg reverse_cpuid[] = {
        [CPUID_7_1_EAX]       = {         7, 1, CPUID_EAX},
        [CPUID_12_EAX]        = {0x00000012, 0, CPUID_EAX},
        [CPUID_8000_001F_EAX] = {0x8000001f, 0, CPUID_EAX},
+       [CPUID_7_1_EDX]       = {         7, 1, CPUID_EDX},
 };
 
 /*
#19 lore raw

Re: [PATCH 4/6] x86: KVM: Enable AVX-VNNI-INT8 CPUID and expose it to guest

Quoted 35 lines · Sean Christopherson
Quoted 11 lines · Borislav Petkov
Quoted 8 lines · Jiaxi Chen

AVX-VNNI-INT8 is a new set of instructions in the latest Intel platform Sierra Forest. It multiplies the individual bytes of two unsigned or unsigned source operands, then add and accumulate the results into the destination dword element size operand. This instruction allows for the platform to have superior AI capabilities.

The bit definition:
CPUID.(EAX=7,ECX=1):EDX[bit 4]

For this particular one, use scattered.c instead of adding a new leaf.

Unless the kernel wants to use X86_FEATURE_AVX_VNNI_INT8, which seems unlikely, there's no need to create a scattered entry. This can be handled in KVM by adding a KVM-only leaf entry (which will be needed no matter what), plus a #define for X86_FEATURE_AVX_VNNI_INT8 to direct it to the KVM entry.

E.g.

arch/x86/kvm/reverse_cpuid.h +1 -0
index a19d473d0184..25e7bfc61607 100644
@@ -13,6 +13,7 @@
  */
 enum kvm_only_cpuid_leafs {
        CPUID_12_EAX     = NCAPINTS,
+       CPUID_7_1_EDX,
        NR_KVM_CPU_CAPS,
 
        NKVMCAPINTS = NR_KVM_CPU_CAPS - NCAPINTS,
@@ -24,6 +25,16 @@ enum kvm_only_cpuid_leafs {
 #define KVM_X86_FEATURE_SGX1           KVM_X86_FEATURE(CPUID_12_EAX, 0)
 #define KVM_X86_FEATURE_SGX2           KVM_X86_FEATURE(CPUID_12_EAX, 1)

arch/x86/kvm/reverse_cpuid.h enum kvm_only_cpuid_leafs { · line 36

Quoted 10 lines
arch/x86/kvm/reverse_cpuid.h +9 -0
 
+#define KVM_X86_FEATURE_AVX_VNNI_INT8  KVM_X86_FEATURE(CPUID_7_1_EDX, 4)
+
+/*
+ * Alias X86_FEATURE_* to the KVM variant for features in KVM-only leafs that
+ * aren't scattered by cpufeatures.h so that X86_FEATURE_* can be used in KVM,
+ * e.g. to query guest CPUID.  As a bonus, no translation is needed for these
+ * features in __feature_translate().
+ */
+#define X86_FEATURE_AVX_VNNI_INT8      KVM_X86_FEATURE_AVX_VNNI_INT8

Actually, there's no need for KVM_X86_FEATURE_AVX_VNNI_INT8 in this case, just #define X86_FEATURE_AVX_VNNI_INT8 directly. The KVM_ prefixed macro exists purely to redirect the non-KVM_ version, but that's unnecessary in this case.

arch/x86/kvm/reverse_cpuid.h +9 -0
index a19d473d0184..38adafb03490 100644
@@ -13,6 +13,7 @@
  */
 enum kvm_only_cpuid_leafs {
        CPUID_12_EAX     = NCAPINTS,
+       CPUID_7_1_EDX,
        NR_KVM_CPU_CAPS,
 
        NKVMCAPINTS = NR_KVM_CPU_CAPS - NCAPINTS,
@@ -24,6 +25,13 @@ enum kvm_only_cpuid_leafs {
 #define KVM_X86_FEATURE_SGX1           KVM_X86_FEATURE(CPUID_12_EAX, 0)
 #define KVM_X86_FEATURE_SGX2           KVM_X86_FEATURE(CPUID_12_EAX, 1)
 
+/*
+ * Omit the KVM_ prefix for features in KVM-only leafs that aren't scattered by
+ * cpufeatures.h so that X86_FEATURE_* can be used in KVM,* e.g. to query guest
+ * CPUID.  As a bonus, no handling in __feature_translate() is needed.
+ */
+#define X86_FEATURE_AVX_VNNI_INT8      KVM_X86_FEATURE(CPUID_7_1_EDX, 4)
+
 struct cpuid_reg {
        u32 function;
        u32 index;
@@ -48,6 +56,7 @@ static const struct cpuid_reg reverse_cpuid[] = {
        [CPUID_7_1_EAX]       = {         7, 1, CPUID_EAX},
        [CPUID_12_EAX]        = {0x00000012, 0, CPUID_EAX},
        [CPUID_8000_001F_EAX] = {0x8000001f, 0, CPUID_EAX},
+       [CPUID_7_1_EDX]       = {         7, 1, CPUID_EDX},
 };
author
#20 lore raw

Re: [PATCH 4/6] x86: KVM: Enable AVX-VNNI-INT8 CPUID and expose it to guest

Quoted 78 lines · Sean Christopherson
Quoted 35 lines · Sean Christopherson
Quoted 11 lines · Borislav Petkov
Quoted 8 lines · Jiaxi Chen

AVX-VNNI-INT8 is a new set of instructions in the latest Intel platform Sierra Forest. It multiplies the individual bytes of two unsigned or unsigned source operands, then add and accumulate the results into the destination dword element size operand. This instruction allows for the platform to have superior AI capabilities.

The bit definition:
CPUID.(EAX=7,ECX=1):EDX[bit 4]

For this particular one, use scattered.c instead of adding a new leaf.

Unless the kernel wants to use X86_FEATURE_AVX_VNNI_INT8, which seems unlikely, there's no need to create a scattered entry. This can be handled in KVM by adding a KVM-only leaf entry (which will be needed no matter what), plus a #define for X86_FEATURE_AVX_VNNI_INT8 to direct it to the KVM entry.

E.g.

arch/x86/kvm/reverse_cpuid.h +1 -0
index a19d473d0184..25e7bfc61607 100644
@@ -13,6 +13,7 @@
  */
 enum kvm_only_cpuid_leafs {
        CPUID_12_EAX     = NCAPINTS,
+       CPUID_7_1_EDX,
        NR_KVM_CPU_CAPS,
 
        NKVMCAPINTS = NR_KVM_CPU_CAPS - NCAPINTS,
@@ -24,6 +25,16 @@ enum kvm_only_cpuid_leafs {
 #define KVM_X86_FEATURE_SGX1           KVM_X86_FEATURE(CPUID_12_EAX, 0)
 #define KVM_X86_FEATURE_SGX2           KVM_X86_FEATURE(CPUID_12_EAX, 1)

arch/x86/kvm/reverse_cpuid.h enum kvm_only_cpuid_leafs { · line 36

Quoted 10 lines
arch/x86/kvm/reverse_cpuid.h +9 -0
 
+#define KVM_X86_FEATURE_AVX_VNNI_INT8  KVM_X86_FEATURE(CPUID_7_1_EDX, 4)
+
+/*
+ * Alias X86_FEATURE_* to the KVM variant for features in KVM-only leafs that
+ * aren't scattered by cpufeatures.h so that X86_FEATURE_* can be used in KVM,
+ * e.g. to query guest CPUID.  As a bonus, no translation is needed for these
+ * features in __feature_translate().
+ */
+#define X86_FEATURE_AVX_VNNI_INT8      KVM_X86_FEATURE_AVX_VNNI_INT8

Actually, there's no need for KVM_X86_FEATURE_AVX_VNNI_INT8 in this case, just #define X86_FEATURE_AVX_VNNI_INT8 directly. The KVM_ prefixed macro exists purely to redirect the non-KVM_ version, but that's unnecessary in this case.

arch/x86/kvm/reverse_cpuid.h +8 -0
index a19d473d0184..38adafb03490 100644
@@ -13,6 +13,7 @@
  */
 enum kvm_only_cpuid_leafs {
        CPUID_12_EAX     = NCAPINTS,
+       CPUID_7_1_EDX,
        NR_KVM_CPU_CAPS,
 
        NKVMCAPINTS = NR_KVM_CPU_CAPS - NCAPINTS,
@@ -24,6 +25,13 @@ enum kvm_only_cpuid_leafs {
 #define KVM_X86_FEATURE_SGX1           KVM_X86_FEATURE(CPUID_12_EAX, 0)
 #define KVM_X86_FEATURE_SGX2           KVM_X86_FEATURE(CPUID_12_EAX, 1)
 
+/*
+ * Omit the KVM_ prefix for features in KVM-only leafs that aren't scattered by
+ * cpufeatures.h so that X86_FEATURE_* can be used in KVM,* e.g. to query guest
+ * CPUID.  As a bonus, no handling in __feature_translate() is needed.
+ */
+#define X86_FEATURE_AVX_VNNI_INT8      KVM_X86_FEATURE(CPUID_7_1_EDX, 4)
+
 struct cpuid_reg {
        u32 function;
        u32 index;

arch/x86/kvm/reverse_cpuid.h static const struct cpuid_reg reverse_cpuid[] = { · line 60

Quoted 5 lines
arch/x86/kvm/reverse_cpuid.h +1 -0
@@ -48,6 +56,7 @@ static const struct cpuid_reg reverse_cpuid[] = {
        [CPUID_7_1_EAX]       = {         7, 1, CPUID_EAX},
        [CPUID_12_EAX]        = {0x00000012, 0, CPUID_EAX},
        [CPUID_8000_001F_EAX] = {0x8000001f, 0, CPUID_EAX},
+       [CPUID_7_1_EDX]       = {         7, 1, CPUID_EDX},
 };

Use KVM-only leafs is better for edx in this case. Will follow this suggestion in v2. Thanks Sean very much.

Signature · 2 lines
Regards,
Jiaxi
#21 lore raw

Re: [PATCH 4/6] x86: KVM: Enable AVX-VNNI-INT8 CPUID and expose it to guest

In reply to Borislav Petkov
Quoted 15 lines · Borislav Petkov
Quoted 6 lines · Jiaxi Chen

AVX-VNNI-INT8 is a new set of instructions in the latest Intel platform Sierra Forest. It multiplies the individual bytes of two unsigned or unsigned source operands, then add and accumulate the results into the destination dword element size operand. This instruction allows for the platform to have superior AI capabilities.

Quoted 7 lines

The bit definition:
CPUID.(EAX=7,ECX=1):EDX[bit 4]

This patch enables this CPUID in the kernel feature bits and expose it to guest OS. Since the CPUID involves a bit of EDX (EAX=7,ECX=1) which has not been enumerated yet, this patch adds CPUID_7_1_EDX to CPUID subleaves. At the same time, word 20 is newly-defined in CPU features for CPUID level

Quoted 4 lines

For all your commit messages:

Avoid having "This patch" or "This commit" in the commit message. It is tautologically useless.

Will follow this in the future. Thanks.

Quoted 11 lines

Also, do

$ git grep 'This patch' Documentation/process

for more details.

For this particular one, use scattered.c instead of adding a new leaf.

Thx.

author
#22 lore raw

PATCH 5/6 x86: KVM: Enable AVX-NE-CONVERT CPUID and expose it to guest

In reply to Jiaxi Chen

AVX-NE-CONVERT is a new set of instructions in the latest Intel platform Sierra Forest, which can convert low precision floating point like BF16/FP16 to high precision floating point FP32. It can also convert FP32 elements to BF16. This instruction allows the platform to have improved AI capabilities and better compatibility.

The bit definition:
CPUID.(EAX=7,ECX=1):EDX[bit 5]

This patch enables this CPUID in the kernel feature bits and expose it to guest OS.

---
 arch/x86/include/asm/cpufeatures.h | 1 +
 arch/x86/kvm/cpuid.c               | 3 ++-
 2 files changed, 3 insertions(+), 1 deletion(-)
arch/x86/include/asm/cpufeatures.h +1 -0
index b2aa761ea110..4e0bbffedbb4 100644
@@ -425,6 +425,7 @@
 
 /* Intel-defined CPU features, CPUID level 0x00000007:1 (EDX), word 20 */
 #define X86_FEATURE_AVX_VNNI_INT8       (20*32+ 4) /* Support for VPDPB[SU,UU,SS]D[,S] */
+#define X86_FEATURE_AVX_NE_CONVERT      (20*32+ 5) /* AVX NE CONVERT Instructions */
 
 /*
  * BUG word(s)
arch/x86/kvm/cpuid.c +2 -1
index b1b53a5c788a..fcd00c68e546 100644
@@ -661,7 +661,8 @@ void kvm_set_cpu_caps(void)
 		F(AVX_IFMA));
 
 	kvm_cpu_cap_mask(CPUID_7_1_EDX,
-		F(AVX_VNNI_INT8));
+		F(AVX_VNNI_INT8) | F(AVX_NE_CONVERT)
+	);
 
 	kvm_cpu_cap_mask(CPUID_D_1_EAX,
 		F(XSAVEOPT) | F(XSAVEC) | F(XGETBV1) | F(XSAVES) | f_xfd
Signature · 1 line
2.27.0
author
#23 lore raw

PATCH 6/6 x86: KVM: Enable PREFETCHIT0/1 CPUID and expose it to guest

In reply to Jiaxi Chen

Latest Intel platform Granite Rapids has introduced a new instruction - PREFETCHIT0/1, which moves code to memory (cache) closer to the processor depending on specific hints.

The bit definition:
CPUID.(EAX=7,ECX=1):EDX[bit 14]

This patch enables this CPUID in the kernel feature bits and expose it to guest OS.

---
 arch/x86/include/asm/cpufeatures.h | 1 +
 arch/x86/kvm/cpuid.c               | 2 +-
 2 files changed, 2 insertions(+), 1 deletion(-)
arch/x86/include/asm/cpufeatures.h +1 -0
index 4e0bbffedbb4..de8aa62bbdb1 100644
@@ -426,6 +426,7 @@
 /* Intel-defined CPU features, CPUID level 0x00000007:1 (EDX), word 20 */
 #define X86_FEATURE_AVX_VNNI_INT8       (20*32+ 4) /* Support for VPDPB[SU,UU,SS]D[,S] */
 #define X86_FEATURE_AVX_NE_CONVERT      (20*32+ 5) /* AVX NE CONVERT Instructions */
+#define X86_FEATURE_PREFETCHITI         (20*32+14) /* PREFETCHIT0/1 Instructions */
 
 /*
  * BUG word(s)
arch/x86/kvm/cpuid.c +1 -1
index fcd00c68e546..e73307732d10 100644
@@ -661,7 +661,7 @@ void kvm_set_cpu_caps(void)
 		F(AVX_IFMA));
 
 	kvm_cpu_cap_mask(CPUID_7_1_EDX,
-		F(AVX_VNNI_INT8) | F(AVX_NE_CONVERT)
+		F(AVX_VNNI_INT8) | F(AVX_NE_CONVERT) | F(PREFETCHITI)
 	);
 
 	kvm_cpu_cap_mask(CPUID_D_1_EAX,
Signature · 1 line
2.27.0