[v2] SVM guest shadow stack support

[PATCH v2 9/9] KVM: SVM: Add CET features to supported_xss

Posted by John Allen 1 year, 11 months ago

If the CPU supports CET, add CET XSAVES feature bits to the
supported_xss mask.

Signed-off-by: John Allen <john.allen@amd.com>
---
 arch/x86/kvm/svm/svm.c | 4 ++++
 1 file changed, 4 insertions(+)

diff --git a/arch/x86/kvm/svm/svm.c b/arch/x86/kvm/svm/svm.c
index 1181f017c173..d97d82ebec4a 100644
--- a/arch/x86/kvm/svm/svm.c
+++ b/arch/x86/kvm/svm/svm.c
@@ -5177,6 +5177,10 @@ static __init void svm_set_cpu_caps(void)
 	    boot_cpu_has(X86_FEATURE_AMD_SSBD))
 		kvm_cpu_cap_set(X86_FEATURE_VIRT_SSBD);
 
+	if (kvm_cpu_cap_has(X86_FEATURE_SHSTK))
+		kvm_caps.supported_xss |= XFEATURE_MASK_CET_USER |
+					  XFEATURE_MASK_CET_KERNEL;
+
 	if (enable_pmu) {
 		/*
 		 * Enumerate support for PERFCTR_CORE if and only if KVM has
-- 
2.40.1

Re: [PATCH v2 9/9] KVM: SVM: Add CET features to supported_xss

Posted by Sean Christopherson 1 year, 9 months ago

On Mon, Feb 26, 2024, John Allen wrote:
> If the CPU supports CET, add CET XSAVES feature bits to the
> supported_xss mask.
> 
> Signed-off-by: John Allen <john.allen@amd.com>
> ---
>  arch/x86/kvm/svm/svm.c | 4 ++++
>  1 file changed, 4 insertions(+)
> 
> diff --git a/arch/x86/kvm/svm/svm.c b/arch/x86/kvm/svm/svm.c
> index 1181f017c173..d97d82ebec4a 100644
> --- a/arch/x86/kvm/svm/svm.c
> +++ b/arch/x86/kvm/svm/svm.c
> @@ -5177,6 +5177,10 @@ static __init void svm_set_cpu_caps(void)
>  	    boot_cpu_has(X86_FEATURE_AMD_SSBD))
>  		kvm_cpu_cap_set(X86_FEATURE_VIRT_SSBD);
>  
> +	if (kvm_cpu_cap_has(X86_FEATURE_SHSTK))
> +		kvm_caps.supported_xss |= XFEATURE_MASK_CET_USER |
> +					  XFEATURE_MASK_CET_KERNEL;

Based on Weijiang's series, I believe this is unnecessary.  Common x86 code will
both set supported_xss, and clear bits if their associated features are unsupported.

I also asked Weijiang to modify the "advertise to userspace" patch to explicitly
clear SHSTK and IBT in svm_set_cpu_caps()[*], so if the stars align as I think they
will, this patch should simply need to delete the

	kvm_cpu_cap_clear(X86_FEATURE_SHSTK);

that will be added by the VMX series.

[*] https://lore.kernel.org/all/ZjLRnisdUgeYgg8i@google.com