[PATCH 0/3] KVM: arm64: Fixes for 'writable' ID registers in protected mode

[PATCH 0/3] KVM: arm64: Fixes for 'writable' ID registers in protected mode

[Oliver Upton]

Address a few issues with ID register handling in protected mode:

 - CTR_EL0 isn't trapped on FEAT_EVT systems if userspace changed the
   guest value

 - VPIDR_EL2 is set to 0 if userspace enabled writable 'implementation
   ID' registers

 - Fix in capability documentation

Oliver Upton (3):
  KVM: arm64: Copy guest CTR_EL0 into hyp VM
  KVM: arm64: Copy MIDR_EL1 into hyp VM when it is writable
  KVM: arm64: Fix documentation for KVM_CAP_ARM_WRITABLE_IMP_ID_REGS

 Documentation/virt/kvm/api.rst | 2 +-
 arch/arm64/kvm/hyp/nvhe/pkvm.c | 7 ++++++-
 2 files changed, 7 insertions(+), 2 deletions(-)


base-commit: 164f95739b6893716add9439cf765edd5254d0f6
-- 
2.39.5

[PATCH 1/3] KVM: arm64: Copy guest CTR_EL0 into hyp VM

[Oliver Upton]

Since commit 2843cae26644 ("KVM: arm64: Treat CTR_EL0 as a VM feature
ID register") KVM has allowed userspace to configure the VM-wide view of
CTR_EL0, falling back to trap-n-emulate if the value doesn't match
hardware. It appears that this has worked by chance in protected-mode
for some time, and on systems with FEAT_EVT protected-mode
unconditionally sets TID4 (i.e. TID2 traps sans CTR_EL0).

Forward the guest CTR_EL0 value through to the hyp VM and align the
TID2/TID4 configuration with the non-protected setup.

Fixes: 2843cae26644 ("KVM: arm64: Treat CTR_EL0 as a VM feature ID register")
Signed-off-by: Oliver Upton <oliver.upton@linux.dev>
---
 arch/arm64/kvm/hyp/nvhe/pkvm.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

diff --git a/arch/arm64/kvm/hyp/nvhe/pkvm.c b/arch/arm64/kvm/hyp/nvhe/pkvm.c
index 3927fe52a3dd..2e2975248cd3 100644
--- a/arch/arm64/kvm/hyp/nvhe/pkvm.c
+++ b/arch/arm64/kvm/hyp/nvhe/pkvm.c
@@ -46,7 +46,8 @@ static void pkvm_vcpu_reset_hcr(struct kvm_vcpu *vcpu)
 		vcpu->arch.hcr_el2 |= HCR_FWB;
 
 	if (cpus_have_final_cap(ARM64_HAS_EVT) &&
-	    !cpus_have_final_cap(ARM64_MISMATCHED_CACHE_TYPE))
+	    !cpus_have_final_cap(ARM64_MISMATCHED_CACHE_TYPE) &&
+	    kvm_read_vm_id_reg(vcpu->kvm, SYS_CTR_EL0) == read_cpuid(CTR_EL0))
 		vcpu->arch.hcr_el2 |= HCR_TID4;
 	else
 		vcpu->arch.hcr_el2 |= HCR_TID2;
@@ -374,6 +375,7 @@ static void init_pkvm_hyp_vm(struct kvm *host_kvm, struct pkvm_hyp_vm *hyp_vm,
 	hyp_vm->kvm.arch.pkvm.enabled = READ_ONCE(host_kvm->arch.pkvm.enabled);
 	hyp_vm->kvm.arch.flags = 0;
 	pkvm_init_features_from_host(hyp_vm, host_kvm);
+	hyp_vm->kvm.arch.ctr_el0 = host_kvm->arch.ctr_el0;
 }
 
 static void pkvm_vcpu_init_sve(struct pkvm_hyp_vcpu *hyp_vcpu, struct kvm_vcpu *host_vcpu)
-- 
2.39.5

[PATCH 2/3] KVM: arm64: Copy MIDR_EL1 into hyp VM when it is writable

[Oliver Upton]

KVM recently added a capability that allows userspace to override the
'implementation ID' registers presented to the VM. MIDR_EL1 is a special
example, where the hypervisor can directly set the value when read from
EL1 using VPIDR_EL2.

Copy the VM-wide value for MIDR_EL1 into the hyp VM for non-protected
guests when the capability is enabled so VPIDR_EL2 gets set up
correctly.

Reported-by: Mark Brown <broonie@kernel.org>
Closes: https://lore.kernel.org/kvmarm/ac594b9c-4bbb-46c8-9391-e7a68ce4de5b@sirena.org.uk/
Fixes: 3adaee783061 ("KVM: arm64: Allow userspace to change the implementation ID registers")
Signed-off-by: Oliver Upton <oliver.upton@linux.dev>
---
 arch/arm64/kvm/hyp/nvhe/pkvm.c | 3 +++
 1 file changed, 3 insertions(+)

diff --git a/arch/arm64/kvm/hyp/nvhe/pkvm.c b/arch/arm64/kvm/hyp/nvhe/pkvm.c
index 2e2975248cd3..afc1b5c403c9 100644
--- a/arch/arm64/kvm/hyp/nvhe/pkvm.c
+++ b/arch/arm64/kvm/hyp/nvhe/pkvm.c
@@ -376,6 +376,9 @@ static void init_pkvm_hyp_vm(struct kvm *host_kvm, struct pkvm_hyp_vm *hyp_vm,
 	hyp_vm->kvm.arch.flags = 0;
 	pkvm_init_features_from_host(hyp_vm, host_kvm);
 	hyp_vm->kvm.arch.ctr_el0 = host_kvm->arch.ctr_el0;
+
+	if (test_bit(KVM_ARCH_FLAG_WRITABLE_IMP_ID_REGS, &hyp_vm->kvm.arch.flags))
+		hyp_vm->kvm.arch.midr_el1 = host_kvm->arch.midr_el1;
 }
 
 static void pkvm_vcpu_init_sve(struct pkvm_hyp_vcpu *hyp_vcpu, struct kvm_vcpu *host_vcpu)
-- 
2.39.5

[PATCH 3/3] KVM: arm64: Fix documentation for KVM_CAP_ARM_WRITABLE_IMP_ID_REGS

[Oliver Upton]

The capability actually fails with EINVAL if vCPUs have already been
created.

Signed-off-by: Oliver Upton <oliver.upton@linux.dev>
---
 Documentation/virt/kvm/api.rst | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/Documentation/virt/kvm/api.rst b/Documentation/virt/kvm/api.rst
index 56f3dcdc4477..3468a2a7de6b 100644
--- a/Documentation/virt/kvm/api.rst
+++ b/Documentation/virt/kvm/api.rst
@@ -8264,7 +8264,7 @@ take care to differentiate between these cases.
 :Architectures: arm64
 :Target: VM
 :Parameters: None
-:Returns: 0 on success, -EBUSY if vCPUs have been created before enabling this
+:Returns: 0 on success, -EINVAL if vCPUs have been created before enabling this
           capability.
 
 This capability changes the behavior of the registers that identify a PE
-- 
2.39.5

Re: [PATCH 1/3] KVM: arm64: Copy guest CTR_EL0 into hyp VM

[Marc Zyngier]

On Wed, 05 Mar 2025 23:08:23 +0000,
Oliver Upton <oliver.upton@linux.dev> wrote:
> 
> Since commit 2843cae26644 ("KVM: arm64: Treat CTR_EL0 as a VM feature
> ID register") KVM has allowed userspace to configure the VM-wide view of
> CTR_EL0, falling back to trap-n-emulate if the value doesn't match
> hardware. It appears that this has worked by chance in protected-mode
> for some time, and on systems with FEAT_EVT protected-mode
> unconditionally sets TID4 (i.e. TID2 traps sans CTR_EL0).
> 
> Forward the guest CTR_EL0 value through to the hyp VM and align the
> TID2/TID4 configuration with the non-protected setup.
> 
> Fixes: 2843cae26644 ("KVM: arm64: Treat CTR_EL0 as a VM feature ID register")
> Signed-off-by: Oliver Upton <oliver.upton@linux.dev>
> ---
>  arch/arm64/kvm/hyp/nvhe/pkvm.c | 4 +++-
>  1 file changed, 3 insertions(+), 1 deletion(-)
> 
> diff --git a/arch/arm64/kvm/hyp/nvhe/pkvm.c b/arch/arm64/kvm/hyp/nvhe/pkvm.c
> index 3927fe52a3dd..2e2975248cd3 100644
> --- a/arch/arm64/kvm/hyp/nvhe/pkvm.c
> +++ b/arch/arm64/kvm/hyp/nvhe/pkvm.c
> @@ -46,7 +46,8 @@ static void pkvm_vcpu_reset_hcr(struct kvm_vcpu *vcpu)
>  		vcpu->arch.hcr_el2 |= HCR_FWB;
>  
>  	if (cpus_have_final_cap(ARM64_HAS_EVT) &&
> -	    !cpus_have_final_cap(ARM64_MISMATCHED_CACHE_TYPE))
> +	    !cpus_have_final_cap(ARM64_MISMATCHED_CACHE_TYPE) &&
> +	    kvm_read_vm_id_reg(vcpu->kvm, SYS_CTR_EL0) == read_cpuid(CTR_EL0))
>  		vcpu->arch.hcr_el2 |= HCR_TID4;
>  	else
>  		vcpu->arch.hcr_el2 |= HCR_TID2;
> @@ -374,6 +375,7 @@ static void init_pkvm_hyp_vm(struct kvm *host_kvm, struct pkvm_hyp_vm *hyp_vm,
>  	hyp_vm->kvm.arch.pkvm.enabled = READ_ONCE(host_kvm->arch.pkvm.enabled);
>  	hyp_vm->kvm.arch.flags = 0;
>  	pkvm_init_features_from_host(hyp_vm, host_kvm);
> +	hyp_vm->kvm.arch.ctr_el0 = host_kvm->arch.ctr_el0;

Maybe it'd be better to stick this into pkvm_init_features_from_host()
in order to keep all the "inherit stuff from host" together?

Same thing for the following patch.

Thanks,

	M.

-- 
Without deviation from the norm, progress is not possible.

Re: [PATCH 0/3] KVM: arm64: Fixes for 'writable' ID registers in protected mode

[Marc Zyngier]

On Wed, 05 Mar 2025 23:08:22 +0000,
Oliver Upton <oliver.upton@linux.dev> wrote:
> 
> Address a few issues with ID register handling in protected mode:
> 
>  - CTR_EL0 isn't trapped on FEAT_EVT systems if userspace changed the
>    guest value
> 
>  - VPIDR_EL2 is set to 0 if userspace enabled writable 'implementation
>    ID' registers
> 
>  - Fix in capability documentation
> 
> Oliver Upton (3):
>   KVM: arm64: Copy guest CTR_EL0 into hyp VM
>   KVM: arm64: Copy MIDR_EL1 into hyp VM when it is writable
>   KVM: arm64: Fix documentation for KVM_CAP_ARM_WRITABLE_IMP_ID_REGS
> 
>  Documentation/virt/kvm/api.rst | 2 +-
>  arch/arm64/kvm/hyp/nvhe/pkvm.c | 7 ++++++-
>  2 files changed, 7 insertions(+), 2 deletions(-)

Irrespective of the comments for the first two patches:

Reviewed-by: Marc Zyngier <maz@kernel.org>

	M.

-- 
Without deviation from the norm, progress is not possible.

Re: [PATCH 0/3] KVM: arm64: Fixes for 'writable' ID registers in protected mode

[Oliver Upton]

On Wed, 05 Mar 2025 15:08:22 -0800, Oliver Upton wrote:
> Address a few issues with ID register handling in protected mode:
> 
>  - CTR_EL0 isn't trapped on FEAT_EVT systems if userspace changed the
>    guest value
> 
>  - VPIDR_EL2 is set to 0 if userspace enabled writable 'implementation
>    ID' registers
> 
> [...]

Marc, I fixed it in post :)

Applied to next, thanks!

[1/3] KVM: arm64: Copy guest CTR_EL0 into hyp VM
      https://git.kernel.org/kvmarm/kvmarm/c/9d9122736433
[2/3] KVM: arm64: Copy MIDR_EL1 into hyp VM when it is writable
      https://git.kernel.org/kvmarm/kvmarm/c/03e1b89d051f
[3/3] KVM: arm64: Fix documentation for KVM_CAP_ARM_WRITABLE_IMP_ID_REGS
      https://git.kernel.org/kvmarm/kvmarm/c/5980a6937012

--
Best,
Oliver

Re: [PATCH 2/3] KVM: arm64: Copy MIDR_EL1 into hyp VM when it is writable

[Sebastian Ott]

On Wed, 5 Mar 2025, Oliver Upton wrote:
> KVM recently added a capability that allows userspace to override the
> 'implementation ID' registers presented to the VM. MIDR_EL1 is a special
> example, where the hypervisor can directly set the value when read from
> EL1 using VPIDR_EL2.
>
> Copy the VM-wide value for MIDR_EL1 into the hyp VM for non-protected
> guests when the capability is enabled so VPIDR_EL2 gets set up
> correctly.
>
> Reported-by: Mark Brown <broonie@kernel.org>
> Closes: https://lore.kernel.org/kvmarm/ac594b9c-4bbb-46c8-9391-e7a68ce4de5b@sirena.org.uk/
> Fixes: 3adaee783061 ("KVM: arm64: Allow userspace to change the implementation ID registers")
> Signed-off-by: Oliver Upton <oliver.upton@linux.dev>
> ---
> arch/arm64/kvm/hyp/nvhe/pkvm.c | 3 +++
> 1 file changed, 3 insertions(+)
>
> diff --git a/arch/arm64/kvm/hyp/nvhe/pkvm.c b/arch/arm64/kvm/hyp/nvhe/pkvm.c
> index 2e2975248cd3..afc1b5c403c9 100644
> --- a/arch/arm64/kvm/hyp/nvhe/pkvm.c
> +++ b/arch/arm64/kvm/hyp/nvhe/pkvm.c
> @@ -376,6 +376,9 @@ static void init_pkvm_hyp_vm(struct kvm *host_kvm, struct pkvm_hyp_vm *hyp_vm,
> 	hyp_vm->kvm.arch.flags = 0;
> 	pkvm_init_features_from_host(hyp_vm, host_kvm);
> 	hyp_vm->kvm.arch.ctr_el0 = host_kvm->arch.ctr_el0;
> +
> +	if (test_bit(KVM_ARCH_FLAG_WRITABLE_IMP_ID_REGS, &hyp_vm->kvm.arch.flags))
> +		hyp_vm->kvm.arch.midr_el1 = host_kvm->arch.midr_el1;
> }

Oh, I didn't even think that all that's missing was to update the trusted
kvm instance ... I was looking on where to set VPIDR_EL2 for pkvm.

Thanks for the fix, I guess I'll have some more code reading todo...

Sebastian