Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]
Groups > linux.kernel > #1701142 > unrolled thread
| Started by | Radim Krčmář <rkrcmar@redhat.com> |
|---|---|
| First post | 2017-08-01 17:20 +0200 |
| Last post | 2017-08-01 20:40 +0200 |
| Articles | 2 — 2 participants |
Back to article view | Back to linux.kernel
This discussion starts older than the indexed window; earlier articles aren't shown. The article labeled Started by
below is the oldest one visible, not the original post.
Re: [PATCH v5 3/3] KVM: nVMX: Emulate EPTP switching for the L1 hypervisor Radim Krčmář <rkrcmar@redhat.com> - 2017-08-01 17:20 +0200
Re: [PATCH v5 3/3] KVM: nVMX: Emulate EPTP switching for the L1 hypervisor Bandan Das <bsd@redhat.com> - 2017-08-01 20:40 +0200
| From | Radim Krčmář <rkrcmar@redhat.com> |
|---|---|
| Date | 2017-08-01 17:20 +0200 |
| Subject | Re: [PATCH v5 3/3] KVM: nVMX: Emulate EPTP switching for the L1 hypervisor |
| Message-ID | <u9H7H-5CS-7@gated-at.bofh.it> |
2017-07-28 15:52-0400, Bandan Das:
> When L2 uses vmfunc, L0 utilizes the associated vmexit to
> emulate a switching of the ept pointer by reloading the
> guest MMU.
>
> Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
> Signed-off-by: Bandan Das <bsd@redhat.com>
> ---
> diff --git a/arch/x86/kvm/vmx.c b/arch/x86/kvm/vmx.c
> @@ -7767,6 +7781,85 @@ static int handle_preemption_timer(struct kvm_vcpu *vcpu)
> return 1;
> }
>
> +static bool check_ept_address_valid(struct kvm_vcpu *vcpu, u64 address)
> +{
> + struct vcpu_vmx *vmx = to_vmx(vcpu);
> + u64 mask = VMX_EPT_RWX_MASK;
> + int maxphyaddr = cpuid_maxphyaddr(vcpu);
> + struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
> +
> + /* Check for execute_only validity */
> + if ((address & mask) == VMX_EPT_EXECUTABLE_MASK) {
> + if (!(vmx->nested.nested_vmx_ept_caps &
> + VMX_EPT_EXECUTE_ONLY_BIT))
> + return false;
> + }
This checks looks wrong ... bits 0:2 define the memory type:
0 = Uncacheable (UC)
6 = Write-back (WB)
If those are supported MSR IA32_VMX_EPT_VPID_CAP, so I think it should
return false when
(address & 0x7) == 0 && !(vmx->nested.nested_vmx_ept_caps & VMX_EPTP_UC_BIT))
the same for 6 and VMX_EPTP_WB_BIT and unconditionally for the remaining
types.
Btw. when is TLB flushed after EPTP switching?
> @@ -10354,10 +10456,20 @@ static int check_vmentry_prereqs(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
> vmx->nested.nested_vmx_entry_ctls_high))
> return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
>
> - if (nested_cpu_has_vmfunc(vmcs12) &&
> - (vmcs12->vm_function_control &
> - ~vmx->nested.nested_vmx_vmfunc_controls))
> - return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
> + if (nested_cpu_has_vmfunc(vmcs12)) {
> + if (vmcs12->vm_function_control &
> + ~vmx->nested.nested_vmx_vmfunc_controls)
> + return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
> +
> + if (nested_cpu_has_eptp_switching(vmcs12)) {
> + if (!nested_cpu_has_ept(vmcs12) ||
> + (vmcs12->eptp_list_address >>
> + cpuid_maxphyaddr(vcpu)) ||
> + !IS_ALIGNED(vmcs12->eptp_list_address, 4096))
page_address_valid() would make this check a bit nicer,
thanks.
> + return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
[toc] | [next] | [standalone]
| From | Bandan Das <bsd@redhat.com> |
|---|---|
| Date | 2017-08-01 20:40 +0200 |
| Subject | Re: [PATCH v5 3/3] KVM: nVMX: Emulate EPTP switching for the L1 hypervisor |
| Message-ID | <u9Kff-7x4-1@gated-at.bofh.it> |
| In reply to | #1701142 |
Radim Krčmář <rkrcmar@redhat.com> writes:
> 2017-07-28 15:52-0400, Bandan Das:
>> When L2 uses vmfunc, L0 utilizes the associated vmexit to
>> emulate a switching of the ept pointer by reloading the
>> guest MMU.
>>
>> Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
>> Signed-off-by: Bandan Das <bsd@redhat.com>
>> ---
>> diff --git a/arch/x86/kvm/vmx.c b/arch/x86/kvm/vmx.c
>> @@ -7767,6 +7781,85 @@ static int handle_preemption_timer(struct kvm_vcpu *vcpu)
>> return 1;
>> }
>>
>> +static bool check_ept_address_valid(struct kvm_vcpu *vcpu, u64 address)
>> +{
>> + struct vcpu_vmx *vmx = to_vmx(vcpu);
>> + u64 mask = VMX_EPT_RWX_MASK;
>> + int maxphyaddr = cpuid_maxphyaddr(vcpu);
>> + struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
>> +
>> + /* Check for execute_only validity */
>> + if ((address & mask) == VMX_EPT_EXECUTABLE_MASK) {
>> + if (!(vmx->nested.nested_vmx_ept_caps &
>> + VMX_EPT_EXECUTE_ONLY_BIT))
>> + return false;
>> + }
>
> This checks looks wrong ... bits 0:2 define the memory type:
>
> 0 = Uncacheable (UC)
> 6 = Write-back (WB)
Oops, sorry, I badly messed this up! I will incorporate these
changes and the suggestions by David to a new version.
> If those are supported MSR IA32_VMX_EPT_VPID_CAP, so I think it should
> return false when
>
> (address & 0x7) == 0 && !(vmx->nested.nested_vmx_ept_caps & VMX_EPTP_UC_BIT))
>
> the same for 6 and VMX_EPTP_WB_BIT and unconditionally for the remaining
> types.
>
> Btw. when is TLB flushed after EPTP switching?
From what I understand, mmu_sync_roots() calls kvm_mmu_flush_or_zap()
that sets KVM_REQ_TLB_FLUSH.
Bandan
>> @@ -10354,10 +10456,20 @@ static int check_vmentry_prereqs(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
>> vmx->nested.nested_vmx_entry_ctls_high))
>> return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
>>
>> - if (nested_cpu_has_vmfunc(vmcs12) &&
>> - (vmcs12->vm_function_control &
>> - ~vmx->nested.nested_vmx_vmfunc_controls))
>> - return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
>> + if (nested_cpu_has_vmfunc(vmcs12)) {
>> + if (vmcs12->vm_function_control &
>> + ~vmx->nested.nested_vmx_vmfunc_controls)
>> + return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
>> +
>> + if (nested_cpu_has_eptp_switching(vmcs12)) {
>> + if (!nested_cpu_has_ept(vmcs12) ||
>> + (vmcs12->eptp_list_address >>
>> + cpuid_maxphyaddr(vcpu)) ||
>> + !IS_ALIGNED(vmcs12->eptp_list_address, 4096))
>
> page_address_valid() would make this check a bit nicer,
>
> thanks.
>
>> + return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
[toc] | [prev] | [standalone]
Back to top | Article view | linux.kernel
csiph-web