DMARC-Filter: OpenDMARC Filter v1.3.2 mx1.redhat.com 0039144825
From: Bandan Das <bsd@redhat.com>
To: Radim =?utf-8?B?S3LEjW3DocWZ?= <rkrcmar@redhat.com>
Cc: kvm@vger.kernel.org, pbonzini@redhat.com, david@redhat.com,
        jmattson@google.com, linux-kernel@vger.kernel.org
Subject: Re: [PATCH v5 3/3] KVM: nVMX: Emulate EPTP switching for the L1 hypervisor
References: <20170728195245.1018-1-bsd@redhat.com>
        <20170728195245.1018-4-bsd@redhat.com> <20170801151720.GC302@flask>
Date: Tue, 01 Aug 2017 14:30:30 -0400
In-Reply-To: <20170801151720.GC302@flask> ("Radim \=\?utf-8\?B\?S3LEjW3DocWZ\?\=
 \=\?utf-8\?B\?Iidz\?\= message of "Tue, 1
        Aug 2017 17:17:21 +0200")
Message-ID: <jpgtw1ruomx.fsf@linux.bootlegged.copy>
User-Agent: Gnus/5.13 (Gnus v5.13) Emacs/25.2 (gnu/linux)
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Transfer-Encoding: 8BIT
Sender: linux-kernel-owner@vger.kernel.org
Content-Length: 2559
Lines: 76

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;