On 2/12/19 9:06 AM, Paolo Bonzini wrote:
> On 12/02/19 15:44, Singh, Brijesh wrote:
>> - if (unlikely(insn && !insn_len))
>> - return 1;
>> + if (unlikely(insn && !insn_len)) {
>> + if (!kvm_x86_ops->emulate_instruction_possible(vcpu))
>> + return 1;
>> + }
>
> Are the instruction bytes valid, that is can we just ignore insn_len and
> use the bytes but not the length? That would work for SEV too.
>
The instruction bytes are not valid so we will not able to workaround
for the SEV.
> Paolo
>
>> diff --git a/arch/x86/kvm/vmx/vmx.c b/arch/x86/kvm/vmx/vmx.c
>> index 95d618045001..6767bad8367e 100644
>> --- a/arch/x86/kvm/vmx/vmx.c
>> +++ b/arch/x86/kvm/vmx/vmx.c
>> @@ -7530,6 +7530,11 @@ static int enable_smi_window(struct kvm_vcpu *vcpu)
>> return 0;
>> }
>>
>> +static bool emulate_instruction_possible(struct kvm_vcpu *vcpu)
>> +{
>> + return 1;
>
> This should be "return 0;" to keep previous behavior.
>
Sure, I will fix in v2.
>> +}
>> +
>> static __init int hardware_setup(void)
>> {
>> unsigned long host_bndcfgs;
>> @@ -7832,6 +7837,7 @@ static struct kvm_x86_ops vmx_x86_ops __ro_after_init = {
>> .set_nested_state = NULL,
>> .get_vmcs12_pages = NULL,
>> .nested_enable_evmcs = NULL,
>> + .emulate_instruction_possible = emulate_instruction_possible,
>> };
>>
>
