Re: [Qemu-devel] [QEMU PATCH v3 7/9] KVM: i386: Add support for save and restore nested state

[Liran Alon]

> On 18 Jun 2019, at 18:48, Dr. David Alan Gilbert <address@hidden> wrote:
> 
> * Liran Alon (address@hidden) wrote:
>> 
>>> On 18 Jun 2019, at 12:03, Dr. David Alan Gilbert <address@hidden> wrote:
>>> 
>>> * Liran Alon (address@hidden) wrote:
>>>> 
>>>> +static const VMStateDescription vmstate_vmx_vmcs12 = {
>>>> +  .name = "cpu/kvm_nested_state/vmx/vmcs12",
>>>> +  .version_id = 1,
>>>> +  .minimum_version_id = 1,
>>>> +  .needed = vmx_vmcs12_needed,
>>>> +  .fields = (VMStateField[]) {
>>>> +      VMSTATE_UINT8_ARRAY(data.vmx[0].vmcs12,
>>>> +                          struct kvm_nested_state, 0x1000),
>>> 
>>> Where did that magic 0x1000 come from?
>> 
>> Currently, KVM folks (including myself), haven’t decided yet to expose 
>> vmcs12 struct layout to userspace but instead to still leave it opaque.
>> The formal size of this size is VMCS12_SIZE (defined in kernel as 0x1000). I 
>> was wondering if we wish to expose VMCS12_SIZE constant to userspace or not.
>> So currently I defined these __u8 arrays as 0x1000. But in case Paolo agrees 
>> to expose VMCS12_SIZE, we can use that instead.
> 
> Well if it's not defined it's bound to change at some state!

I agree it’s better to expose VMCS12_SIZE to userspace but I didn’t want to be 
the one that decides this.
Let’s let Paolo decide and modify this patch accordingly if he decides to 
expose it.

> Also, do we need to clear it before we get it from the kernel - e.g.
> is the kernel guaranteed to give us 0x1000 ?

Userspace don’t need to clear it before getting it from kernel.
It does guarantee to give you 0x1000. See vmx_get_nested_state() implementation 
in kernel.

-Liran

> 
> Dave
> 
>> -Liran
>> 
>>> --
>>> Dr. David Alan Gilbert / address@hidden / Manchester, UK
>> 
> --
> Dr. David Alan Gilbert / address@hidden / Manchester, UK