[PATCH v3 14/14] ARM64: KVM: fix big endian issue in access_vm_reg for 32bit guest

[Marc Zyngier]

On 28/05/14 15:09, Christoffer Dall wrote:
> On Wed, May 28, 2014 at 06:56:44AM -0700, Victor Kamensky wrote:
>> On 28 May 2014 02:14, Christoffer Dall <christoffer.dall at linaro.org> wrote:
>>
>> They cannot be separate it is the same data with two ways
>> to access: as 64bit value or low word 32bit value. Typically it is LPAE
>> cp15 memory related registers like TTBR0, as in mcrr, mcr example
>> in my previous response. mcrr will update high and low word values
>> of TTBR0, mcr will update only low word of it.
>>
> You just have to decide whether you want to represent a 64-bit register as
> a concatenation of two u32's or a 32-bit register access as a part
> of the 64-bit register value.
> 
> I think the architecture, at least for 32-bit register views on ARMv8,
> defines the 32-bit accesses as a view on the 64-bit register.

Indeed, and that's how KVM implements it.

> You should of course only store those registers once, but instead of
> storing them in an array of u32's you could store the 64-bit wide
> registers in a separate array of u64's.  32-bit accesses to those
> registers would look in the array of u64's for such a value.
> 
> I didn't try it out, so not sure how it looks like, just saying it's an
> option worth considering.

I think that's the best idea so far. If the access is 64bit, use the
64bit version of the union (going through the corresponding
vcpu_sys_reg() interface).

This should be a very minimal refactor.

	M.
-- 
Jazz is not dead. It just smells funny...