Re: [PATCH 2/3] x86/svm: Enumeration for CET

[Andrew Cooper <andrew.cooper3@xxxxxxxxxx>]

On 28/04/2021 10:14, Jan Beulich wrote:
> On 27.04.2021 19:47, Andrew Cooper wrote:
>> On 27/04/2021 16:53, Jan Beulich wrote:
>>> On 26.04.2021 19:54, Andrew Cooper wrote:
>>>> @@ -497,7 +501,9 @@ struct vmcb_struct {
>>>>      u64 rip;
>>>>      u64 res14[11];
>>>>      u64 rsp;
>>>> -    u64 res15[3];
>>>> +    u64 _msr_s_cet;             /* offset 0x400 + 0x1E0 - cleanbit 12 */
>>>> +    u64 _ssp;                   /* offset 0x400 + 0x1E8   | */
>>>> +    u64 _msr_isst;              /* offset 0x400 + 0x1F0   v */
>>>>      u64 rax;
>>>>      u64 star;
>>>>      u64 lstar;
>>> Any reason for the leading underscores, when none of the neighboring
>>> fields have such?
>> Yes - they're covered by a cleanbit, and for better or worse, this is
>> our style.
> The underscore prefixes are, to my understanding, there only to
> emphasize that the fields shouldn't be accessed directly, but ...
>
>>> Did you perhaps mean to add VMCB_ACCESSORS()
>>> instances for them?
>> TBH, I opencoded the cleanbit handling because I thoroughly hate that
>> entire infrastructure.
> ... via this (or something with similar abstracting effect). So
> for any fields you mean to access directly they imo shouldn't be
> there. I particularly don't view them as indicators of being
> covered by cleanbits (if the respective accessors aren't used).

The leading underscores are enforced by 'vmcb->_ ## name' in
VMCB_ACCESSORS().

The cleanbits are the only reason we can't treat this as a simple struct.

~Andrew