Re: [PATCH v4 2/4] x86/boot: Refactor BIOS/PVH start

[Jan Beulich <jbeulich@xxxxxxxx>]

On 26.09.2024 11:24, Frediano Ziglio wrote:
> On Thu, Sep 26, 2024 at 7:50 AM Jan Beulich <jbeulich@xxxxxxxx> wrote:
>>
>> On 25.09.2024 21:33, Andrew Cooper wrote:
>>> On 25/09/2024 7:00 am, Frediano Ziglio wrote:
>>>> @@ -449,62 +458,40 @@ __pvh_start:
>>>>          mov     %ecx, %es
>>>>          mov     %ecx, %ss
>>>>
>>>> -        /* Skip bootloader setup and bios setup, go straight to 
>>>> trampoline */
>>>> -        movb    $1, sym_esi(pvh_boot)
>>>> -        movb    $1, sym_esi(skip_realmode)
>>>> +        /* Load null selector to unused segment registers. */
>>>> +        xor     %ecx, %ecx
>>>> +        mov     %ecx, %fs
>>>> +        mov     %ecx, %gs
>>>
>>> Honestly, the more I look at this, the more bizarre it is.
>>>
>>> We should just set up %fs/gs like we do %ds/es, which in this case is
>>> simply to drop the comment and the xor.
>>
>> What's bizarre here? As long as we don't use %fs/%gs, it doesn't matter
>> much what we set them to. So yes, they may be set to what %ds etc are set
>> to, but they may as well be marked unusable. Documentation-wise that's
>> cleaner imo, as down the road - when a need to use one arises - it then
>> won't require auditing of all code to figure where the register(s) is(are)
>> actually used (just to find: nowhere). Even if a comment to this effect
>> was left here, I for one wouldn't trust it in a couple of years time, but
>> rather fear it went stale.
> 
> Hi,
>   are you against this change and asking to roll it back?

Well, first of all I'm hoping for a reply by Andrew. Maybe I'm overlooking
something. As it stands right now I indeed think we'd be better off keeping
nul selectors in %fs and %gs.

Jan