[PATCH v5 29/32] x86/mm: Add support to encrypt the kernel in-place

[Tom Lendacky]

On 5/26/2017 11:25 AM, Borislav Petkov wrote:
> On Thu, May 25, 2017 at 05:24:27PM -0500, Tom Lendacky wrote:
>> I guess I could do that, but this will probably only end up clearing a
>> single PGD entry anyway since it's highly doubtful the address range
>> would cross a 512GB boundary.
> 
> Or you can compute how many 512G-covering, i.e., PGD entries there are
> and clear just the right amnount. :^)
> 
>> I can change the name. As for the use of ENTRY... without the
>> ENTRY/ENDPROC combination I was receiving a warning about a return
>> instruction outside of a callable function. It looks like I can just
>> define the "sme_enc_routine:" label with the ENDPROC and the warning
>> goes away and the global is avoided. It doesn't like the local labels
>> (.L...) so I'll use the new name.
> 
> Is that warning from objtool or where does it come from?

Yes, it's from objtool:

arch/x86/mm/mem_encrypt_boot.o: warning: objtool: .text+0xd2: return 
instruction outside of a callable function

> 
> How do I trigger it locally

I think having CONFIG_STACK_VALIDATION=y will trigger it.

> 
>> The hardware will try to optimize rep movsb into large chunks assuming
>> things are aligned, sizes are large enough, etc. so we don't have to
>> explicitly specify and setup for a rep movsq.
> 
> I thought the hw does that for movsq too?

It does.

Thanks,
Tom

>