Re: [RFC PATCH v2 10/20] x86: Insure that memory areas are encrypted when possible

From: Tom Lendacky
Date: Wed Sep 14 2016 - 10:11:39 EST

On 09/12/2016 11:33 AM, Borislav Petkov wrote:
> On Mon, Sep 12, 2016 at 10:05:36AM -0500, Tom Lendacky wrote:
>> I can look into that. The reason I put this here is this is all the
>> early page fault support that is very specific to this file. I modified
>> an existing static function to take advantage of the mapping support.
> Yeah, but all this code is SME-specific and doesn't belong there.
> AFAICT, it uses global/public symbols so there shouldn't be a problem to
> have it in mem_encrypt.c.

Ok, I'll look into moving this into mem_encrypt.c. I'd like to avoid
duplicating code so I may have to make that static function external
unless I find a better way.


>> Hmmm, maybe... With the change to the early_memremap() the initrd is now
>> identified as BOOT_DATA in relocate_initrd() and so it will be mapped
>> and copied as non-encyrpted data. But since it was encrypted before the
>> call to relocate_initrd() it will copy encrypted bytes which will later
>> be accessed encrypted. That isn't clear though, so I'll rework
>> reserve_initrd() to perform the sme_early_mem_enc() once at the end
>> whether the initrd is re-located or not.
> Makes sense.