Re: [edk2-devel] SecCore evacuation in PeiCore?

["Michael Kubacki" <mikuback@linux.microsoft.com>]

Hi Marvin,

Your understanding of SecMigrationPei is correct. It is not ideal as 
it's an unfamiliar pattern that could give the false impression that it 
is a universal SEC migration solution, which it is not. But if platforms 
understand that any additional data published in SecCore must be 
explicitly migrated (potentially via library extension to 
SecMigrationPei), it can be used to serve the SEC post-memory migration 
role.

I assumed it was related to the reset vector due to the 16-bit 
alignment. I think it would be great to have SecCore aligned properly if 
possible.

Thanks,
Michael

On 8/14/2021 8:29 AM, Marvin Häuser wrote:
> Hey Michael,
>
> Thank you for your response! It was actually quicker than I imagined. :)
>
> I think I understand, but please let me try to get this absolutely 
> right. Can I think of "SecMigrationPei" as a sort of "SecCorePostMem", 
> which either is loaded into permanent RAM directly or is shadowed 
> because it is a PEIM unlike SecCore - and it republishes all public 
> data, most especially PPIs, such that the entire PEI stage no longer 
> has any references to the original SecCore at all, and the SecCore 
> module basically just sits there in the ROM, and its exposed data is 
> either discarded or orphaned? Is that about right?
>
> I think I hit the alignment issue of SecCore too, but only for X64 
> builds (likely just because the size happens to be lucky for IA32) of 
> OVMF. Pretty much sure it's just ResetVector positioning. What would 
> be the issue with moving the ResetVector into a separate component, 
> with its fixed position in FD (this is actually how UefiCpuPkg/VTF0 
> works), and having SecCore aligned correctly? Not specifically to 
> restore MigrateSecModulesInFv(), but as future-proofing to ensure 
> expected outputs. In fact, I noticed because my new PE loader code was 
> upset about the unaligned XIP load address.
>
> Also thanks for your patch!
>
> Best regards,
> Marvin
>
> On 13/08/2021 18:51, Michael Kubacki wrote:
>> Hi Marvin,
>>
>> I apologize for the delayed response, I missed this message earlier. 
>> The function was called from EvacuateTempRam() in the initial set of 
>> patches:
>> [PATCH 1/6] MdeModulePkg/PeiCore: Enable T-RAM evacuation in PeiCore 
>> (CVE-2019-11098) (groups.io) 
>> <https://edk2.groups.io/g/devel/message/61823>
>>
>> I was not involved in the patch series on the mailing list (job role 
>> change at the time) but as a comment in that patch notes, there was 
>> an inconsistency observed in PE32 section alignment in SEC modules. I 
>> don't see where this was resolved other than the calls being removed 
>> later in the series. SecCore migration would not occur implicitly in 
>> the PeiCore flow but there is functionality for SEC data migration in 
>> UefiCpuPkg/SecMigrationPei.
>>
>> Based on what I see now, I'd be happy to send a patch to remove 
>> MigrateSecModulesInFv().
>>
>> Thanks,
>> Michael
>>
>> On 8/7/2021 2:54 PM, Marvin Häuser wrote:
>>> Good day everyone,
>>> Good day Michael,
>>>
>>> The commit that introduced T-RAM evacuation [1] also introduced the 
>>> function "MigrateSecModulesInFv()". It also is explicitly mentioned 
>>> as part of the control flow in the commit message. As far as I can 
>>> see, since then till today this function has never been called 
>>> anywhere. Was this some draft function that accidentally made it 
>>> into the patch, or did the caller get lost somewhere? The 
>>> description makes sense to me and I'm not experienced enough with 
>>> the PeiCore control flow to tell whether the PEIM migration somehow 
>>> covers SecCore implicitly. Also I noticed it only supports SecCore 
>>> in a PE/COFF section, not a TE section. Is there a rationale for that?
>>>
>>> Thank you for your time!
>>>
>>> Best regards,
>>> Marvin
>>>
>>>
>>> [1] 
>>> https://github.com/tianocore/edk2/commit/9bedaec05b7b8ba9aee248361bb61a85a26726cb
>>>
>>>
>>> 
>>>
>>