Re: [edk2-devel] [PATCH v7 0/7] Add safe unaccepted memory behavior

["Alexey Kardashevskiy" <aik@ozlabs.ru>]

On 26/10/2022 12:07, Dionna Amalie Glaze wrote:
> On Tue, Oct 25, 2022 at 5:23 PM Alexey Kardashevskiy <aik@ozlabs.ru> wrote:
>>
>> Hi Dionna,
>>
>> Thanks for updating the tree, builds nicely now! However the VM's kernel
>> does not boot - the guest kernel reports
>>
>> EFI stub: ERROR: exit_boot() failed!
>>
>> and hangs. I am not quite sure how it is supposed to work (still
>> learning) but "Accepting all memory" happens twice (should it?) and the
>> actual reason for the CoreExitBootService() failure is that MapKey !=
>> mMemoryMapKey in CoreTerminateMemoryMap(), these are 0x7AD7 vs 0x7AD9
>> or  0x7AE1 vs 0x7AE3 (the diff is always 2).
>>
> 
> "Accepting all memory" may happen twice, but it's idempotent. The
> debug_info log happening twice might be confusing, so I can change
> that if you'd like.

Nah, it is fine as long as the thing boots.


> The first accept will remove all unaccepted memory regions from the
> address space map.
> CoreExitBootService should fail the first time since the first accept
> will change the memory map.
> That failure means that the caller should GetMemoryMap again and try
> CoreExitBootService again.

Ah, ok.

>> How do you test it exactly, is there any command line change needed in
>> addition to enabling SNP?
>>
>> My guest kernel uses
>> https://patchew.org/linux/cover.1664298261.git.thomas.lendacky@amd.com/
>> with the TDX prerequisite. Thanks,
>>
> 
> It's a few name changes behind, but this branch of Linux is what I've
> been using:
> 
> https://github.com/deeglaze/amdese-linux/tree/v12unacceptedv7v5-enableum

This does not build though due to unresolved rebase conflict, the fix is 
kinda trivial but I do not get those "Accepting all memory" anymore so I 
wonder what else is missing.

> 
> Specific enablement patch here
> https://github.com/AMDESE/linux/commit/5a708081d58d773e767b11735ee1fd17ef5e5f61
> 
> I incorporate Kirill's patch set v7 for basic unaccepted memory
> support, Tom Lendacky's v5 patch set for SEV-SNP support of unaccepted
> memory, and I have a single patch that calls the protocol.
> This branch doesn't have Kirill's TDX patches.
> I've run it with a regular SEV-SNP enabled guest kernel too. At this
> point all the tests have used kernel injection rather than having the
> kernels all baked into the image.

I am trying with "-m 2G" and "-m 8G" and see no difference - "Accepting 
all memory" is not appearing with this kernel. What do I miss? Thanks,


-- 
Alexey