Hi Ray,
Your description is correct. I believe there are efforts trying to make Standalone MM launch in PEI happen? i.e. Add standalone mm ipl pei driver by hongbin123 · Pull Request #5236 · tianocore/edk2 (github.com) will bring in support for #1 and #2. Thanks to Hongbin explaining to me that such efforts are mainly targeting 64bit PEI. But I do not think it takes much more work to support 32bit PEI + Standalone MM once this groundwork is done (again, we can help upstream the code that supports #3 from Project MU if so desired 😊).
Moreover, just because there are no platforms in the public have such configuration, I would perceive it as a great opportunity to introduce the new MM communicate header definition to prevent further contamination from the old definition. In as the new platforms can setup MM foundation in PEI (32 or 64 bit, Intel or AMD) and communicate under the new definition of data structure and PPI.
Please let me know your thoughts. Your feedback is really appreciated!
Regards,
Kun
From: Ni, Ray <ray.ni@intel.com>
Sent: Thursday, August 8, 2024 8:48 PM
To: Kun Qin <Kun.Qin@microsoft.com>; devel@edk2.groups.io
Cc: Wu, Jiaxin <jiaxin.wu@intel.com>; Tan, Dun <dun.tan@intel.com>; Xu, Wei6 <wei6.xu@intel.com>; Zhang, Hongbin1 <hongbin1.zhang@intel.com>; Kinney, Michael D <michael.d.kinney@intel.com>
Subject: [EXTERNAL] Re: Proposing v3 of MM communicate buffer
Kun, thanks for explaining that ARM does not require this.
IMO, the proposal is only useful in X86 when:
* BIOS uses standalone MM
* BIOS launches standalone MM in PEI
* BIOS PEI runs in 32bit mode
I do not see any such platform in open source as the X86 standalone MM is not available yet in edk2 trunk.
Thanks,
Ray
From: Kun Qin <Kun.Qin@microsoft.com>
Sent: Friday, August 9, 2024 1:41
To: Ni, Ray <ray.ni@intel.com>;
devel@edk2.groups.io <devel@edk2.groups.io>
Cc: Wu, Jiaxin <jiaxin.wu@intel.com>; Tan, Dun <dun.tan@intel.com>; Xu, Wei6 <wei6.xu@intel.com>; Zhang, Hongbin1 <hongbin1.zhang@intel.com>;
Ni, Ray <ray.ni@intel.com>; Kinney, Michael D <michael.d.kinney@intel.com>
Subject: RE: Proposing v3 of MM communicate buffer
Hi Ray,
Thanks for your feedback. The ARM platforms I was exposed to have consistent operation mode is only AARCH64, so this proposal is not particularly attached to any ARM problem.
I agree that 32bit PEI/DXE communicate into MM will have issue on x86 platforms as of today. But I have only heard Intel processors moving to support x64 PEI/DXE. I think introducing a new MM communicate header will help to prevent the issue to propagate much further as it might take non-Intel x86 platforms years to fully move away from 32bit PEI/DXE. Please let me know if you have any thoughts.
Regards,
Kun
P.S. Project MU have a thunking module that can launch x64 MM core from 32bit environment: mu_feature_mm_supv/MmSupervisorPkg/Drivers/MmPeiLaunchers/MmIplX64Relay.inf at main · microsoft/mu_feature_mm_supv (github.com). We can upstream it to edk2 if folks think it will help.
From: Ni, Ray <ray.ni@intel.com>
Sent: Thursday, August 8, 2024 1:15 AM
To: devel@edk2.groups.io; Kun Qin <Kun.Qin@microsoft.com>
Cc: Wu, Jiaxin <jiaxin.wu@intel.com>; Tan, Dun <dun.tan@intel.com>; Xu, Wei6 <wei6.xu@intel.com>; Zhang, Hongbin1 <hongbin1.zhang@intel.com>;
Ni, Ray <ray.ni@intel.com>; Kinney, Michael D <michael.d.kinney@intel.com>
Subject: [EXTERNAL] Re: Proposing v3 of MM communicate buffer
Kun,
I like your proposed solution as it is backward compatible.
But, I think the new PPI/Protocol is only useful when the CPU mode where PPI/Protocol is produced does not match the CPU mode in MM.
In X86, it could be: 32bit PEI + 64bit MM, 32bit DXE + 64bit MM, or vice versa. But I doubt the value of support these combinations in X86. Because that means the IPL (either PEI or DXE module) needs to support invoking MM Core in a different CPU mode.
And the latest X86 platforms are switching to 64bit PEI + 64bit DXE + 64bit MM.
Does the proposal try to solve some ARM problem? Can you explain the necessity? I would like to avoid the complicated interfaces which do not solve a practical problem.
Thanks,
Ray
From: devel@edk2.groups.io
<devel@edk2.groups.io> on behalf of Kun Qin via groups.io
<Kun.Qin=microsoft.com@groups.io>
Sent: Thursday, August 8, 2024 2:14
To: devel@edk2.groups.io <devel@edk2.groups.io>
Subject: [edk2-devel] Proposing v3 of MM communicate buffer
Hi all,
I am trying to propose a change into PI spec and would like to gather some feedback in this forum.
Essentially, the current communicate header contains a UINTN field in place, which is causing programing
errors when trying to communicate the message between different operation mode (i.e. PEI in IA32
communicate into MM in x64). There are various implementations at large to compensate for this
size discrepancy through the edk2 codebase, thus fixing the existing communicate buffer definition
will be less feasible. Thus I think proposing a new structure and implement the corresponding header
parser will be a simpler approach, which also allows a bit more flexibility to inject new features/checks
into the communication channel.
The proposed change for the spec is detailed here:
https://github.com/kuqin12/edk2/blob/BZ3398-MmCommunicate-Length-v4/CodeFirst/BZ3430-SpecChange.md
And the code first change is listed here:
https://github.com/kuqin12/edk2/blob/BZ3398-MmCommunicate-Length-v4/
Could you please provide me with any feedback that you think might be helpful for future usage of MM
communicate? Any input is appreciated.
Regards,
Kun