Re: [PATCH 00/13] Extend secure variable service to be usable from Standalone MM

[Ard Biesheuvel <ard.biesheuvel@linaro.org>]

On Thu, 3 Jan 2019 at 08:43, Jagadeesh Ujja <jagadeesh.ujja@arm.com> wrote:
>
> hi Ard
>
> On Wed, Jan 2, 2019 at 10:45 PM Ard Biesheuvel
> <ard.biesheuvel@linaro.org> wrote:
> >
> > On Thu, 20 Dec 2018 at 15:23, Gao, Liming <liming.gao@intel.com> wrote:
> > >
> > > Jagadeesh:
> > >   MdeModulePkg Variable service/Fault tolerant/Nor Flash driver depends on StandaloneMmServicesTableLib library class header file. This header file is added into MdePkg. It has two interfaces. One is global gMmst, another is function InMm(). So, there is no dependency issue here.
> > > And, MdePkg adds one StandaloneMmServicesTableLib library INF with empty implementation, this library is just for build. It sets gMmst=NULL, and always return FASLE in InMm(). This library can be used in MdeModulePkg.dsc to make Variable driver pass build. There is also no dependency issue here. Last, Platform DSC file will refer to the real StandaloneMmServicesTableLib library INF from StandaloneMmPkg.
> > >
> >
> > I think we should avoid the need for InMm() altogether for standalone
> > MM. It will always return TRUE for standalone MM modules, and it will
> > always return FALSE for other modules, so the distinction should be
> > made at build time.
> >
> > This means that we need to refactor the SMM 'server' modules and/or
> > libraries so that any code they cannot share (like boot services
> > invocations) are only included in the classic SMM versions.
> >
> > I have pushed my own prototype code here:
> > https://github.com/ardbiesheuvel/edk2/commits/standalone-mm
> >
> > There is some overlap with Jagadeesh's work. I will work with him
> > directly to resolve this before posting any new revisions.
> >
> InMm()”  and “PcdStandaloneMmVariableEnabled” are defined to reuse the
> existing code as much as possible.
> Initially I have done separate copy of the file to avoid “if..else”
> but had a comment about “duplicating code primarily due to the
> maintenance overhead”
>

We shouldn't rely on runtime functions and PCDs to make build time decision.

Lots of the SMM code can be refactored. As Jian suggested, we could
introduce a helper library with implementations for the MM protocol
handling and memory allocation routines exposed via the system table,
so that the users can invoke the abstract library.

As for the various boot services calls: these just have to be factored
out or replaced.
- gBS->CalculcateCrc32 () in the FTW driver can just be replaced with
the version from BaseLib (but only for the standalone MM version)
- MorLockInitAtEndOfDxe() in the variable driver attempts to locate
some TCG protocols to infer whether the variable may have been set by
platform firmware, this code just needs to be dropped in the
standalone MM version
- the ArmPkg NOR flash driver needs some refactoring in any case,
since all the block I/O and disk I/O routines can be dropped for
standalone MM.

So of course, we have to take the maintenance burden into account, and
so we have to refactor carefully so that we share as much code as
possible. But relying on InMm() and PCDs is not acceptable.

> So we are using “InMm()” and “PcdStandaloneMmVariableEnabled” PCD flag
> and trying to use the same code as much as possible.
>
> The patchset “Extend secure variable service to be usable from
> Standalone MM” as POC was submitted as RFC patches on “October 31,
> 2018”.
> Subsequent comments are fixed and we had 7 version of the patch set
> under review.
>

I'm not sure why you are bringing this up, but it is irrelevant. This
is important stuff that touches a lot of different packages, so if it
takes 20 revisions, so be it.