Re: [PATCH edk2-platforms 6/7] Platform/DeveloperBox: add .DSC/.FDF description of MM components

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

On Thu, 17 Jan 2019 at 13:08, Leif Lindholm <leif.lindholm@linaro.org> wrote:
>
> On Thu, Jan 17, 2019 at 12:10:01PM +0100, Ard Biesheuvel wrote:
> > > >  ################################################################################
> > > >  #
> > > > @@ -294,8 +295,10 @@ [PcdsFixedAtBuild.common]
> > > >  !endif
> > > >    gEfiMdeModulePkgTokenSpaceGuid.PcdFirmwareRevision|$(BUILD_NUMBER)
> > > >
> > > > -  gArmTokenSpaceGuid.PcdMmBufferBase|0xFFC00000
> > > > -  gArmTokenSpaceGuid.PcdMmBufferSize|0x00200000
> > > > +  gEfiSecurityPkgTokenSpaceGuid.PcdUserPhysicalPresence|TRUE
> > >
> > > So, I can see why you add this hard-wired for the purpose of testing.
> > > But please, add a *very* conspicuous, and strongly worded, comment
> > > statement preceding it.
> >
> > Well, I was talking to Peter about this the other day: according to
> > the spec, this setting should only matter before exit boot services,
> > and since this platform only supports serial and GOP consoles, one
> > could argue that only a physically present user could interact with it
> > before that time.
>
> But that also makes the Pcd pointless.
>
> > The obvious way of implementing this non-trivially on this platform is
> > to use a DIP switch, but that requires you to open the case to
> > enroll/delete the platform key. Perhaps that does not matter, and it
> > would in fact produce a less dangerous reference implementation.
>
> I would be totally OK with that.
> I would also be totally OK with a DynamicPcd settable through the UI
> (which is what most machines I come across have).
>

That won't work for this implementation: the state of dynamic PCDs
does not propagate into the MM world (nor should it), and so no MM
driver implementing the dynamic PCD protocol exists.

> But I would also be cool with a sufficiently evil "here be dragons"
> statement, pointing out that we don't care that much
> *on*this*specific*system* because the NOR isn't actually hw protected
> anyway, and this implementation is all about exercising the software
> stack..
>

I'll go with that for the time being.