Re: [edk2-devel] [PATCH v2 2/2] UefiCpuPkg/PiSmmCpu: PcdCpuSmmAccessOut controls SMM access-out policy

["Laszlo Ersek" <lersek@redhat.com>]

On 08/01/19 01:46, Laszlo Ersek wrote:

> So ultimately, I would argue for the following patch series:
> 
> - patch#1: Revert commit c60d36b4, and explain why, in the commit
>   message.
> 
> - patch#2: Introduce the new PCD, also mentioning MMIO.
> 
> - Patch#3: modify *all* of the following functions, internally, to
>   return immediately, if "mSmmAccessOut" is TRUE:
> 
>   - GetUefiMemoryMap()
>   - SetUefiMemMapAttributes()
>   - IsSmmCommBufferForbiddenAddress()
> 
>   Basically, the new PCD should short-circuit all three functions
>   declared in "PiSmmCpuDxeSmm.h" by commit d2fc7711136a.

Sigh, why is this so friggin complicated.

The functions added by commit d2fc7711136a were extended with the
following commits, later:

* ac6613db4697 ("UefiCpuPkg/PiSmmCpu: Check for untested memory in GCD",
2018-07-26)

The commit message says, "It treats GCD untested memory as invalid SMM
communication buffer".

* 8a2e1a9d54ee ("UefiCpuPkg/PiSmmCpu: Check EFI_RUNTIME_RO in UEFI mem
attrib table.", 2018-07-26)

The commit message says, "It treats the UEFI runtime page with
EFI_MEMORY_RO attribute as invalid SMM communication buffer.


Both of these commits come from the patch series

  [edk2] [PATCH 0/6] Check untested memory and EFI_MEMORY_RO

  http://mid.mail-archive.com/20180720052626.24932-1-hao.a.wu@intel.com
  https://lists.01.org/pipermail/edk2-devel/2018-July/027326.html

And now the question becomes: when a platform permits "access-out", does
it also enable access to untested memory, and EFI_MEMORY_RO?

(a) If the answer is yes, then short-circuiting the three functions that
I listed, on (mSmmAccessOut == TRUE), *immediately* after entering them,
is correct.

(b) If the answer is no -- that is, access-out should *continue*
preventing access to untested memory, and EFI_MEMORY_RO -- then we
cannot completely short-circuit the functions that I listed. Instead, we
can only short-circuit those parts that work with the "mUefiMemoryMap"
variable.

The current description of the PCD suggests that (b) is the right approach.

Thanks,
Laszlo