Re: [edk2-devel] [PATCH 3/3] UefiCpuPkg/PiSmmCpu: Enable 5 level paging when CPU supports

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

On 06/28/19 08:47, Ni, Ray wrote:
> REF:https://bugzilla.tianocore.org/show_bug.cgi?id=1946
>
> The patch changes SMM environment to use 5 level paging when CPU
> supports it.
>
> Signed-off-by: Ray Ni <ray.ni@intel.com>
> Cc: Eric Dong <eric.dong@intel.com>
> ---
>  .../PiSmmCpuDxeSmm/SmmCpuMemoryManagement.c   |  20 +-
>  UefiCpuPkg/PiSmmCpuDxeSmm/SmmProfile.c        | 272 ++++++----
>  UefiCpuPkg/PiSmmCpuDxeSmm/X64/PageTbl.c       | 483 ++++++++++++------
>  UefiCpuPkg/PiSmmCpuDxeSmm/X64/SmiEntry.nasm   |  12 +
>  .../PiSmmCpuDxeSmm/X64/SmmProfileArch.c       |  72 ++-
>  5 files changed, 559 insertions(+), 300 deletions(-)

This patch does not build with GCC, because:

> diff --git a/UefiCpuPkg/PiSmmCpuDxeSmm/X64/PageTbl.c b/UefiCpuPkg/PiSmmCpuDxeSmm/X64/PageTbl.c
> index 3d5d663d99..c088010327 100644
> --- a/UefiCpuPkg/PiSmmCpuDxeSmm/X64/PageTbl.c
> +++ b/UefiCpuPkg/PiSmmCpuDxeSmm/X64/PageTbl.c
> @@ -16,6 +16,7 @@ SPDX-License-Identifier: BSD-2-Clause-Patent
>  LIST_ENTRY                          mPagePool = INITIALIZE_LIST_HEAD_VARIABLE (mPagePool);
>  BOOLEAN                             m1GPageTableSupport = FALSE;
>  BOOLEAN                             mCpuSmmStaticPageTable;
> +BOOLEAN                             m5LevelPagingSupport;

as-is, two translation units *define* (allocate) "m5LevelPagingSupport":
"PageTbl.c" above, and "SmiEntry.nasm". And that breaks the build with
GCC (it should break the build with VS as well, because it is a bug in
the code).

However, I'm not suggesting that we add "extern" to "PageTbl.c".

Because, I don't think we should reintroduce DBs into the PiSmmCpuDxeSmm
assembly code. That would be a regression for:

  https://bugzilla.tianocore.org/show_bug.cgi?id=866

We should *especially* not reintroduce the dual use of a byte
- both for binary instruction encoding,
- and as a data value (for steering C-language code).

PiSmmCpuDxeSmm had that kind of code before, but I eliminated it. For
example, in commit 3c5ce64f23c4 ("UefiCpuPkg/PiSmmCpuDxeSmm: patch
"XdSupported" with PatchInstructionX86()", 2018-04-04).

Therefore, please incorporate the following update, into patch #3:

> diff --git a/UefiCpuPkg/PiSmmCpuDxeSmm/X64/SmiEntry.nasm b/UefiCpuPkg/PiSmmCpuDxeSmm/X64/SmiEntry.nasm
> index b5e0405b3b00..ae79bf024bf0 100644
> --- a/UefiCpuPkg/PiSmmCpuDxeSmm/X64/SmiEntry.nasm
> +++ b/UefiCpuPkg/PiSmmCpuDxeSmm/X64/SmiEntry.nasm
> @@ -69,7 +69,7 @@ extern ASM_PFX(mXdSupported)
>  global ASM_PFX(gPatchXdSupported)
>  global ASM_PFX(gPatchSmiStack)
>  global ASM_PFX(gPatchSmiCr3)
> -global ASM_PFX(m5LevelPagingSupport)
> +global ASM_PFX(gPatch5LevelPagingSupport)
>  global ASM_PFX(gcSmiHandlerTemplate)
>  global ASM_PFX(gcSmiHandlerSize)
>
> @@ -126,8 +126,8 @@ ASM_PFX(gPatchSmiCr3):
>      mov     cr3, rax
>      mov     eax, 0x668                   ; as cr4.PGE is not set here, refresh cr3
>
> -    DB      0xb1                        ; mov cl, m5LevelPagingSupport
> -ASM_PFX(m5LevelPagingSupport): DB 0
> +    mov     cl, strict byte 0           ; source operand will be patched
> +ASM_PFX(gPatch5LevelPagingSupport):
>      cmp     cl, 0
>      je      SkipEnable5LevelPaging
>      ;
> diff --git a/UefiCpuPkg/PiSmmCpuDxeSmm/X64/PageTbl.c b/UefiCpuPkg/PiSmmCpuDxeSmm/X64/PageTbl.c
> index c08801032766..c31160735a37 100644
> --- a/UefiCpuPkg/PiSmmCpuDxeSmm/X64/PageTbl.c
> +++ b/UefiCpuPkg/PiSmmCpuDxeSmm/X64/PageTbl.c
> @@ -17,6 +17,7 @@ LIST_ENTRY                          mPagePool = INITIALIZE_LIST_HEAD_VARIABLE (m
>  BOOLEAN                             m1GPageTableSupport = FALSE;
>  BOOLEAN                             mCpuSmmStaticPageTable;
>  BOOLEAN                             m5LevelPagingSupport;
> +X86_ASSEMBLY_PATCH_LABEL            gPatch5LevelPagingSupport;
>
>  /**
>    Disable CET.
> @@ -337,6 +338,7 @@ SmmInitPageTable (
>    m1GPageTableSupport    = Is1GPageSupport ();
>    m5LevelPagingSupport   = Is5LevelPagingSupport ();
>    mPhysicalAddressBits   = CalculateMaximumSupportAddress ();
> +  PatchInstructionX86 (gPatch5LevelPagingSupport, m5LevelPagingSupport, 1);
>    DEBUG ((DEBUG_INFO, "5LevelPaging Support     - %d\n", m5LevelPagingSupport));
>    DEBUG ((DEBUG_INFO, "1GPageTable Support      - %d\n", m1GPageTableSupport));
>    DEBUG ((DEBUG_INFO, "PcdCpuSmmStaticPageTable - %d\n", mCpuSmmStaticPageTable));

With this update, the build succeeds, and a quick regression-test has
passed for me (using OVMF/IA32X64).

I'll try to do deeper testing if you agree with this update.

Thanks,
Laszlo