Re: [PATCH v2 0/3] XCODE5 toolchain binary patching fix

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

On 05/06/20 18:32, Tom Lendacky wrote:
>
> BZ: https://bugzilla.tianocore.org/show_bug.cgi?id=2340
>
> Commit 2db0ccc2d7fe ("UefiCpuPkg: Update CpuExceptionHandlerLib pass
> XCODE5 tool chain") introduced binary patching in the
> ExceptionHandlerAsm.nasm in order to support the XCODE5 toolchain.
> However, the CpuExceptionHandlerLib can be used during SEC phase which
> would result in binary patching of flash.
>
> This series creates a new CpuExceptionHandlerLib file to support the
> required binary patching for the XCODE5 toolchain, while reverting the
> changes from commit 2db0ccc2d7fe in the standard file. As the Pei, Dxe
> and SMM versions of the library operate in memory (as opposed to
> flash), only the SEC/PEI version is of the library is updated to use
> the version of the ExceptionHandlerAsm.nasm that does not perform
> binary patching.
>
> This is accomplished in phases:
>   - Create a new XCODE5 specific version of the
>     ExceptionHandlerAsm.nasm file and update all CpuExceptionHandler
>     INF files to use it while also creating a new SEC/PEI
>     CpuExceptionHandler INF file specifically for the XCODE5
>     toolchain.
>   - Update all package DSC files that use the
>     SecPeiCpuExceptionHandlerLib version of the library to use the
>     XCODE5 version of the library, Xcode5SecPeiCpuExceptionHandlerLib,
>     when the XCODE5 toolchain is used.
>   - Revert the changes made by commit 2db0ccc2d7fe in the standard
>     file and update the SecPeiCpuExceptionHandlerLib.inf file to use
>     the standard file.
>
> I don't have access to an XCODE5 toolchain setup, so I have not tested
> this with XCODE5. I would like to request that someone who does please
> test this.
>
> Also, will this change have an impact on any of the platform builds
> outside of this tree?

This series takes care of all the "SecPeiCpuExceptionHandlerLib.inf"
occurrences in edk2.

Regarding edk2-platforms, with master being at 644e223bb371
("Platform/RaspberryPi: create DXE phase SerialPortLib version for
RPi3", 2020-05-06):

$ git grep -l \
      'UefiCpuPkg/Library/CpuExceptionHandlerLib/SecPeiCpuExceptionHandlerLib.inf'

Platform/Intel/MinPlatformPkg/Include/Dsc/CorePeiLib.dsc
Platform/Intel/QuarkPlatformPkg/Quark.dsc
Platform/Intel/QuarkPlatformPkg/QuarkMin.dsc
Platform/Intel/Vlv2TbltDevicePkg/PlatformPkgIA32.dsc
Platform/Intel/Vlv2TbltDevicePkg/PlatformPkgX64.dsc

I don't know what's best for these platforms: an XCODE5-compatible
source code that does the wrong thing when run from flash, or an
XCODE5-incompatible source code that does the right thing.

If all of these platforms lived in edk2 proper, I would suggest
inserting patches for them just before patch#3 in this series --
similarly to your OvmfPkg patch. But, these platforms live outside of
edk2, and I don't know if they are ever built with XCODE5...

... You could propose a separate edk2-platforms series:

Platform/Intel/MinPlatformPkg/Include/Dsc/CorePeiLib.dsc
  Chasel Chiu <chasel.chiu@intel.com>
  Nate DeSimone <nathaniel.l.desimone@intel.com>
  Liming Gao <liming.gao@intel.com>

Platform/Intel/QuarkPlatformPkg/Quark.dsc
  Michael D Kinney <michael.d.kinney@intel.com>
  Kelly Steele <kelly.steele@intel.com>

Platform/Intel/QuarkPlatformPkg/QuarkMin.dsc
  Michael D Kinney <michael.d.kinney@intel.com>
  Kelly Steele <kelly.steele@intel.com>

Platform/Intel/Vlv2TbltDevicePkg/PlatformPkgIA32.dsc
  Zailiang Sun <zailiang.sun@intel.com>
  Yi Qian <yi.qian@intel.com>

Platform/Intel/Vlv2TbltDevicePkg/PlatformPkgX64.dsc
  Zailiang Sun <zailiang.sun@intel.com>
  Yi Qian <yi.qian@intel.com>

and we could then delay just the pushing of patch#3 in this series until
the edk2-platforms patches have been merged too.

> In other words, should the new INF be the one that uses the reverted
> ExceptionHandlerAsm.nasm file and it be called something like
> BaseSecPeiCpuExceptionHandlerLib.inf?

Keeping the current (= self-patching) logic associated with the current
filename ("SecPeiCpuExceptionHandlerLib.inf") would certainly eliminate
the headache about out-of-tree platforms. But it would also mean we'd
have to use a special prefix for the *non-broken* SEC INF file. And that
irks me quite a bit. The quirk is in the patching variant, and IMO that
should be reflected by the file name.

Thanks
Laszlo