Re: [edk2-devel] [PATCH v2] OvmfPkg: strip build paths in release builds

["Andrew Fish" <afish@apple.com>]

[-- Attachment #1: Type: text/plain, Size: 9411 bytes --]

Liming,

I understand how the tool works, but that is not how it is used. To make it work like other toolchains *_*_*_GENFW_FLAGS for ELF targets should pass —zero for builds that do not contain symbols. This is how it works today [1].

The proposed fix is a global hammer ` RELEASE_*_*_GENFW_FLAGS = --zero` and it prevents some one from overriding the toolchain definition to turn on source level debugging for Release builds. In all the other places this is a tool chain choice. So my complaint is solving this globally vs. on a per toolchain basis. All the other toolchains don’t produce.

If you look at the XCODE5 toolchian that we override the edk2 default   DEBUG_XCODE5_X64_CC_FLAGS has -g and RELEASE_XCODE5_X64_CC_FLAGS does not. So the source level debugging or not is part of the compiler target choice. 
  DEBUG_XCODE5_*_MTOC_FLAGS = -align 0x20 -d $(DEBUG_DIR)/$(MODULE_NAME).dll
RELEASE_XCODE5_*_MTOC_FLAGS = -align 0x20

For our override we pass -g to RELEASE builds CC_FLAGS and do `RELEASE_XCODE5_*_MTOC_FLAGS = -align 0x20 -d $(MODULE_GUID)` so the override to force --zero on every call to GenFw will break this. 

So my point is if the toolchain is generating a Debug Directory entry in a RELEASE target that does not support debugging then that target should setting `RELEASE_<target>_*_GENFW_FLAGS = —zero` to undo the unwanted work. I understand why the debug info gets added as it reduces the complexity of the ELF to PE/COFF conversion, so I’m not complaining about that just that the ELF targets don’t handle this themselves. So to me this fix is working the bug of the ELF targets not passing —zero on RELEASE builds to GenFw.

[1] $ git grep _GENFW_FLAGS -- *.template
BaseTools/Conf/tools_def.template:2872:DEBUG_CLANGPDB_X64_GENFW_FLAGS      = --keepexceptiontable
BaseTools/Conf/tools_def.template:2877:RELEASE_CLANGPDB_X64_GENFW_FLAGS    =
BaseTools/Conf/tools_def.template:2882:NOOPT_CLANGPDB_X64_GENFW_FLAGS      = --keepexceptiontable
BaseTools/Conf/tools_def.template:3124:*_*_*_GENFW_FLAGS                  =

Thanks,

Andrew Fish

> On Mar 24, 2021, at 7:24 PM, gaoliming <gaoliming@byosoft.com.cn> wrote:
> 
> Andrew:
>  GenFw generates debug entry, and zero debug entry. Its functionality is same to the image without debug entry. So, new option is not introduced to control whether GenFw generates debug entry when it converts ELF image to EFI image. 
>
> Thanks
> Liming
> 发件人: devel@edk2.groups.io <mailto:devel@edk2.groups.io> <devel@edk2.groups.io <mailto:devel@edk2.groups.io>> 代表 Andrew Fish via groups.io <http://groups.io/>
> 发送时间: 2021年3月25日 7:26
> 收件人: edk2-devel-groups-io <devel@edk2.groups.io <mailto:devel@edk2.groups.io>>; ross@burtonini.com <mailto:ross@burtonini.com>
> 抄送: lersek@redhat.com <mailto:lersek@redhat.com>
> 主题: Re: [edk2-devel] [PATCH v2] OvmfPkg: strip build paths in release builds
>
> This breaks some usage we have have in our fork. We have symbols turned on for Release builds, so this change would break that. 
>
> It looks to me that the root cause of this issue might be that the GenFw is blindly writing the debug directory entry into the PE/COFF? For native PE/COFF I think this is controlled by linker flags? For Xcode/clang it is controlled by the *_XCODE5_*_MTOC_FLAGS. So at this point it is kind of up to each toolchain how they want to deal with symbols on release builds. 
>
>
> It seems kind of strange to insert a section and then zero it. Almost seems like the intent of —zero was to post process compare images? 
> 
> 
>   -z, --zero            Zero the Debug Data Fields in the PE input image file.
>                         It also zeros the time stamp fields.
>                         This option can be used to compare the binary efi image.
>                         It can't be combined with other action options
>                         except for -o, -r option. It is a action option.
>                         If it is combined with other action options, the later
>                         input action option will override the previous one.
> 
> 
> And in case you are going to ask our fork uses relative paths from the Build directory and/or a UUID string for the Debug Directory entry file name so it is a constant value and does not impact build reproducibility. 
> 
> 
> From a feature stand point this change will break any hope of source level debugging with RELEASE builds. I also think it changes the exception handler code output in OVMF [1] for ELF toolchains. You are going to get the (No PDB) vs. the file and path you were getting today. I assume if you had tools that natively produce PE/COFF and did not have a Debug Directory entry the same thing would happen prior to this change. 
> 
> 
>     Status = PeCoffLoaderGetEntryPoint ((VOID *) Pe32Data, &EntryPoint);
>     if (EFI_ERROR (Status)) {
>       EntryPoint = NULL;
>     }
>     InternalPrintMessage ("!!!! Find image based on IP(0x%x) ", CurrentEip);
>     PdbPointer = PeCoffLoaderGetPdbPointer ((VOID *) Pe32Data);
>     if (PdbPointer != NULL) {
>       InternalPrintMessage ("%a", PdbPointer);
>     } else {
>       InternalPrintMessage ("(No PDB) " );
>     }
>     InternalPrintMessage (
>       " (ImageBase=%016lp, EntryPoint=%016p) !!!!\n",
>       (VOID *) Pe32Data,
>       EntryPoint
>       );
> 
> 
> Not saying we have to "stop the presses", but just trying to point out the side effects of this change. It is not so much that this change is bad, but that we have no way to turn off the Debug Directory Entry for ELF conversion, so we seem to be working around that issue with a bigger hammer?
>
> [1] https://github.com/tianocore/edk2/blob/master/UefiCpuPkg/Library/CpuExceptionHandlerLib/CpuExceptionCommon.c#L117 <https://github.com/tianocore/edk2/blob/master/UefiCpuPkg/Library/CpuExceptionHandlerLib/CpuExceptionCommon.c#L117>
>
> Thanks,
> 
> 
> Andrew Fish
> 
> 
>> On Mar 24, 2021, at 4:58 AM, Ross Burton <ross@burtonini.com <mailto:ross@burtonini.com>> wrote:
>>
>> GenFw will embed a NB10 section which contains the path to the input file,
>> which means the output files have build paths embedded in them.  To reduce
>> information leakage and ensure reproducible builds, pass --zero in release
>> builds to remove this information.
>> 
>> Ref: https://bugzilla.tianocore.org/show_bug.cgi?id=3256 <https://bugzilla.tianocore.org/show_bug.cgi?id=3256>
>> Signed-off-by: Ross Burton <ross.burton@arm.com <mailto:ross.burton@arm.com>>
>> ---
>> OvmfPkg/AmdSev/AmdSevX64.dsc | 1 +
>> OvmfPkg/Bhyve/BhyveX64.dsc   | 1 +
>> OvmfPkg/OvmfPkgIa32.dsc      | 1 +
>> OvmfPkg/OvmfPkgIa32X64.dsc   | 1 +
>> OvmfPkg/OvmfPkgX64.dsc       | 1 +
>> OvmfPkg/OvmfXen.dsc          | 1 +
>> 6 files changed, 6 insertions(+)
>> 
>> diff --git a/OvmfPkg/AmdSev/AmdSevX64.dsc b/OvmfPkg/AmdSev/AmdSevX64.dsc
>> index 65c42284d9..69a05feea9 100644
>> --- a/OvmfPkg/AmdSev/AmdSevX64.dsc
>> +++ b/OvmfPkg/AmdSev/AmdSevX64.dsc
>> @@ -78,6 +78,7 @@
>>   GCC:*_*_X64_GENFW_FLAGS   = --keepexceptiontable
>>   INTEL:*_*_X64_GENFW_FLAGS = --keepexceptiontable
>> !endif
>> +  RELEASE_*_*_GENFW_FLAGS = --zero
>> 
>>   #
>>   # Disable deprecated APIs.
>> diff --git a/OvmfPkg/Bhyve/BhyveX64.dsc b/OvmfPkg/Bhyve/BhyveX64.dsc
>> index 4a1cdf5aca..132f55cf69 100644
>> --- a/OvmfPkg/Bhyve/BhyveX64.dsc
>> +++ b/OvmfPkg/Bhyve/BhyveX64.dsc
>> @@ -76,6 +76,7 @@
>>   GCC:*_*_X64_GENFW_FLAGS   = --keepexceptiontable
>>   INTEL:*_*_X64_GENFW_FLAGS = --keepexceptiontable
>> !endif
>> +  RELEASE_*_*_GENFW_FLAGS = --zero
>> 
>>   #
>>   # Disable deprecated APIs.
>> diff --git a/OvmfPkg/OvmfPkgIa32.dsc b/OvmfPkg/OvmfPkgIa32.dsc
>> index 1eaf3e99c6..93c209950c 100644
>> --- a/OvmfPkg/OvmfPkgIa32.dsc
>> +++ b/OvmfPkg/OvmfPkgIa32.dsc
>> @@ -80,6 +80,7 @@
>> !if $(TOOL_CHAIN_TAG) != "XCODE5" && $(TOOL_CHAIN_TAG) != "CLANGPDB"
>>   GCC:*_*_*_CC_FLAGS                   = -mno-mmx -mno-sse
>> !endif
>> +  RELEASE_*_*_GENFW_FLAGS = --zero
>> 
>>   #
>>   # Disable deprecated APIs.
>> diff --git a/OvmfPkg/OvmfPkgIa32X64.dsc b/OvmfPkg/OvmfPkgIa32X64.dsc
>> index 4a5a430147..97cc438250 100644
>> --- a/OvmfPkg/OvmfPkgIa32X64.dsc
>> +++ b/OvmfPkg/OvmfPkgIa32X64.dsc
>> @@ -84,6 +84,7 @@
>>   GCC:*_*_X64_GENFW_FLAGS   = --keepexceptiontable
>>   INTEL:*_*_X64_GENFW_FLAGS = --keepexceptiontable
>> !endif
>> +  RELEASE_*_*_GENFW_FLAGS = --zero
>> 
>>   #
>>   # Disable deprecated APIs.
>> diff --git a/OvmfPkg/OvmfPkgX64.dsc b/OvmfPkg/OvmfPkgX64.dsc
>> index d4d601b444..f544fb04bf 100644
>> --- a/OvmfPkg/OvmfPkgX64.dsc
>> +++ b/OvmfPkg/OvmfPkgX64.dsc
>> @@ -84,6 +84,7 @@
>>   GCC:*_*_X64_GENFW_FLAGS   = --keepexceptiontable
>>   INTEL:*_*_X64_GENFW_FLAGS = --keepexceptiontable
>> !endif
>> +  RELEASE_*_*_GENFW_FLAGS = --zero
>> 
>>   #
>>   # Disable deprecated APIs.
>> diff --git a/OvmfPkg/OvmfXen.dsc b/OvmfPkg/OvmfXen.dsc
>> index 507029404f..fcaa35acf1 100644
>> --- a/OvmfPkg/OvmfXen.dsc
>> +++ b/OvmfPkg/OvmfXen.dsc
>> @@ -74,6 +74,7 @@
>>   GCC:*_*_X64_GENFW_FLAGS   = --keepexceptiontable
>>   INTEL:*_*_X64_GENFW_FLAGS = --keepexceptiontable
>> !endif
>> +  RELEASE_*_*_GENFW_FLAGS = --zero
>> 
>>   #
>>   # Disable deprecated APIs.
>> -- 
>> 2.25.1
>> 
>> 
>> 
>> 
>> 
>> 
> 
>
> 


[-- Attachment #2: Type: text/html, Size: 41552 bytes --]