Re: [MdeModulePkg] SetVirtualAddressMap() crashed due to DxeReportStatusCodeLib assuming the state of the BootService Memory at runtime.

[Andrew Fish <afish@apple.com>]

> On Aug 8, 2016, at 7:26 PM, Zeng, Star <star.zeng@intel.com> wrote:
> 
> On 2016/8/9 10:07, Andrew Fish wrote:
>> 
>>> On Aug 8, 2016, at 6:21 PM, Zeng, Star <star.zeng@intel.com> wrote:
>>> 
>>> Andrew,
>>> 
>>> Should MdeModulePkg/Library/RuntimeDxeReportStatusCodeLib be used for your case if there are really runtime status code reporting needed?
>>> 
>> 
>> Star,
>> 
>> If the Library instance does not fully support DXE_RUNTIME_DRIVER, why is it listed in the LIBRARY_CLASS as supported?
> 
> This kind of library can support DXE_RUNTIME_DRIVER at boot time, for example UefiHiiLib, UefiHiiServicesLib, UefiBootServicesTableLib, DxePcdLib and etc.
> 

Star,

I understand giving access to Boot Services resources to a Runtime Driver for its constructor. I think the difference here is the DxeReportStatusCodeLib is abstracting a runtime service, but not doing it in a way to really works properly at runtime in all cases. I would expect that a library abstraction a runtime feature would work at runtime. Which is why I ended up with a "bad" mapping in the first place.

Thanks,

Andrew Fish 

>> 
>> https://github.com/tianocore/edk2/blob/master/MdeModulePkg/Library/DxeReportStatusCodeLib/DxeReportStatusCodeLib.inf#L25
>> 
>> LIBRARY_CLASS                  = ReportStatusCodeLib|DXE_CORE DXE_DRIVER DXE_RUNTIME_DRIVER DXE_SAL_DRIVER DXE_SMM_DRIVER UEFI_APPLICATION UEFI_DRIVER SMM_CORE
>> 
>> Actually I tried to add the UefiRuntimeDriverLib and the build failed as UefiRuntimeDriverLib was not supported for the DXE_CORE type. Maybe the bug is this library instance lists DXE_RUNTIME_DRIVER,  DXE_SAL_DRIVER and DXE_SMM_DRIVER when it has special case code to support DXE_CORE? Maybe this library is trying to do too many things?
>> 
>> What ReportStatusCodeLib would you recommend to link with RuntimeDxe driver: https://github.com/tianocore/edk2/blob/master/MdeModulePkg/Core/RuntimeDxe/RuntimeDxe.inf
> 
> [LibraryClasses.common.DXE_CORE]
> ReportStatusCodeLib|MdeModulePkg/Library/DxeReportStatusCodeLib/DxeReportStatusCodeLib.inf
> 
> [LibraryClasses.common.DXE_RUNTIME_DRIVER]
> ReportStatusCodeLib|MdeModulePkg/Library/RuntimeDxeReportStatusCodeLib/RuntimeDxeReportStatusCodeLib.inf
> 
> [LibraryClasses.common.UEFI_DRIVER]
> ReportStatusCodeLib|MdeModulePkg/Library/DxeReportStatusCodeLib/DxeReportStatusCodeLib.inf
> 
> [LibraryClasses.common.DXE_DRIVER]
> ReportStatusCodeLib|MdeModulePkg/Library/DxeReportStatusCodeLib/DxeReportStatusCodeLib.inf
> 
> [LibraryClasses.common.UEFI_APPLICATION]
> ReportStatusCodeLib|MdeModulePkg/Library/DxeReportStatusCodeLib/DxeReportStatusCodeLib.inf
> 
> Thanks,
> Star
> 
>> 
>> Thanks,
>> 
>> Andrew Fish
>> 
>>> Thanks,
>>> Star
>>> -----Original Message-----
>>> From: edk2-devel [mailto:edk2-devel-bounces@lists.01.org] On Behalf Of Andrew Fish
>>> Sent: Tuesday, August 9, 2016 7:08 AM
>>> To: edk2-devel <edk2-devel@lists.01.org>
>>> Subject: [edk2] [MdeModulePkg] SetVirtualAddressMap() crashed due to DxeReportStatusCodeLib assuming the state of the BootService Memory at runtime.
>>> 
>>> I was messing about with an ExitBootServices test that fills boot services memory with 0xAFAFAFAFAFAFAFAF (It was Vincent's idea to use my Initials but it has the handy property of being a non-cononical address and causes on GP fault on X64) and SetVirtualAddressMap() started crashing.
>>> 
>>> It looks like this code is assuming the 1st call to ReportStatus code will not happen at runtime. This is not the case for the RuntimeDxe driver.
>>> https://github.com/tianocore/edk2/blob/master/MdeModulePkg/Library/DxeReportStatusCodeLib/ReportStatusCodeLib.c#L43
>>> VOID
>>> InternalGetReportStatusCode (
>>> VOID
>>> )
>>> {
>>> EFI_STATUS  Status;
>>> 
>>> if (mReportStatusCodeLibStatusCodeProtocol != NULL) {
>>>  return;
>>> }
>>> 
>>> //
>>> // Check gBS just in case ReportStatusCode is called before gBS is initialized.
>>> //
>>> if (gBS != NULL && gBS->LocateProtocol != NULL) {
>>>  Status = gBS->LocateProtocol (&gEfiStatusCodeRuntimeProtocolGuid, NULL, (VOID**) &mReportStatusCodeLibStatusCodeProtocol);
>>>  if (EFI_ERROR (Status)) {
>>>    mReportStatusCodeLibStatusCodeProtocol = NULL;
>>>  }
>>> }
>>> }
>>> 
>>> I'm guessing this seems to work due to:https://github.com/tianocore/edk2/blob/master/MdeModulePkg/Core/Dxe/DxeMain/DxeMain.c#L803
>>> 
>>> //
>>> // Zero out the Boot Service Table
>>> //
>>> ZeroMem (gBS, sizeof (EFI_BOOT_SERVICES));
>>> 
>>> 
>>> Thus if I'm looking at this code correctly it only looks like it works at Runtime since it is depending on the value of a boot services memory buffer not changing. This is not a valid assumption as that code is owned by the caller of ExitBootServices, so it should be legal for my test to change the value.
>>> 
>>> I wanted to get a few more eyes on this prior to filling a bug?
>>> 
>>> Thanks,
>>> 
>>> Andrew Fish
>>> 
>>> 
>>> _______________________________________________
>>> edk2-devel mailing list
>>> edk2-devel@lists.01.org
>>> https://lists.01.org/mailman/listinfo/edk2-devel
>>> _______________________________________________
>>> edk2-devel mailing list
>>> edk2-devel@lists.01.org
>>> https://lists.01.org/mailman/listinfo/edk2-devel
> 
> _______________________________________________
> edk2-devel mailing list
> edk2-devel@lists.01.org
> https://lists.01.org/mailman/listinfo/edk2-devel