Re: [edk2-devel] [PATCH] OvmfPkg: Set PciLib for TdxDxe driver

["Lendacky, Thomas" <thomas.lendacky@amd.com>]

On 4/19/22 00:06, Yao, Jiewen wrote:
> OK. Let me describe what I think.
> 
> PCI Express BAR need to be initialized by someone in the platform.
> This initialization may require CFG8. That is understandable.
> 
> A good design is that: After the PCIE BAR is initialized, it can be accessed.
> Requires additional step (such as clear C-bit) means the PCIE BAR is not fully initialized originally. I don't think it is a good idea.
> 
> So far, the problem is TdxDxe, but what if a PEI driver also wants to use access PCIE space? It may run into same problem.
> 
> I think the best way is to clear C-bit in PciExBarInitialization(), as SEV specific step to finish initialization. https://github.com/tianocore/edk2/blob/master/OvmfPkg/Library/PlatformInitLib/Platform.c#L261
> 
> As such, no matter how many drivers want to use PCIE, they can.
> 
> 
> Splitting PCIE bar programming and C bit clearing is a big problem. In this window, no one can actually touch the PCIE bar, although it seems being initialized...

I tried this approach and it does not work. It is because new page tables 
are used in the DXE phase and so the c-bit has to be cleared in the new 
page tables vs the page tables used in PEI.

Thanks,
Tom

> 
> 
> Thank you
> Yao Jiewen
> 
>> -----Original Message-----
>> From: devel@edk2.groups.io <devel@edk2.groups.io> On Behalf Of Yao, Jiewen
>> Sent: Tuesday, April 19, 2022 12:47 PM
>> To: Xu, Min M <min.m.xu@intel.com>; devel@edk2.groups.io
>> Cc: Brijesh Singh <brijesh.singh@amd.com>; Aktas, Erdem
>> <erdemaktas@google.com>; James Bottomley <jejb@linux.ibm.com>; Tom
>> Lendacky <thomas.lendacky@amd.com>
>> Subject: Re: [edk2-devel] [PATCH] OvmfPkg: Set PciLib for TdxDxe driver
>>
>> Can SEV clear the C-bit in SEC phase?
>>
>> I think that is right way to ensure PCI Express can always be accessed by anyone.
>>
>>
>>> -----Original Message-----
>>> From: Xu, Min M <min.m.xu@intel.com>
>>> Sent: Tuesday, April 19, 2022 12:39 PM
>>> To: Yao, Jiewen <jiewen.yao@intel.com>; devel@edk2.groups.io
>>> Cc: Brijesh Singh <brijesh.singh@amd.com>; Aktas, Erdem
>>> <erdemaktas@google.com>; James Bottomley <jejb@linux.ibm.com>; Tom
>>> Lendacky <thomas.lendacky@amd.com>
>>> Subject: RE: [edk2-devel] [PATCH] OvmfPkg: Set PciLib for TdxDxe driver
>>>
>>> In AmdSevDxe's entry point it clears the C-bit from PcdPciExpressBaseAddress
>>> and other memory spaces if needed. Please see
>>>
>> https://github.com/tianocore/edk2/blob/master/OvmfPkg/AmdSevDxe/AmdSev
>>> Dxe.c#L81-L95. After that OVMF can use PCI express.
>>>
>>> This broken is caused by the call sequence of TdxDxe driver and AmdSevDxe
>>> driver. Currently TdxDxe driver is loaded before AmdSevDxe, so in SEV-ES guest
>>> the C-bit of PcdPciExpressBaseAddress hasn't been cleared. In this situation
>> the
>>> access to PciExpressBaseAddress trigger exceptions (lib constructor in TdxDxe).
>>>
>>> There are 2 options to fix this issue.
>>> 1. Adjust the load sequence of AmdSevDxe and TdxDxe (Load AmdSevDxe
>> before
>>> TdxDxe)
>>> 2. Make TdxDxe to import BasePciLibCf8.inf instead of DxePciLibI440FxQ35.inf
>>> (just like AmdSevDxe)
>>>
>>> Tom and I tested above 2 options in SEV and TDX and all work.
>>>
>>>> -----Original Message-----
>>>> From: Yao, Jiewen <jiewen.yao@intel.com>
>>>> Sent: Tuesday, April 19, 2022 12:16 PM
>>>> To: Xu, Min M <min.m.xu@intel.com>; devel@edk2.groups.io
>>>> Cc: Brijesh Singh <brijesh.singh@amd.com>; Aktas, Erdem
>>>> <erdemaktas@google.com>; James Bottomley <jejb@linux.ibm.com>; Tom
>>>> Lendacky <thomas.lendacky@amd.com>
>>>> Subject: RE: [edk2-devel] [PATCH] OvmfPkg: Set PciLib for TdxDxe driver
>>>>
>>>> Do you mean, with SEV introduced, OVMF cannot use PCI express any more?
>>>>
>>>> Thank you
>>>> Yao Jiewen
>>>>
>>>>
>>>>> -----Original Message-----
>>>>> From: Xu, Min M <min.m.xu@intel.com>
>>>>> Sent: Tuesday, April 19, 2022 11:05 AM
>>>>> To: Yao, Jiewen <jiewen.yao@intel.com>; devel@edk2.groups.io
>>>>> Cc: Brijesh Singh <brijesh.singh@amd.com>; Aktas, Erdem
>>>>> <erdemaktas@google.com>; James Bottomley <jejb@linux.ibm.com>;
>> Tom
>>>>> Lendacky <thomas.lendacky@amd.com>
>>>>> Subject: RE: [edk2-devel] [PATCH] OvmfPkg: Set PciLib for TdxDxe
>>>>> driver
>>>>>
>>>>> On April 19, 2022 10:54 AM, Yao Jiewen wrote:
>>>>>>
>>>>>> Why does TdxDxe call TdxMailbox in an SEV platform?
>>>>>> Or why does TdxMailbox call SynchronizationLib in an SEV platform?
>>>>>>
>>>>> TdxDxe will not call TdxMailbox/SynchronizationLib in SEV platform.
>>>>> The problem is in the lib constructor. When TdxDxe driver is loaded,
>>>>> before its entry point is called, the lib constructors will be called even in a
>>>> SEV platform.
>>>>>>
>>>>>> There are many places we can do CcProbe to stop action. Why we need
>>>>>> do it in DSC?
>>>>> So we cannot stop the lib constructor with CcProbe in this case.
>>>>>
>>>>> Thanks
>>>>> Min
>>
>>
>> 
>>
>