Re: [Patch v5 48/48] UefiCpuPkg/PiSmmCpuDxeSmm: Add gEfiVariableArchProtocolGuid dependency

[Laszlo Ersek <lersek@redhat.com>]

On 08/19/16 04:57, Zeng, Star wrote:
> On 2016/8/19 10:45, Zeng, Star wrote:
>> On 2016/8/19 10:26, Laszlo Ersek wrote:
>>> On 08/19/16 04:00, Fan, Jeff wrote:
>>>> Laszlo,
>>>>
>>>> I could revert this patch firstly.
>>>
>>> Thank you, that would be very kind.
>>>
>>>> Could you please dig out the OVMF to address the potential issue,
>>>> then I could re-commit it for those platforms required this patch.
>>>
>>> The problem is that this week (what remains of it) and the next week I
>>> won't really have time for this -- tomorrow I'm hoping to finish up
>>> something else in a mortal hurry. It was actually in preparation for
>>> rebasing / pushing that work that I pulled "master", and found out about
>>> the regression.
>>>
>>> Can we perhaps get help from the variable stack maintainers? What's the
>>> design of the (lack of) depexes on those drivers?
>>
>> Variable driver consumes
>> PcdFlashNvStorageVariableBase(64)/PcdFlashNvStorageVariableSize to
>> produce *gEfiVariableArchProtocolGuid* protocol.
>> Variable driver registers (SMM)FTW protocol notification function
>> SmmFtwNotificationEvent() or FtwNotificationEvent() to produce
>> *gEfiVariableWriteArchProtocolGuid* protocol.
>> (SMM)FTW driver has dependency on gEfiSmmFirmwareVolumeBlockProtocolGuid
>> or gEfiFirmwareVolumeBlockProtocolGuid.
>>
>> I am not so sure what you said about the (lack of) depexes on those
>> drivers.
>>
>> Did you see variable driver loaded and started when ASSERT appeared on
>> OVMF?
> 
> 
> You may could compare the serial logs to get if there is some driver
> execution flow differences for the images built without and with this
> patch.

Thanks, I'll try to do that the week after next or so, hopefully.

Laszlo