[edk2-devel] [edk2-platforms][PATCH v1 1/1] IntelSiliconPkg/PeiSmmAccessLib: Remove S3 requirement
Michael Kubacki
mikuback at linux.microsoft.com
Thu Aug 19 19:45:06 UTC 2021
Ray,
In the end, PI modules are allowed to depend on non-deprecated PI PPIs.
The scope of this change is about the API that installs the PPI. So I do
not want to diverge the conversation about changing other code to avoid
the PPI.
I've already been waiting since April to get other IntelSiliconPkg
changes merged (https://edk2.groups.io/g/devel/message/78600) and I
would like to keep this conversation focused on the patch.
This conversation has taken several tangents as I tried to clear up in
this message: https://edk2.groups.io/g/devel/message/79544. I did not
see a response, please read that if you have not.
The maintainability issues of the code have become evident. The code is
coupling unnecessary assumptions and dependencies and that is increasing
the difficulty of understanding and modifying the code.
The goal of this change was for the PeiInstallSmmAccessPpi() function to
simply install the PPI without the boot mode being BOOT_ON_S3_RESUME
which it already documents should be the case.
Because the existing code attempted to couple platform design
assumptions into a generic silicon package, we had to talk about:
- Historical platform S3 requirements
- Interface documentation that is inaccurate
- Clarification around security ambiguity and implications to modifying
the implementation
- General dependencies for Standalone MM
- Potential synchronization issues between the PI Spec defined
interfaces EFI_SMARAM_HOB_DESCRIPTOR_BLOCK and EFI_PEI_MM_ACCESS_PPI
- The role of MM Access PPI in other architecture boot flows
In the process, some of these topics did yield issues for follow up:
1. The documentation interface should accurately reflect actual behavior.
2. The potential synchronization issue should be better understood and
resolved. Note that this issue is not specific to non-S3 flows. PEIMs on
S3 flows could easily read the HOB.
BZ for 1: https://bugzilla.tianocore.org/show_bug.cgi?id=3575
BZ for 2: https://bugzilla.tianocore.org/show_bug.cgi?id=3576
It also seems the library would be better as a PEIM but that's another
topic.
Thanks,
Michael
On 8/19/2021 10:27 AM, Yao, Jiewen wrote:
> Do you want to open/close/lock SMRAM? If yes, then you need PPI.
> If no, then you don’t need PPI. Hob should be enough to provide such info.
>
> Thank you
> Yao Jiewen
>
>> -----Original Message-----
>> From: Ni, Ray <ray.ni at intel.com>
>> Sent: Thursday, August 19, 2021 6:02 PM
>> To: Yao, Jiewen <jiewen.yao at intel.com>; Michael Kubacki
>> <michael.kubacki at outlook.com>; devel at edk2.groups.io;
>> mikuback at linux.microsoft.com; Chaganty, Rangasai V
>> <rangasai.v.chaganty at intel.com>
>> Subject: RE: [edk2-devel] [edk2-platforms][PATCH v1 1/1]
>> IntelSiliconPkg/PeiSmmAccessLib: Remove S3 requirement
>>
>> Jiewen,
>> We have MmAccess PPI and gEfiMmPeiSmramMemoryReserveGuid HOB.
>>
>> If Standalone IPL can get the SMRAM range from the HOB, what's the purpose
>> of MmAccess PPI?
>>
>> X86 silicon doesn't rely on MmAccess.Close/Lock to close SMRAM anymore.
>> Does ARM need?
>>
>> Michael,
>> Have you considered to remove the dep on MmAccess PPI from standalone MM?
>>
>> Thanks,
>> Ray
>>
>> -----Original Message-----
>> From: Yao, Jiewen <jiewen.yao at intel.com>
>> Sent: Thursday, August 19, 2021 10:02 AM
>> To: Michael Kubacki <michael.kubacki at outlook.com>; devel at edk2.groups.io;
>> Ni, Ray <ray.ni at intel.com>; mikuback at linux.microsoft.com; Chaganty,
>> Rangasai V <rangasai.v.chaganty at intel.com>
>> Subject: RE: [edk2-devel] [edk2-platforms][PATCH v1 1/1]
>> IntelSiliconPkg/PeiSmmAccessLib: Remove S3 requirement
>>
>> The history was that we didn’t need MmAccessPei without S3.
>> MmAccessPei was added for S3 resume purpose only.
>>
>> Today, if there is real use case to rely on MmAccessPei in normal boot path.
>> Then we can add it.
>>
>> I could see the potential impact is: If MmAccessPei changes the SMRAM
>> attribute in normal boot path, it MUST be reflected in the SmramHob. Otherwise,
>> there is inconsistency issue in PEI/DXE, when DXE consumes the SmramHob.
>> This is NOT required in S3, because we don’t run DXE in S3 path.
>>
>> Thank you
>> Yao Jiewen
>>
>>> -----Original Message-----
>>> From: Michael Kubacki <michael.kubacki at outlook.com>
>>> Sent: Thursday, August 19, 2021 2:47 AM
>>> To: devel at edk2.groups.io; Ni, Ray <ray.ni at intel.com>;
>>> mikuback at linux.microsoft.com; Chaganty, Rangasai V
>>> <rangasai.v.chaganty at intel.com>; Yao, Jiewen <jiewen.yao at intel.com>
>>> Subject: Re: [edk2-devel] [edk2-platforms][PATCH v1 1/1]
>>> IntelSiliconPkg/PeiSmmAccessLib: Remove S3 requirement
>>>
>>> Jiewen/Sai, are you thinking about this?
>>>
>>> Thanks,
>>> Michael
>>>
>>> On 8/12/2021 1:20 AM, Ni, Ray wrote:
>>>> Michael,
>>>> I need Jiewen's input on why MmAccess and MmCommunication PPIs were
>>>> not
>>> installed in normal boot path. Without understanding the reason, I
>>> don't have confidence to approve the change.
>>>>
>>>> Sai,
>>>> Do you see other impacts to Intel platforms with this behavior change?
>>>>
>>>> Thanks,
>>>> Ray
>>>>
>>>> -----Original Message-----
>>>> From: devel at edk2.groups.io <devel at edk2.groups.io> On Behalf Of
>>>> Michael
>>> Kubacki
>>>> Sent: Tuesday, August 10, 2021 11:36 PM
>>>> To: devel at edk2.groups.io; Ni, Ray <ray.ni at intel.com>; Chaganty,
>>>> Rangasai V
>>> <rangasai.v.chaganty at intel.com>
>>>> Cc: Yao, Jiewen <jiewen.yao at intel.com>
>>>> Subject: Re: [edk2-devel] [edk2-platforms][PATCH v1 1/1]
>>> IntelSiliconPkg/PeiSmmAccessLib: Remove S3 requirement
>>>>
>>>> Installation is a platform decision. The buried dependency on boot
>>>> mode in this
>>> particular function is just a roadblock platforms have to work around.
>>> The role of this API is to install the PPI.
>>>>
>>>> Thanks,
>>>> Michael
>>>>
>>>> On 8/9/2021 9:47 PM, Ni, Ray wrote:
>>>>> Michael,
>>>>> Allowing the gPeiSmmAccessPpiGuid PPI installation in normal boot
>>>>> will further allow gEfiPeiSmmCommunicationPpiGuid installation in
>>>>> normal path,
>>> while without your change neither of the PPIs is installed in normal boot.
>>>>>
>>>>> + Jiewen for potential security concern.
>>>>>
>>>>> Thanks,
>>>>> Ray
>>>>>
>>>>>> -----Original Message-----
>>>>>> From: Chaganty, Rangasai V <rangasai.v.chaganty at intel.com>
>>>>>> Sent: Tuesday, August 10, 2021 6:46 AM
>>>>>> To: mikuback at linux.microsoft.com; devel at edk2.groups.io
>>>>>> Cc: Ni, Ray <ray.ni at intel.com>
>>>>>> Subject: RE: [edk2-platforms][PATCH v1 1/1]
>>>>>> IntelSiliconPkg/PeiSmmAccessLib: Remove S3 requirement
>>>>>>
>>>>>> Reviewed-by: Sai Chaganty <rangasai.v.chaganty at intel.com>
>>>>>>
>>>>>> -----Original Message-----
>>>>>> From: mikuback at linux.microsoft.com <mikuback at linux.microsoft.com>
>>>>>> Sent: Monday, August 09, 2021 6:40 AM
>>>>>> To: devel at edk2.groups.io
>>>>>> Cc: Ni, Ray <ray.ni at intel.com>; Chaganty, Rangasai V
>>>>>> <rangasai.v.chaganty at intel.com>
>>>>>> Subject: [edk2-platforms][PATCH v1 1/1]
>>>>>> IntelSiliconPkg/PeiSmmAccessLib: Remove S3 requirement
>>>>>>
>>>>>> From: Michael Kubacki <michael.kubacki at microsoft.com>
>>>>>>
>>>>>> REF:https://bugzilla.tianocore.org/show_bug.cgi?id=3539
>>>>>>
>>>>>> PeiInstallSmmAccessPpi() currently requires the boot mode be set
>>>>>> to S3 to
>>> actually install gEfiPeiMmAccessPpiGuid.
>>>>>>
>>>>>> This change removes this requirement in the function
>>>>>> implementation for
>>> two reasons:
>>>>>>
>>>>>> 1. Practical use cases exist to require this PPI in cases other than
>>>>>> the boot mode being set to BOOT_ON_S3_RESUME.
>>>>>>
>>>>>> 2. It is poor API design to implicitly bury this requirement within
>>>>>> a function whose responsibility is to install the PPI. The caller
>>>>>> can easily place arbitrary constraints around whether to call
>>>>>> based on conditions such as the boot mode being
>>>>>> BOOT_ON_S3_RESUME.
>>>>>>
>>>>>> Cc: Ray Ni <ray.ni at intel.com>
>>>>>> Cc: Rangasai V Chaganty <rangasai.v.chaganty at intel.com>
>>>>>> Signed-off-by: Michael Kubacki <michael.kubacki at microsoft.com>
>>>>>> ---
>>>>>>
>>> Silicon/Intel/IntelSiliconPkg/Feature/SmmAccess/Library/PeiSmmAccessLi
>>> b/PeiS
>>> mmAccessLib.c | 12 ------------
>>>>>> 1 file changed, 12 deletions(-)
>>>>>>
>>>>>> diff --git
>>>>>> a/Silicon/Intel/IntelSiliconPkg/Feature/SmmAccess/Library/PeiSmmAc
>>>>>> ces
>>>>>> sLib/PeiSmmAccessLib.c
>>>>>> b/Silicon/Intel/IntelSiliconPkg/Feature/SmmAccess/Library/PeiSmmAc
>>>>>> ces sLib/PeiSmmAccessLib.c index d9bf4fba983e..4df0d695fdaf 100644
>>>>>> ---
>>>>>> a/Silicon/Intel/IntelSiliconPkg/Feature/SmmAccess/Library/PeiSmmAc
>>>>>> ces
>>>>>> sLib/PeiSmmAccessLib.c
>>>>>> +++ b/Silicon/Intel/IntelSiliconPkg/Feature/SmmAccess/Library/PeiS
>>>>>> +++ mmA
>>>>>> +++ cce
>>>>>> +++ ssLib/PeiSmmAccessLib.c
>>>>>> @@ -252,19 +252,7 @@ PeiInstallSmmAccessPpi (
>>>>>> EFI_SMRAM_HOB_DESCRIPTOR_BLOCK *DescriptorBlock;
>>>>>> SMM_ACCESS_PRIVATE_DATA *SmmAccessPrivate;
>>>>>> VOID *HobList;
>>>>>> - EFI_BOOT_MODE BootMode;
>>>>>>
>>>>>> - Status = PeiServicesGetBootMode (&BootMode);
>>>>>> - if (EFI_ERROR (Status)) {
>>>>>> - //
>>>>>> - // If not in S3 boot path. do nothing
>>>>>> - //
>>>>>> - return EFI_SUCCESS;
>>>>>> - }
>>>>>> -
>>>>>> - if (BootMode != BOOT_ON_S3_RESUME) {
>>>>>> - return EFI_SUCCESS;
>>>>>> - }
>>>>>> //
>>>>>> // Initialize private data
>>>>>> //
>>>>>> --
>>>>>> 2.28.0.windows.1
>>>>>
>>>>>
>>>>>
>>>>>
>>>>>
>>>>
>>>>
>>>>
>>>>
>>>>
>>>>
>>>>
>>>>
>>>>
>>>>
>
>
>
>
>
-=-=-=-=-=-=-=-=-=-=-=-
Groups.io Links: You receive all messages sent to this group.
View/Reply Online (#79590): https://edk2.groups.io/g/devel/message/79590
Mute This Topic: https://groups.io/mt/84768258/1813853
Group Owner: devel+owner at edk2.groups.io
Unsubscribe: https://edk2.groups.io/g/devel/unsub [edk2-devel-archive at redhat.com]
-=-=-=-=-=-=-=-=-=-=-=-
More information about the edk2-devel-archive
mailing list