[AMD Official Use Only - General] See my answer below,
> -----Original Message----- > From: devel@edk2.groups.io <devel@edk2.groups.io> On Behalf Of > Konstantin Aladyshev via groups.io > Sent: Friday, September 1, 2023 12:02 AM > To: Chang, Abner <abner.ch...@amd.com> > Cc: devel@edk2.groups.io; disc...@edk2.groups.io > Subject: Re: [edk2-devel] [edk2-discuss] PLDM messages via MCTP over KCS > > Caution: This message originated from an External Source. Use proper caution > when opening attachments, clicking links, or responding. > > > As I've said there is nothing like "KCS completion code" in the MCTP > over KCS binding specification > (https://www.dmtf.org/sites/default/files/standards/documents/DSP0254_1 > .0.0.pdf). > The response packet should have the same structure as a request. I.e. > a packet with MANAGEABILITY_MCTP_KCS_HEADER and PEC. > > Currently I'm writing "MCTP over KCS" binding for the OpenBMC libmctp > project. So I can send whatever I want, I don't think my output would > be any useful to you. But I've asked this question in the community > and they also confirmed that the response packet has the same > structure. > (https://discord.com/channels/775381525260664832/7787906385638850 > 86/1146782595334549554) > > Now I'm a little bit confused about the `KcsTransportSendCommand` > function. It has the following arguments for the function output: > ``` > OUT UINT8 *ResponseData OPTIONAL, > IN OUT UINT32 *ResponseDataSize OPTIONAL > ``` > Should we include MCTP_TRANSPORT_HEADER/MCTP_MESSAGE_HEADER to > this > output or not? If the MCTP KCS packet for host->BMC and BMC->host are in the same structure, then yes, the response data from BMC should includes MCTP_TRANSPORT_HEADER/MCTP_MESSAGE_HEADER in my opinion, as this is defined in MCTP base protocol. So let me explain the implementation for MCTP over KCS and what do we miss now. A. MCTP protocol driver linked with KCS Transport interface library - In MCTP protocol driver, if the transport interface is KCS then 1. MCTP protocol driver builds up the MCTP KCS transport header, which is DefBody, NetFunc and ByeCount 2. MCTP protocol driver builds up the MCTP payload 3. MCTP protocol driver builds up the MCTP KCS transport trailer, which is PEC. Above three steps are already implemented. PEC is calculated by MCTP protocol driver and should be verified by KCS using the same algorithm in my understanding of spec. B. In KCS Transport interface library 1. KCS Transport interface library sends the transport header got from TransportToken. Same behavior for IPMI protocol, but different content. KCS Transport interface library doesn't have to understand this. 2. KCS Transport interface library sends the payload 3. KCS Transport interface library sends the transport trailer got from TransportToken. Same behavior for IPMI protocol, but different content. KCS Transport interface library doesn't have to understand this. Above three steps are already implemented. Then, if Manageability protocol is MCTP, we skip reading responses header (Not implemented) For reading response data 1. If the ResponseData and ResponseSize is valid in the given TransportToken, then we read ResponseSize data from KCS. (Already implemented) 2. if Manageability protocol is MCTP, then we skip reading responses header again (Not implemented) Now the response is returned to MCTP protocol driver C. In MCTP protocol driver, we expect to get the whole MCTP over KCS packet response, that includes DefBody, NetFunc and ByeCount, MCTP message and PEC. 1. MCTP protocol driver verifies the returned PEC with the payload. 2. Strip out DefBody, NetFunc, ByeCount and PEC and then returns it to upper layer (e.g., MCTP transport interface library). Returns only MCTP Transport header and MCTP packet payload as it shows in MCTP base protocol spec. Above is not implemented D. In MCTP transport interface library, we can strip out MCTP transport header and then return it to upper layer (e.g., PLDM protocol driver). Above is not implemented. E. In PLDM protocol driver, 1. we verify the Message Integrity Check if the Message Type requests it. 2. we can remove MCTP message type then return it to upper layer (e.g., PLDM SMBIOS transfer) Above is not implemented. We didn’t implement BMC->Host in step C, D and E as our current demand is to send the SMBIOS table to BMC, which doesn't require the response data if I am not wrong. Let me know if it is problematic in the above process. Thanks and regards, Abner > > Best regards, > Konstantin Aladyshev > > On Thu, Aug 31, 2023 at 6:52 PM Chang, Abner <abner.ch...@amd.com> > wrote: > > > > [AMD Official Use Only - General] > > > > But wait, wee my another comment below, > > > > > -----Original Message----- > > > From: Chang, Abner > > > Sent: Thursday, August 31, 2023 11:42 PM > > > To: devel@edk2.groups.io; aladyshe...@gmail.com > > > Cc: disc...@edk2.groups.io > > > Subject: RE: [edk2-devel] [edk2-discuss] PLDM messages via MCTP over > KCS > > > > > > > > > > > > > -----Original Message----- > > > > From: devel@edk2.groups.io <devel@edk2.groups.io> On Behalf Of > > > > Konstantin Aladyshev via groups.io > > > > Sent: Thursday, August 31, 2023 10:57 PM > > > > To: Chang, Abner <abner.ch...@amd.com> > > > > Cc: disc...@edk2.groups.io; devel@edk2.groups.io > > > > Subject: Re: [edk2-devel] [edk2-discuss] PLDM messages via MCTP over > KCS > > > > > > > > Caution: This message originated from an External Source. Use proper > > > caution > > > > when opening attachments, clicking links, or responding. > > > > > > > > > > > > > (I don see what is the response header for MCTP KCS in spec though, > does > > > it > > > > mention the KCS response?). > > > > > > > > The spec doesn't explicitly mention that the format of a send and > > > > response packets differ. So I assume it is the same and it is > > > > described at the "Figure 1 – MCTP over KCS Packet Format" > > > > > > > > (https://www.dmtf.org/sites/default/files/standards/documents/DSP0254_1 > > > > .0.0.pdf) > > > > Therefore the format of a response would look like this: > > > > ``` > > > > MANAGEABILITY_MCTP_KCS_HEADER > > > > (https://github.com/tianocore/edk2- > > > > > > > > platforms/blob/master/Features/ManageabilityPkg/Include/Library/Managea > > > > bilityTransportMctpLib.h) > > > > MCTP_TRANSPORT_HEADER > > > > > > > > (https://github.com/tianocore/edk2/blob/master/MdePkg/Include/Industry > > > > Standard/Mctp.h) > > > > MCTP_MESSAGE_HEADER > > > > > > > > (https://github.com/tianocore/edk2/blob/master/MdePkg/Include/Industry > > > > Standard/Mctp.h) > > > > < response data> > > > What do you see the KCS response from BMC? You probably right as the > > > header and trailer are labeled in different colors 😊. Could you please > enable > > > the debug message to capture it? > > > > > > > PEC > > > > (Probably we need to define > MANAGEABILITY_MCTP_KCS_TRAILER) > > > > ``` > > > We have MANAGEABILITY_MCTP_KCS_HEADER defined but no > > > MANAGEABILITY_MCTP_KCS_TRAILER as it is hardcoded to one byte. > > > If the KCS response is PEC, then yes, we can create > > > MANAGEABILITY_MCTP_KCS_TRAILER for KCS transport interface. > > > > In the implementation, PEC is calculated in MCTP protocol and send through > KCS as the KCS packet trailer. So, when we send the MCTP request through > KCS, KCS shouldn't respond the PEC to upper stack, right? I still think the > response should be the KCS completion code. The debug message from your > end may help to clarify this as your BMC has the MCTP KCS implementation. > > > > Thanks > > Abner > > > > > > > > > > > > > So in the "KcsTransportSendCommand" > > > > (https://github.com/tianocore/edk2- > > > > > platforms/blob/14553d31c72afa7289f6a2555b6e91f4f715a05a/Features/ > > > > > > > > ManageabilityPkg/Library/ManageabilityTransportKcsLib/Common/KcsComm > > > > on.c#L414) > > > > we can check if we transfer is MCTP (based on > > > > "TransportToken->ManagebilityProtocolSpecification == MCTP" like > > > > you've suggested) and handle response accordingly. > > > > > > > > But which headers should we check in this function? Only > > > > > > > > MANAGEABILITY_MCTP_KCS_HEADER/MANAGEABILITY_MCTP_KCS_TRAILER > > > > ? > > > Yes, only check header and trailer for transport interface. > > > > > > > What about MCTP_TRANSPORT_HEADER/MCTP_MESSAGE_HEADER? > Do > > > we > > > > need to > > > > check them here as well? Or do we need to check them somewhere > upper > > > > the call stack? > > > We should leave this to MCTP protocol driver as this is belong to protocol > > > layer, the upper layer stack. > > > > > > Thanks > > > Abner > > > > > > > > > > > Best regards, > > > > Konstantin Aladyshev > > > > > > > > On Thu, Aug 31, 2023 at 7:59 AM Chang, Abner > <abner.ch...@amd.com> > > > > wrote: > > > > > > > > > > [AMD Official Use Only - General] > > > > > > > > > > Hi Aladyshev, > > > > > > > > > > > -----Original Message----- > > > > > > From: Konstantin Aladyshev <aladyshe...@gmail.com> > > > > > > Sent: Wednesday, August 30, 2023 11:09 PM > > > > > > To: Chang, Abner <abner.ch...@amd.com> > > > > > > Cc: disc...@edk2.groups.io; devel@edk2.groups.io > > > > > > Subject: Re: [edk2-discuss] PLDM messages via MCTP over KCS > > > > > > > > > > > > Caution: This message originated from an External Source. Use proper > > > > caution > > > > > > when opening attachments, clicking links, or responding. > > > > > > > > > > > > > > > > > > Hi! > > > > > > > > > > > > I've started to implement MCTP over KCS binding for the libmctp > > > > > > (https://github.com/openbmc/libmctp) and test it with the current > code > > > > > > in the ManageabilityPkg. > > > > > > > > > > > > I was able successfully send the MCTP packet to the BMC, but right > now > > > > > > I'm having some troubles with receiving the answer back. > > > > > > > > > > > > I think I've found some bug in the `KcsTransportSendCommand` code. > > > > > > > > > > > > After it sends data over KCS in expects a responce starting with a > > > > > > 'IPMI_KCS_RESPONSE_HEADER' > > > > > > https://github.com/tianocore/edk2- > > > > > > > > > > > platforms/blob/14553d31c72afa7289f6a2555b6e91f4f715a05a/Features/ > > > > > > > > > > > > > > ManageabilityPkg/Library/ManageabilityTransportKcsLib/Common/KcsComm > > > > > > on.c#L476 > > > > > > > > > > > > Isn't it wrong, assuming that the right header in case of MCTP > > > > > > should > > > > > > be 'MANAGEABILITY_MCTP_KCS_HEADER' ? > > > > > > > > > > > > I guess the 'IpmiHelperCheckCompletionCode' check after the data > > > > > > receive is also not relevant for the MCTP. > > > > > > > > > > > > > > > This is something I don’t really sure as I can't verify the response > payload > > > > because our BMC doesn't have the code to handle MCTP over KCS > > > command. > > > > However it is appreciated if community can help to verify this. As I can > > > > remember, I can see the return KCS status is 0xC1, the invalid command. > > > Thus I > > > > think if we do a MCTP over KCS, the first response is still KCS response > > > header. > > > > > This is not what do you see on the BCM it does support MCTP over > KCS? If > > > > so, then I would like to have your help to correct this code. > > > > > > > > > > > > > > > > > Since 'ManageabilityTransportKcsLib' can be used both for IPMI and > > > > > > MCTP, how should we deal with this? > > > > > If KcsCommon.c, we can have different code path for the given protocol > > > > GUID. e.g., if (TransportToken->ManagebilityProtocolSpecification == > MCTP). > > > > > Then skip reading the KCS_REPOSNSE_HEADER or to read the > > > > MCTP_RESPONSE_HEADER (I don see what is the response header for > MCTP > > > > KCS in spec though, does it mention the KCS response?). > > > > > > > > > > Thanks > > > > > Abner > > > > > > > > > > > > > > > > > Best regards, > > > > > > Konstantin Aladyshev > > > > > > > > > > > > On Wed, Aug 23, 2023 at 5:18 AM Chang, Abner > > > > <abner.ch...@amd.com> > > > > > > wrote: > > > > > > > > > > > > > > [AMD Official Use Only - General] > > > > > > > > > > > > > > Please see my answers inline. > > > > > > > > > > > > > > > -----Original Message----- > > > > > > > > From: disc...@edk2.groups.io <disc...@edk2.groups.io> On > Behalf > > > Of > > > > > > > > Konstantin Aladyshev via groups.io > > > > > > > > Sent: Wednesday, August 23, 2023 1:54 AM > > > > > > > > To: Chang, Abner <abner.ch...@amd.com> > > > > > > > > Cc: disc...@edk2.groups.io; devel@edk2.groups.io > > > > > > > > Subject: Re: [edk2-discuss] PLDM messages via MCTP over KCS > > > > > > > > > > > > > > > > Caution: This message originated from an External Source. Use > proper > > > > > > caution > > > > > > > > when opening attachments, clicking links, or responding. > > > > > > > > > > > > > > > > > > > > > > > > Thanks for the answer! > > > > > > > > > > > > > > > > I was a little bit confused about the part, that in the same > > > > > > > > package > I > > > > > > > > actually need to provide different library implementations for > > > > > > > > the > > > > > > > > same 'ManageabilityTransportLib', thanks for the clarification! > > > > > > > > I think your DSC example should go into the package > documentation. > > > > > > > Yes, this is a good idea. I will update it. > > > > > > > > > > > > > > > > > > > > > > > As for me, I'm working with the OpenBMC distribution > > > > > > > > (https://github.com/openbmc/openbmc) and my goal is to > transfer > > > > data > > > > > > > > from the BIOS to the BMC via MCTP/PLDM. > > > > > > > > Currently there is no solution for the MCTP over KCS binding in > Linux, > > > > > > > > so I need to add this support: > > > > > > > > - either to the MCTP userspace library > > > > > > > > (https://github.com/openbmc/libmctp) [old OpenBMC way, but > > > > probably > > > > > > > > easier] > > > > > > > > - or to the MCTP kernel binding > > > > > > > > > (https://github.com/torvalds/linux/tree/master/drivers/net/mctp) > > > > > > > > [modern mctp Linux driver approach] > > > > > > > > > > > > > > > > Both don't sound like an easy task, so can I ask, what MC (i.e. > > > > > > > > management controller) device and firmware do you use on the > other > > > > > > > > side of the MCTP KCS transmissions? > > > > > > > > > > > > > > We use OpenBMC as well, but as you mention there are some > missing > > > > pieces > > > > > > to fully support manageability between host and BMC. > > > > > > > We don’t have code to handle MCTP IPMI either, the edk2 > > > > ManageabilityPkg > > > > > > provides the framework while MCTP/PLDM/KCS implementation > > > provides > > > > a > > > > > > sample other than IPMI/KCS to prove the flexibility of > ManageabilityPkg. > > > > > > > Actually, MCTP over KCS is not supported in our BMC firmware yet, > > > thus > > > > > > BMC just returns the invalid command. However, the transport > > > framework > > > > > > has been verified to make sure the implementation works fine as > expect. > > > > > > > We need help from community to provide more manageability > > > protocols > > > > and > > > > > > transport interface libraries to this package. > > > > > > > > > > > > > > > > > > > > > > > You've also mentioned PLDM SMBIOS, isn't it covered by the > > > > > > > > https://github.com/tianocore/edk2- > > > > > > > > > > > > > > > > > > > > > > platforms/blob/master/Features/ManageabilityPkg/Universal/PldmSmbiosTr > > > > > > > > ansferDxe/PldmSmbiosTransferDxe.c > > > > > > > > ? > > > > > > > Ah hah, yes I forget I upstream it. > > > > > > > > > > > > > > Please just feel free to send patch to make more functionalities > > > > > > > to > this > > > > > > package. > > > > > > > Thanks > > > > > > > Abner > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > Best regards, > > > > > > > > Konstantin Aladyshev > > > > > > > > > > > > > > > > On Tue, Aug 22, 2023 at 7:26 PM Chang, Abner > > > > > > <abner.ch...@amd.com> > > > > > > > > wrote: > > > > > > > > > > > > > > > > > > [AMD Official Use Only - General] > > > > > > > > > > > > > > > > > > Hi Aladyshev, > > > > > > > > > We use library class to specify the desire transport > > > > > > > > > interface for > the > > > > > > > > management protocol, such as MCTP, PLDM and IPMI. This way > we > > > can > > > > > > > > flexibly support any transport interface for the management > protocol. > > > > > > > > > > > > > > > > > > Here is the example of using ManageabilityPkg, which is PLDM > over > > > > MCTP > > > > > > > > over KCS. > > > > > > > > > > > > ManageabilityPkg/Universal/IpmiProtocol/Dxe/IpmiProtocolDxe.inf > > > > { > > > > > > > > > <LibraryClasses> > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > ManageabilityTransportLib|ManageabilityPkg/Library/ManageabilityTranspor > > > > > > > > tKcsLib/Dxe/DxeManageabilityTransportKcs.inf > > > > > > > > > } > > > > > > > > > > > > > ManageabilityPkg/Universal/MctpProtocol/Dxe/MctpProtocolDxe.inf { > > > > > > > > > <LibraryClasses> > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > ManageabilityTransportLib|ManageabilityPkg/Library/ManageabilityTranspor > > > > > > > > tKcsLib/Dxe/DxeManageabilityTransportKcs.inf > > > > > > > > > } > > > > > > > > > > > > > ManageabilityPkg/Universal/PldmProtocol/Dxe/PldmProtocolDxe.inf { > > > > > > > > > <LibraryClasses> > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > ManageabilityTransportLib|ManageabilityPkg/Library/ManageabilityTranspor > > > > > > > > tMctpLib/Dxe/DxeManageabilityTransportMctp.inf > > > > > > > > > } > > > > > > > > > > > > > > > > > > So you can implement ManageabilityTransport library for either > > > > industry > > > > > > > > standard or proprietary implementation for the specific > management > > > > > > > > protocol. > > > > > > > > > > > > > > > > > > BTW, We do have PLDM SMBIOS over MCTP implementation but > > > not > > > > > > > > upstream yet. > > > > > > > > > > > > > > > > > > Hope this information helps. > > > > > > > > > Thanks > > > > > > > > > Abner > > > > > > > > > > > > > > > > > > > > > > > > > > > > -----Original Message----- > > > > > > > > > > From: disc...@edk2.groups.io <disc...@edk2.groups.io> On > > > > Behalf Of > > > > > > > > > > Konstantin Aladyshev via groups.io > > > > > > > > > > Sent: Tuesday, August 22, 2023 7:00 PM > > > > > > > > > > To: discuss <disc...@edk2.groups.io>; devel@edk2.groups.io > > > > > > > > > > Subject: [edk2-discuss] PLDM messages via MCTP over KCS > > > > > > > > > > > > > > > > > > > > Caution: This message originated from an External Source. > > > > > > > > > > Use > > > > proper > > > > > > > > caution > > > > > > > > > > when opening attachments, clicking links, or responding. > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > Hi! > > > > > > > > > > > > > > > > > > > > I'm trying to build `ManageabilityPkg` from the > > > > > > > > > > edk2-platforms > > > > > > > > > > repo to issue PLDM messages via MCTP over KCS. Is it > > > > > > > > > > possible > > > with > > > > > > > > > > the current code? I see all the building blocks, but have > > > > > > > > > > trouble > > > > > > > > > > putting it all together. > > > > > > > > > > > > > > > > > > > > The main question that bothers me is what implementation > should > > > I > > > > set > > > > > > > > > > for the `ManageabilityTransportLib`? > > > > > > > > > > By default it is set to dummy > `BaseManageabilityTransportNull.inf` > > > > > > > > > > (https://github.com/tianocore/edk2- > > > > > > > > > > > > > > > > > > > > > > > > > > > > > platforms/blob/master/Features/ManageabilityPkg/ManageabilityPkg.dsc). > > > > > > > > > > > > > > > > > > > > On one case to get PLDM via MCTP it looks that I need to > > > > > > > > > > set it > to > > > > > > > > > > `DxeManageabilityTransportMctp.inf` > > > > > > > > > > ManageabilityTransportLib| > > > > <...>/DxeManageabilityTransportMctp.inf > > > > > > > > > > (https://github.com/tianocore/edk2- > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > platforms/blob/master/Features/ManageabilityPkg/Library/ManageabilityTra > > > > > > > > > > nsportMctpLib/Dxe/DxeManageabilityTransportMctp.inf) > > > > > > > > > > > > > > > > > > > > But on the other case if I want MCTP over KCS I need to set > > > > > > > > > > it to > > > > > > > > > > `DxeManageabilityTransportKcs.inf` > > > > > > > > > > ManageabilityTransportLib| > > > <...>/DxeManageabilityTransportKcs.inf > > > > > > > > > > (https://github.com/tianocore/edk2- > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > platforms/blob/master/Features/ManageabilityPkg/Library/ManageabilityTra > > > > > > > > > > nsportKcsLib/Dxe/DxeManageabilityTransportKcs.inf) > > > > > > > > > > > > > > > > > > > > What is the right way to resolve this? > > > > > > > > > > > > > > > > > > > > There are no platforms in the repo that actually implement > > > > PLDM/MCTP > > > > > > > > > > functionality, so there is no example that I can use as a > reference. > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > Best regards, > > > > > > > > > > Konstantin Aladyshev > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > -=-=-=-=-=-=-=-=-=-=-=- Groups.io Links: You receive all messages sent to this group. View/Reply Online (#108226): https://edk2.groups.io/g/devel/message/108226 Mute This Topic: https://groups.io/mt/100897530/21656 Group Owner: devel+ow...@edk2.groups.io Unsubscribe: https://edk2.groups.io/g/devel/unsub [arch...@mail-archive.com] -=-=-=-=-=-=-=-=-=-=-=-
