Re: RDE Enablement

[Deepak Kodihalli <deepak.kodihalli.83@gmail.com>]

Hi Mike,

On Fri, Jun 11, 2021 at 2:29 AM Garrett, Mike (HPE Server Firmware)
<mike.garrett@hpe.com> wrote:
>
> Great!  We are interested in RDE becoming ubiquitous on adapter cards so that Redfish configuration of storage and networking doesn't have to include adapter specific code.  A good success metric would be the ability to create a storage logical volume using nothing but standard Redfish operations.  In pursuit of this, a solid open source OpenBMC implementation seems like a good way to put RDE on the right footing.
>
> My initial thoughts would be to build an RDE systemd service on top of the existing pldmd service and have an upper interface into the standard dbus interfaces for inventory, status, and configuration.  However, I suspect there's some additional dbus work needed to join RDE to bmcweb because there will be a need to dynamically change the Redfish model and support things like Actions.  A requirement for this to work would be the ability to discover PLDM devices and assign IDs (MCTP EID) in order to interrogate them for RDE capabilities.  It is unclear to me that the current PLDM and MCTP code handles discovery or if it assumes devices.

The current PLDM stack upstream is mostly for the BMC as a PLDM
responder, but there is work underway to make the BMC act as a PLDM
requester, discover devices, etc. You should start seeing patches for
these shortly (there's already
https://gerrit.openbmc-project.xyz/c/openbmc/pldm/+/43465).

What is the reasoning behind having RDE as a separate service as
opposed to being part of the PLDM service? I think one of the key
aspects would be the Redfish to RDE bridge (and vice versa). We're
also interested in RDE. I'd be happy to discuss/review further on the
ML or on Gerrit (if you're planning to put up a design doc).

Thanks,
Deepak