On Wed, Oct 14, 2026 at 02:15:19PM +0200, Michal Simek wrote: > Hi, > > On 10/14/26 10:12, Rahul Pathak wrote: > > Hi Michal, > > > > It is not correct to say that SCMI already exists and so RPMI is not > > needed. If you read the RPMI specification you will find that the two > > differ significantly. I don't mean overlapping names, but how things > > are defined and implemented in the spec. Compare the transport, > > compare the defined messages. > > > > RPMI defines many service groups that have no counterpart in SCMI like > > System MSI/System IRQ, TEE, RAS Agent, Management Mode, CPPC, > > Logging, Request Forward and others. Together they cover use cases > > ranging from typical systems to confidential computing and > > virtualization, all in one specification. > > > > There is also a difference in governance, which is visible from the > > licences of the two specifications. > > > > RPMI is under CC-BY 4 license which allows anyone to adapt it and share > > modified versions with attribution, and changes are developed in > > the open under RISC-V International. The development does not happen > > behind closed door but rather any indivudial/organization can also take > > part in > > the discussions. > > I don't think license is an argument here because someone decided not to > talk to ARM about it. Based on my information none really talked to ARM to > change that license for SCMI (if that was an issue) or move it to neutral > place. > I have already talked to ARM about it. > > > I think open source projects always give users/vendors freedom > > to use any protocol which suits their platform. It must not be dictated > > that everyone must stick to one interface. As Heinrich pointed out that > > U-Boot > > already supports alternatives for various things and > > Linux and OpenSBI already support RPMI. Adding RPMI support to U-Boot > > does not take anything away from SCMI. > > > > Platforms that use SCMI can continue to do so, and platforms > > built on RPMI get the support they need. > > > > So RPMI is not another interface that does the same thing. It is a > > broader specification, developed openly by the community that uses > > it in which you and anyone can also take part. > > It wasn't my cover letter which described > "The goals and purpose of RPMI are similar to ARM's SCMI." > > I am just saying that it looks like that you are creating parallel firmware > interface instead of extending/reusing existing one without really even > trying to talk to other vendors. I would prefer if we can work together on > one spec and extend it to cover both architectures. > > When you have heterogeneous systems where you have mix of ARM and RISC-V > cores make no sense to have SCMI and RPMI servers both running to serve the > same purpose.
I believe we have ended up in an "either/or" situation. It would have been great if N years ago, SCMI was adopted for RISC-V. I hope one or the other, and not a 3rd one, is adopted for new architectures (looking at loongarch...). But for today, I think U-Boot will need to follow along with what the kernel has already been doing and have RPMI for the RISC-V platforms which don't use SCMI. -- Tom
signature.asc
Description: PGP signature
