2025-10-01 TSC Meeting
Date
Oct 1, 2025
Disclosures
Participants
Agenda
Past Meeting Minutes Approval
Charter Review
Ratification Plan Review (TSC as Governing Committee)
Additional other business
Presentations
Title | Presenter | File | |
|---|---|---|---|
| 1 | Motion on Winding down of Unified Discovery TG | Philipp Tomsich | |
Votes
Notes & Action Items
Meeting recording and transcript: link
Meeting Summary
Unified DB Charter Approval
The discussion focused on the Unified Database (UDB) charter and its capabilities. Derek explained that UDB aims to capture and organize RISC-V specifications in a machine-readable format, including instruction opcodes, dependencies, and encoding parameters. He highlighted that while UDB serves as the central source of truth for RISC-V specifications, the ISA manuals remain the authoritative source of text. The tool can generate various outputs, such as profile specifications, PRMs, and instruction set simulators, and is being used by Synopsys for product generation. Derek also noted that UDB is currently in a validation phase, ensuring industrial-quality data before further applications in the ecosystem.
The group discussed the comparison between UDB and Sail models, with Derek explaining that UDB generates comparable performance to Sail but is faster due to formal proofs. Andrea mentioned that a detailed comparison between Sail and UDB is being worked on, which will be presented to the TSC. The team also discussed the need for a survey to understand which members are using Sail and UDB, with Tariq noting that Cherry extensively uses Sail due to its development of the tool. Jeff inquired about action items before sending out a vote on the UDBC charter, and Greg suggested adding a statement about the scope of UDB encompassing machine-readable encoding of RISC-V ISA specs and, where feasible, non-ISA specs.
RVA23.1/RVB23.1 Ratification Plan Approval
The TSC discussed the ratification plans for RVA 23.1 and RVB 23.1 profiles, which include five extensions each. David Weaver presented the schedule, which aims for plan approval by October 31st and TSC ratification approval by early March, with ratification at the board by the end of March. The group agreed that Jeff would create an OPA vote with separate ballots for each profile, as members may have different positions. They also discussed the possibility of generating the specifications using Unified DB, which could speed up the process.
The group discussed the need for an independent CSRIND extension for version 23.1, with David Brash questioning its necessity. David Weaver agreed to check and report back to the group, noting that the issue could be addressed outside the ratification plan vote. The meeting minutes from the August 6th and September meetings were approved by the group, with Manu making the motion and David Weaver seconding. The conversation ended with a brief discussion of additional business items.
Motion on Winding down of Unified Discovery TG
The discussion focused on the winding down of the Unified Discovery TG due to a lack of engagement, leadership, and a viable community. Philipp proposed a motion to wind down the TG, which would also eliminate the Unified Discovery waiver, as the TSC could handle this. The group discussed the ecosystem impact, noting that while Unified Discovery was not the only discovery method, its absence would mean individual TGs would need to handle discoverability through various methods like CSRs or WARL fields. Greg clarified that Unified Discovery was intended to be a low-level discovery method, not the sole one, and the group acknowledged the need for further discussion on how architecture extensions could rely on high-level discovery methods like ACPI or device tree.
The group discussed the challenges and future direction of discovery mechanisms in profiles and extensions. David Brash raised concerns about the reliability of extension discovery, particularly for mandatory extensions, and questioned the purpose of RBA23 if dependencies cannot be trusted. Philipp acknowledged these issues and highlighted the need for a low-level discovery mechanism, similar to Unified Discovery's original goal. Derek asked about the fate of the MConfig pointer CSR and the process for standardizing extension discovery methods, to which Greg suggested compliance with the current specification and proposed exploring higher-level discovery methods like device tree and ACPI tables. Greg also proposed focusing on specific use cases, such as external debug, to identify the true need for low-level discovery methods.
The group discussed enhancing the ISA string mechanism to better handle profile names and simplify the current complex format that has become difficult to read. Derek suggested compressing the config file to serve as a machine-readable description of implemented features, which aligns with the goals of unified discovery. Philipp noted that unified discovery was previously developed to replace mconfig and provide a comprehensive description of implemented features. The team considered directing the DTPM SIG to assess the need for an external debug-oriented discovery mechanism and proposed referring a simpler alternative to the Architecture Review Committee for further consideration.
Greg and Philipp agreed to defer the motion to the next TSC meeting, with an OpaVote to be conducted in the meantime.
Industrial/Intelligent Edge SIG exploration meeting at Summit
The group discussed plans for an Industrial/Intelligent Edge SIG, with Andrea proposing to gather interested members at the upcoming summit to discuss its creation.
Tech Journal - First Full-Month Coverage
Rafael provided an update on the tech journal coverage, which is being improved using AI to summarize RISC-V activities.
Details
Unified DB
Discussion:
What does UDB produce?
ISS
Machine readable representation of the spec
Documentation, e.g. full profile specification or for a particular extension
Assembler format
Working on downstream content for binutils, gcc, QEMU
Other information like testsuites (not explored) may be explored
Is there a roadmap for where things are going?
Answer: Presenting at Summit. Should articulate that.Which ISSs would UDB be able to generate?
Only IDL executable description. Sail snippets included (as found).How much of IDL has been covered?
IDL is comparable to Sail in coverage.How does UDB relate to ISA?
ISA will be definitive source of truth. UDB will be the machine-readable version of that.Does UDB extend to custom extensions?
Answer: Yes, Qualcomm has released a custom extension using this.Does UDB handle multi-core?
Answer: It builds as a library and off-loads the capability to the consumer.How does UDB overlap with Sail?
Answer: UDB is a broader umbrella, e.g. it understands overlaps among extensions. Similar in that it models instruction semantics.Why build another ISS?
Answer: IDL started 2 years ago when Sail was not complete. IDL not focused on formal verification and thus is lightweight.Is the intent that UDB replaces Sail?
Answer: It might be an option. UDB has enough information to likely generate a Sail model.We need to provide an objective comparison of the state of Sail and IDL to take back to Board.
The CSC is planning to use UDB to build the CTP (test plan) documentation.
Want send a survey to community about the use of a Golden Reference Model. The survey draft will be sent to the TSC mailing list and then discussed next meeting.
AIs
Can we update the Introduction section to include machine readable ISA deliverable and other HW as feasible? (Derek)
Updates to the Charter around “the standard”. What does this mean? Can it be made more specific? (Derek)
Send OpaVote to TSC for approval (RISC-V Staff)
RVA23.1 Ratification Plan
Minor update to the CTR extension since
Discussion:
Deliverable is the specification which is the base + 5 extensions listed in Plan.
May generate specification with UDB
Do we need CsrInd or not?
Answer: Need to go back and review it.
AIs:
None
Next Steps:
OpaVote to TSC to approve the plan (RISC-V Staff)
RVB23.1
Difference is in the Sscfg
Plans are identical
Next Steps:
OpaVote to TSC to approve the plan (RISC-V Staff)
Unified Discovery TG Close
No leadership candidates from last Call for Candidates.
Draft of the existing specification suggest major work still exists.
Consensus has not been reached in the ecosystem.
Discussion:
Would the cancellation include removing the Waiver?
Answer: No, but we would amend motion to do so since TSC approved it originally.What is the defacto way for identifying an extension?
Answer: We won’t have any. Some look at CSRs, other use WARLs, other methods exist.UD was never intended for user mode. It was intended for low-level identification, e.g. FW. Each OS has its own methods for identifying capabilities, e.g. Linux /proc/cpuinfo.
What is the discovery method in support of Profiles?
Answer: Most of this is user-mode and thus not solved by UD.What happens to the MConfig Pointer CSR?
Answer: CSR is already in spec, but compliance may be as simple as no value.One path forward for low-level understanding would be to use the DTPM SIG to look at this for Debug use case and then roll out. Maybe we will find other use cases.
Another path would be to update ISA strings to allow different bases, e.g. profiles.
UDB might be able to play a roll here.
Next steps:
Ask DTPM SIG to review need, starting with a use case of Debug. (Andy Dellow)
Ask ARC to provide guidance on ISA string simplification (Greg)
Continue discussion on the TSC mailing list (All)
Defer vote to OpaVote
Industrial/Intelligent Edge SIG
This idea came out of the TSC Strategic Priorities work
Would like to gather at Summit to explore how to proceed
Reach out to Andrea, Philipp, Jeff if interested in helping kickstart this.
Andrea will send email to TSC list soliciting interest
RISC-V Tech Journal
Trying to create summary to track what’s being done
Please contribute (open for edits by members)
Let us know if you have any suggestions for improvements
There will NOT be an Overflow Meeting next week