Status at a glance:
- Copy Non-ISA Platform Ratification Template
- keep specification status spreadsheet up-to-date ( projected dates, DoD component effort sizing, DoD tasks your group needs resource help with,...)
- Keep dashboard Non-ISA Extensions On Deck for Freeze Milestone up-to-date
- Develop final charter and plan with the group and send to Chairs (firstname.lastname@example.org) for approval within 4 weeks
- Fill in DoD Plan tab in checklist and send to chairs for approval of any deliverables that don't match the defaults along with any requested waivers
- Freeze-time preliminary POC definition approved by IC
- Develop Rationale document and send to chairs for approval
- Notify TSC and Chairs when you complete this milestone and include any waivers and an updated DoD spreadsheet.
- Plan: RISCV has prioritized RVA22. expect CLIC to possibly be ratified with RVM23 with other embedded extensions like Zfinx, new compressed, rv32e, p-extension. So CLIC still on roadmap but probably next gen RISC-V grouping. Continue working on issues and refining spec.
- Hypervisor support?
- See issue #92 discussion. Lastest discussion: The AIA spec is being developed to be the standard support for interrupts for standard hypervisors. Possible hypervisor directions for fast interrupts for embedded include: 1) a minimal CLIC extension that adds HS mode as another privilege layer in the stack, so the hypervisor could optionally receive interrupts and optionally enable/hide interrupts from lower privilege modes. This would allow the 2-stage translation and other features of the hypervisor to be used in a CLIC-based system, but would not support sending interrupts directly to descheduled guest OS contexts (have to route via hypervisor interrupt). 2) support multiple descheduled CLIC contexts to which interrupts could be sent directly while they were sleeping, which would be a much bigger project. 3) add an extension to AIA (instead of working on CLIC) to somehow reduce interrupt latency using the standard hypervisor stack.
- Impact of N-extension on CLIC? Update CLIC spec to not rely on N-extension but add commentary on if N-extension is ratified how it would work with CLIC.
- Latest Draft Fast Interrupt Specification (v0.9-draft-20220510)
issue #235 - change “exception” to “trap” to match priv spec wording.
issue #233 - mnxti pseudo-code clarification (added meaning of clic.priv,clic.level,clic.id)
issue #225 - bounded time to respond to interrupts
- What's next:
- 41 outstanding issues.
- 31 active spec issues.
- 1 issue (new CSR address) dependent on opcode review
- 5 issues to track Definition of Done tasks
- 5 issues kept open for future enhancements (post rev1.0)
Encoding/OpCode consistency review
- Need to propose new CLIC CSR Registers and addresses
- What's next: when outstanding issues are reduced, start planning for review
- email@example.com - when spec is solid but not a final spec - primarily want to nail down opcode/CSR assignment and have a solid draft spec (but not a final spec ready for official Arch Review)
- Also, to remind people of what gets reviewed (as is appropriate for a given extension), see the following list. In addition to the extension spec, please submit information about the PoCs and about utility/efficiency (although we don't need all the gory detail - a paragraph or so for each can be fine). For items considered to not be consequential, a sentence or so explaining why should suffice.
- Consistency with the RISC-V architecture and philosophy
- Documentation clarity and completeness
- Including proper distinction between normative and non-normative text
- Motivation and rationale for the features, instructions, and CSRs
- Utility and efficiency (relative to existing architectural features and mechanisms)
- Is there enough value or benefit to justify the cost of implementation
- Is the cost in terms of area, timing, and complexity reasonable
- Proof of Concept (PoC)
- Software PoC to ensure feature completeness and appropriateness for intended use cases
- Hardware PoC to demonstrate reasonable implementability
- Inappropriate references to protected IP (i.e. covered by patents, copyright, etc.)
- Deterministic Test plan for the fast-interrupt is available. Discussion on-going on how to add async/undeterministic testing of interrupts.
- YAML config needs to be created. See info here.
- Discussion in Arch tests group to add automation (docker?) to validate check-in so that arch-test-suite is run against sail, spike, gcc/toolchain, with versions used recorded. So sail and spike will need to work for CLIC before CLIC tests can be added to riscv-config github.
Compilers / Toolchains
GCC and Binutils
- No new instructions are added. Needs to be aware CSR names? Need to choose arch string. Given that the key CSRs are in M-mode, it should probably be named something like "Smclic"
- No new instructions are added. Needs to be aware of CSR names?
Though all listed under "simulators", these are actually a collection of formal model / virtual machine / architectural simulators / DV simulators etc.
|Project Name||Base Architecture||Level of implementation||Notes|
RTL simulation, FPGA Implementation, Synthesis
|closed / commercial source https://www.seagate.com/innovation/risc-v/|
|high-performance core||RV32||RTL simulation, FPGA Implementation, Synthesis||closed / commercial source https://www.seagate.com/innovation/risc-v/|
|microcontroller-class core||RV32IMAFC||RTL, fully synthesizable||Apache License, Version 2.0 https://github.com/T-head-Semi/opene906/blob/main/doc/opene906_datasheet.pdf|
|E2/S2 series||RV32/64||RTL, fully synthesizable||https://www.sifive.com/core-designer|
|N22||RV32||RTL, fully synthesizable||http://www.andestech.com/en/products-solutions/andescore-processors/riscv-n22/|
|BM-310/BI-651||RV32/64||RTL, fully synthesizable||https://cloudbear.ru/bm_310.html https://cloudbear.ru/bi_651.html|
|n200/n900/nx900/ux900||RV32/64||RTL, fully synthesizable (ECLIC)||https://www.nucleisys.com/product.php?site=n200 https://www.nucleisys.com/product.php?site=n900|
ABI Extensions (no new ABI required)
- Regular C function that save/restores all caller-save registers
- Inline handler gcc interrupt attribute to always callee-save every register (save as you go)
- EABI Task Group - improve interrupt latency by reducing the number of caller-save registers