ACT Support for Priv 1.13

[jjscheel]

Technical Group

Privileged Spec IC

ratification-pkg

Priv 1.13

Technical Liaison

Greg Favor

Task Category

Arch Tests

Task Sub Category

• gcc
• binutils
• gdb
• intrinsics
• Java
• KVM
• ld
• llvm
• Linux kernel
• QEMU
• Spike

Ratification Target

3Q2024

Statement of Work (SOW)

Component names: Priv 1.13

Requirements:
Provide tests for the following Priv 1.13 updates:

1. Defined the misa.V field to reflect that the V extension has been implemented.
2. Defined the RV32-only medelegh and hedelegh CSRs (1)
3. Allocated interrupt 13 for Sscofpmf LCOFI interrupt (2)
4. Exposed count-overflow interrupts to VS-mode (1,2)
5. Redefined misa.MXL to be read-only, making MXLEN a constant
6. Added the constraint that SXLEN≥UXLEN (3)

The following footnotes clarify level of support:

1. Basic CSR testing of existence, read, and write until Hypervisor implemented.
2. Limited by counter support enablement.
3. May have some challenges with the writability of those fields configurable in the mode in Sail (only).

Note: the following features of Priv 1.13 require ACT infrastructure that prevent their implementation at this time:

• Defined hardware error and software check exception codes.
• Defined the misaligned atomicity granule PMA, superseding the proposed Zam extension.

Deliverables:

• Self-checking assembly language tests.
  Note: because of the nature of the Priv 1.13 tests, they should be written as self-checking so that they can be used both in ACTs and in the Sail model BVTs (build verification tests).
• Coverage models

Acceptance Criteria:

• Assembly language tests that meet the TestFormatSpec AND are self-checking.
• Coverage models using riscv-ctg YAML formatted schema or the compressed coverpoint schema translated into YAML
• Tests pass using the riscof framework

Projected timeframe: (best guess date)
2 person months (part time)

SOW Signoffs: (delete those not needed)

• Task group liaison sign-off
• Development partner sign-off
• ACT SIG sign-off (if ACT work)

Waiver

• Freeze
• Ratification

Pull Request Details

No response

[jjscheel]

@allenjbaum, I'd love your feedback on this SOW.

[jjscheel]

Per discussion in the 5/14 DevPartner meeting, I've updated the SOW description in the first entry to require the ACTs to be self-checking so that they may be used by Sail as BVTs.

@billmcspadden-riscv and @allenjbaum, please review and let me know if anything else should be adjusted.

[allenjbaum]

Finally getting a moment to breathe and review this. #4 also need footnote1: need hypervisor support #6 footnote 3 is misleading, in my opinion. The issue is entirely Sail configurability. RE: tests should be self checking, because ACTs compare the Sail results with DUT results, and if Sail can't be configured, but the DUT can be, you'll get a mismatch. If Sail can' t be configured, they're not useful for BVTs. You can make them self checking, and they might work for some DUTs, but not for others. The may not even work for Sail vs. Spike..... The coverage models can be riscv-ctg YAML or the compressed coverpoint schema translated into YAML

…

On Thu, May 9, 2024 at 9:57 AM Jeff Scheel ***@***.***> wrote: @allenjbaum <https://github.com/allenjbaum>, I'd love your feedback on this SOW. — Reply to this email directly, view it on GitHub <#43 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AHPXVJU5TL7ZX5MEED5FYM3ZBOTF3AVCNFSM6AAAAABHM7GPSGVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDCMBTGA2DONJVGE> . You are receiving this because you were mentioned.Message ID: ***@***.***>

[jjscheel]

Thanks, @allenjbaum. My responses:

#4 also need footnote1: need hypervisor support

Added.

#6 footnote 3 is misleading, in my opinion. The issue is entirely
Sail configurability.

I extended statement to read "3. May have some challenges with the writability of those fields configurable in the mode in Sail (only).".

RE: tests should be self checking, because ACTs compare the Sail results
with DUT results,
and if Sail can't be configured, but the DUT can be, you'll get a mismatch.
If Sail can' t be configured, they're not useful for BVTs.
You can make them self checking, and they might work for some DUTs, but not
for others.
The may not even work for Sail vs. Spike.....

I'm not sure I follow. Are you arguing the tests for Priv 1.13 CANNOT be self-checking? If so, I'll need you and @billmcspadden-riscv to discuss this more.

The coverage models can be riscv-ctg YAML or the compressed coverpoint
schema translated into YAML

Will update.

[allenjbaum]

IT is difficult to make any test self checking if it has multiple possible answers, Basically, you need to have a separate test for each possibility, and either - an architectural method for discovering which of the possibilities is implemented by a DUT, or - a way to convey which possibility a DUT implements that is accessible to the test (e.g. riscv-config YAML which can pass parameters to riscof. IF Sail doesn't have the configurability for some possibility, there is no point in running the test as part of Sail C/I -tt will always fail the test for that possibility (and we should be testing all possibilities)

…

------- Another interesting development is using the Verilog coverpoint definition. Umer is looking to this. It's a major change, but it will be worth it if we can get it to work.

On Fri, May 17, 2024 at 8:23 AM Jeff Scheel ***@***.***> wrote: Thanks, @allenjbaum <https://github.com/allenjbaum>. My responses: #4 <#4> also need footnote1: need hypervisor support Added. #6 <#6> footnote 3 is misleading, in my opinion. The issue is entirely Sail configurability. I extended statement to read "3. May have some challenges with the writability of those fields configurable in the mode in Sail (only).". RE: tests should be self checking, because ACTs compare the Sail results with DUT results, and if Sail can't be configured, but the DUT can be, you'll get a mismatch. If Sail can' t be configured, they're not useful for BVTs. You can make them self checking, and they might work for some DUTs, but not for others. The may not even work for Sail vs. Spike..... I'm not sure I follow. Are you arguing the tests for Priv 1.13 *CANNOT* be self-checking? If so, I'll need you and @billmcspadden-riscv <https://github.com/billmcspadden-riscv> to discuss this more. The coverage models can be riscv-ctg YAML or the compressed coverpoint schema translated into YAML Will update. — Reply to this email directly, view it on GitHub <#43 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AHPXVJRKFQSVA6DGQJFVO3DZCYOFLAVCNFSM6AAAAABHM7GPSGVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDCMJXHAZTONRSGM> . You are receiving this because you were mentioned.Message ID: ***@***.***>

[jjscheel]

No resources at this time.

[pz9115]

FYI, Update Binutils support for Priv 1.13

https://sourceware.org/git/?p=binutils-gdb.git;a=commit;h=433372af69804bd134f250e71479860c33a51c39