Add CSR YAML files for Sscofpmf: scountovf

[syedowaisalishah]

Extension CSR YAML files for the Sscofpmf extension.

CSR YAMLs for:

• SCOUNTOVF

Closes Add Sscofpmf CSR #569

[ThinkOpenly]

@syedowaisalishah, did you intend to close this?

[syedowaisalishah]

@syedowaisalishah, did you intend to close this?

Yes, I closed it because the regression test failed in this PR. I’ll address the issues and reopen it once everything passes.

[syedowaisalishah]

I've made all the suggested changes — just haven’t added the sw_read() yet due to an error. Will follow up on that soon.

[ThinkOpenly]

You can either do this by hand, your own script that doesn't get checked in, or using a .layout file like arch/csr/Zicntr/mcountinhibit.layout. Any way is fine with me.

Derek was suggesting here to either explicitly describe all of the bits OF3-OF31, or parameterize their description using a layout like:

  <%- (3..31).each do |of_num| -%>
  OF<%= of_num %>:
    location: <%= of_num %>
[...]

Since you've already done the former (describing each of them), you do not need the ".layout" file.

(For reference, if you did choose to use a "layout" file, you'd need to have it explicitly handled in the Rakefile similarly to mcountinhibit. That has me wondering if we should make that handling more automatic... a thought for another day, perhaps.)

[syedowaisalishah]

@ThinkOpenly , @dhower-qc Added .layout file for OF3–OF31 YAML generation and updated the Rakefile accordingly (similar to mcountinhibit). Verified functionality with rake gen:arch.

[dhower-qc]

Great, almost there. When using layout files, we also check in the generated files so that they are accessible via browsing GitHub. So, if you can add scountovf.yaml to the PR, then I think this looks good.

[dhower-qc]

This (and others) is actually:

return CSR[mhpmevent<%= of_num %>].EN ? # ...

That's why the regressions are failing.

[syedowaisalishah]

Thanks @dhower-qc — I updated the type() and reset_value() lines in scountovf.layout as you suggested, using CSR[mhpmevent<%= of_num %>].EN instead of the bare mhpmevent<%= of_num %>.EN
However, I’m encountering issues:

When I run regressions and smoke test locally in VS Code, the test appears to hang during type-checking. It stalls here with no further output:
Type checking IDL code for rv32... [INFO] 2025-05-06 13:43:32: Allocating more SymbolTables Instructions: |========================================================| CSRs: |===================== |
When I pushed the changes, 4 tests failed in CI.
if I run the smoke without including the YAML file scountovf, the test completes successfully and all checks pass.

[dhower-qc]

The regressions are failing because it can't find CSR[mhpmeventX].OF. That's because those fields don't exist yet...they are added by this extension!

We'll need to modify mhpmeventN.layout to add the fields [OF, MINH, SINH, UINH, VSINH, VUINH] to the event CSRs.

A few things to note: the OF and MINH fields are definedBy Sscofpmf. The SINH field is definedBy: allOf: [S, Sscofpmf]. Similar for the remaining *INH fields.

[ThinkOpenly]

The regressions are failing because it can't find CSR[mhpmeventX].OF. That's because those fields don't exist yet...they are added by this extension!

Why would the regression tests fail for this PR because of something added by this PR? (Maybe I don't understand.)

I wonder if it could be because we're still defining everything twice, until arch/csr/Sscofpmf/scountovf.yaml is removed?

[syedowaisalishah]

The regressions are failing because it can't find CSR[mhpmeventX].OF. That's because those fields don't exist yet...they are added by this extension!

We'll need to modify mhpmeventN.layout to add the fields [OF, MINH, SINH, UINH, VSINH, VUINH] to the event CSRs.

A few things to note: the OF and MINH fields are definedBy Sscofpmf. The SINH field is definedBy: allOf: [S, Sscofpmf]. Similar for the remaining *INH fields.

I’ve updated mhpmeventN.layout to include the new fields (OF, MINH, SINH, UINH, VSINH, VUINH) as described, and regenerated the mhpmevent3–31 YAML files accordingly. However, the regressions are still failing, and when I run regress in VSCode, it hangs without producing any output.

[dhower-qc]

Thanks for the hard work with this one!