Add vCLIC support

[ezelioli]

This PR adds support for the custom vCLIC extension.

It is a replacement of the previous PR, with only changes strictly relevant to vCLIC. It also makes the extension parametrized on an additional RVVCLIC parameter, which depends on RVH and RVSCLIC parameters and can be used to enable / disable the virtualization extension support.

The main changes are:

1. CSRs: Implemented HGEIE write logic and functionality. Added vCLIC-specific VSTVT and VSINTTHRESH registers.
2. CLIC interface: Adapted core interface and CLIC controller logic to vCLIC.

[niwis]

@ezelioli can you rebase onto pulp-v2?

[ezelioli]

@niwis currently the cheshire-integration check fails because the configuration of Cheshire pointed by the CI does not enable the CLIC but uses a CVA6 configuration which enables RVSCLIC and RVVCLIC (only RVSCLIC is overwritten by Cheshire leading to an invalid config with RVVCLIC enabled when CLIC is disabled).

Possible workarounds that I see are:

1. Re-point the CI job to a version of Cheshire which sets both the RVSCLIC and RVVCLIC parameters (probably the right way to do it). In this case, should I have the CLIC / vCLIC enabled or disabled for this CI check ?
2. Change the defaul config package used by Cheshire to disable vCLIC feature.

Any hints on how to proceed?

[ezelioli]

@niwis currently the cheshire-integration check fails because the configuration of Cheshire pointed by the CI does not enable the CLIC but uses a CVA6 configuration which enables RVSCLIC and RVVCLIC (only RVSCLIC is overwritten by Cheshire leading to an invalid config with RVVCLIC enabled when CLIC is disabled).

Possible workarounds that I see are:

1. Re-point the CI job to a version of Cheshire which sets both the RVSCLIC and RVVCLIC parameters (probably the right way to do it). In this case, should I have the CLIC / vCLIC enabled or disabled for this CI check ?

2. Change the defaul config package used by Cheshire to disable vCLIC feature.

Any hints on how to proceed?

[ezelioli]

@niwis currently the cheshire-integration check fails because the configuration of Cheshire pointed by the CI does not enable the CLIC but uses a CVA6 configuration which enables RVSCLIC and RVVCLIC (only RVSCLIC is overwritten by Cheshire leading to an invalid config with RVVCLIC enabled when CLIC is disabled).

Possible workarounds that I see are:

1. Re-point the CI job to a version of Cheshire which sets both the RVSCLIC and RVVCLIC parameters (probably the right way to do it). In this case, should I have the CLIC / vCLIC enabled or disabled for this CI check ?
2. Change the defaul config package used by Cheshire to disable vCLIC feature.

Any hints on how to proceed?

I followed approach 1. and the cheshire-integration for this branch (push) is now passing (the CLIC/vCLIC parameters are disabled on the cva6/vclic branch of Cheshire).
The pull_request cheshire-integration job seems to fail due to Bender not finding the CVA6 hash. Do you have any hints on what I could have misconfigured here?

[niwis]

Thank you @ezelioli! I agree with approach 1. Let's stick to the same approach as for the CLIC: disable it per default and only enable it for the tests that actually need it (which is currently none?)

Don't worry about the failing pull_request run; it's not your fault. This was fixed in pulp-platform/bender#181 and should be part of the next bender release.

One more point: my understanding is that upstream Cheshire tries to only reference released versions of dependencies. Should vCLIC be integrated into CLIC before updating Cheshire? (to be confirmed with the Cheshire/CLIC maintainers)

I'll do a full review pass asap.

[ezelioli]

Thank you @ezelioli! I agree with approach 1. Let's stick to the same approach as for the CLIC: disable it per default and only enable it for the tests that actually need it (which is currently none?)

Don't worry about the failing pull_request run; it's not your fault. This was fixed in pulp-platform/bender#181 and should be part of the next bender release.

One more point: my understanding is that upstream Cheshire tries to only reference released versions of dependencies. Should vCLIC be integrated into CLIC before updating Cheshire? (to be confirmed with the Cheshire/CLIC maintainers)

I'll do a full review pass asap.

Thanks for the updates!
I will double check with the maintainers about the versioning in the meantime.

[ezelioli]

Thank you @ezelioli! I agree with approach 1. Let's stick to the same approach as for the CLIC: disable it per default and only enable it for the tests that actually need it (which is currently none?)

Don't worry about the failing pull_request run; it's not your fault. This was fixed in pulp-platform/bender#181 and should be part of the next bender release.

One more point: my understanding is that upstream Cheshire tries to only reference released versions of dependencies. Should vCLIC be integrated into CLIC before updating Cheshire? (to be confirmed with the Cheshire/CLIC maintainers)

I'll do a full review pass asap.

Yes there is no test currently using vCLIC. I can add a vCLIC test and respective config to Cheshire and update this PR to point there.

[niwis]

Yes there is no test currently using vCLIC. I can add a vCLIC test and respective config to Cheshire and update this PR to point there.

I think that would be useful to avoid regressions with vCLIC.

[ezelioli]

@niwis from my side all changes have been integrated.

• On CLIC side, the branch has been rebased on main and the PR is being reviewed but it is close to merge. Already tested with new baremetal tests and full hypervisor + guest tests in Cheshire on this new version. No changes affecting CVA6 are anyways expected.
• On Cheshire side, I added two more tests which validate the delegation of interrupts to S-mode and VS-mode (latter one only when vclic is enabled).
• Cheshire is updated and ready for a cumulative pulp-v2 PR (pulp-v2.0.0-alpha.1 + vclic).

Let me know if further updates are required :)

[niwis]

Thank you @ezelioli! Only a few points:

• I think according to the RISC-V naming convention, the name for this extension would be Shclic or, since this is no standard extension, Xhclic. could you rename the config parameter accordingly (e.g. RVVCLIC -> RVXHCLIC)?
• Some of these changes seem to fix the standard CLIC extension (wfi, trap logic, and the stvt register in csr_regfile). If this is the case, could you split these changes into a separate PR? This will make future maintenance easier e.g. if single extensions shall be upstreamed.
• Same for changes/fixes that are valid for H extension alone (e.g. hgeie CSR)

[ezelioli]

Thank you @ezelioli! Only a few points:

• I think according to the RISC-V naming convention, the name for this extension would be Shclic or, since this is no standard extension, Xhclic. could you rename the config parameter accordingly (e.g. RVVCLIC -> RVXHCLIC)?
• Some of these changes seem to fix the standard CLIC extension (wfi, trap logic, and the stvt register in csr_regfile). If this is the case, could you split these changes into a separate PR? This will make future maintenance easier e.g. if single extensions shall be upstreamed.
• Same for changes/fixes that are valid for H extension alone (e.g. hgeie CSR)

Opened PR #64 for CLIC-only related fixes

[niwis]

@ezelioli could you please do a last rebase of this branch on pulp-v2 and the related Cheshire integration branch on nw/cva6-pulp-v2? I'll merge this next, then.