Add Smstateen/Ssstateen Extension and CSRs

[neverlandiz]

This PR addresses Issue #547 and adds the Smstateen extension, along with the following Smstateen/Ssstateen CSRs

• mstateen0, mstateen1, mstateen2, mstateen3
• mstateen0h, mstateen1h, mstateen2h, mstateen3h
• hstateen0, hstateen1, hstateen2, hstateen3
• hstateen0h, hstateen1h, hstateen2h, hstateen3h
• sstateen0, sstateen1, sstateen2, sstateen3

[ThinkOpenly]

My comments in hstateen0/hstateen0h generally apply to the others as well.

[dhower-qc]

The comments apply to all the CSRs.

[dhower-qc]

We also need an arch/ext/Ssstateen.yaml file.

[neverlandiz]

I think there is already a Ssstateen.yaml file?

[dhower-qc]

I think there is already a Ssstateen.yaml file?

You're right. Ignore my ignorance please ;)

[jordancarlin]

The {h/s}stateen bits are 0 if the corresponding mstateen bit is 0. It doesn't look like that is captured here right now. I assume this needs a sw_read() call to represent this?

[dhower-qc]

Yep, that's right.

[neverlandiz]

@jordancarlin Thanks for pointing that out! Could you check if this implementation is correct (for hstateen0)?

sw_read(): |
  # for every bit in an mstateen CSR that is zero, the same bit 
  # appears as read-only zero in the matching hstateen CSR

  Bit<64> mstateen0_mask = CSR[mstateen0].csr_value;
  Bit<64> hstateen0_value = csr_value & mstateen0_mask;
  return hstateen0_value;

[jordancarlin]

I think that seems reasonable. For sstateen it will probably be a little more complicated since it changes depending on if the H extension is enabled.

[neverlandiz]

Thanks!

[ThinkOpenly]

Do we need a sw_write() here, too, that enforces read-only for corresponding bits that are zero in mstateen0?

[neverlandiz]

Regression tests are failing due to missing Sscsrind (depends on Issue #557) and Zfinx extensions.

[ThinkOpenly]

Regression tests are failing due to missing Sscsrind

Sscsrind is going in as we speak.

and Zfinx extensions.

I don't even see an issue open for that. :-/

[ThinkOpenly]

There are still a couple of unresolved comments about changing a reset_value and adding a sw_read().

As for Zfinx, I'd be tempted to comment it out in the PR, since Zfinx isn't even started, then open an issue for Zfinx to be added, including a note to uncomment the content here.

[neverlandiz]

Ok, sounds good. Should I open an issue for Zfinx?

I'm still working on resolving the remaining comments, but I'll probably finish them by the end of today.

[jordancarlin]

#711 adds the Zfinx (and other Z*inx) extensions.

[ThinkOpenly]

#711 adds the Zfinx (and other Z*inx) extensions.

Ah, so it does. I searched issues and PRs, but unsurprisingly didn't search for Z.*f.*inx.

Still, that PR and this one depend on each other. I suggest proceeding as planned... comment out Zfinx here, and move this one along.

@neverlandiz , if you would, please add a comment there requesting that your commented out content be uncommented via that PR.

[neverlandiz]

Sounds good, thanks!