Attempts to access non-existent CSRs are reserved

[aswaterman]

Resolves #1116

[allenjbaum]

If there is an attempt to access a non-existent CSR (or write a non-existent CSR in the Read-only region) - is there no guarantee even then that it will get an illegal instruction exception? Or, it is guaranteed only for implemented CSRs at an insufficient priv level? (i.e. which has higher priority, insufficient priv or unimplemented)? I would vote for insufficient priv level here.....

…

On Sun, Jan 28, 2024 at 7:24 PM Andrew Waterman ***@***.***> wrote: Resolves #1116 <#1116> ------------------------------ You can view, comment on, or merge this pull request online at: #1207 Commit Summary - 018a901 <018a901> Attempts to access non-existent CSRs are reserved File Changes (1 file </~https://github.com/riscv/riscv-isa-manual/pull/1207/files>) - *M* src/priv-csrs.adoc </~https://github.com/riscv/riscv-isa-manual/pull/1207/files#diff-9babc2ac9f6753b391e6584ce856baf203f206cb43bc0d3a8a8d5878ed748b7e> (6) Patch Links: - /~https://github.com/riscv/riscv-isa-manual/pull/1207.patch - /~https://github.com/riscv/riscv-isa-manual/pull/1207.diff — Reply to this email directly, view it on GitHub <#1207>, or unsubscribe </~https://github.com/notifications/unsubscribe-auth/AHPXVJWOKSXH7JMY6CIN6WDYQ4I5BAVCNFSM6AAAAABCOW4BAWVHI2DSMVQWIX3LMV43ASLTON2WKOZSGEYDINJWGAZTSMY> . You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>

[aswaterman]

Unimplemented CSRs don’t have a defined privilege level, so they can’t trap on the basis of privilege. (The use of bits 9:8 for privilege is just a convention, as is the use of bits 11:10 for writability. The priv spec could in principle contravene these conventions at any time, even though that would be a deeply unfortunate thing to do.)

[allenjbaum]

Hmm; given the description of the regions in the spec, I thought of it as far more than a convention, but more like a ... spec.

…

On Sun, Jan 28, 2024 at 10:08 PM Andrew Waterman ***@***.***> wrote: Unimplemented CSRs don’t have a defined privilege level, so they can’t trap on the basis of privilege (The use of bits 9:8 for privilege is just a convention, as is the use of bits 11:10 for writability.) — Reply to this email directly, view it on GitHub <#1207 (comment)>, or unsubscribe </~https://github.com/notifications/unsubscribe-auth/AHPXVJUJ4FF56TK6HEWCI7TYQ44HDAVCNFSM6AAAAABCOW4BAWVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMYTSMJUGAZDGNBZGM> . You are receiving this because you commented.Message ID: ***@***.***>

[aswaterman]

The relevant sentence begins with the words "by convention".

[pdonahue-ventana]

Notably, the ratified debug 0.13 spec did not follow that convention. The mscontext CSR in debug 1.0 continues to not follow that convention but it's now deprecated and replaced by something that follows the convention.

[gfavor]

LGTM.