Conflict Bit[1] definition of misa CSR

[max-starfive]

In privileged spec 20211203, the bit[1] of misa CSR is described as "Tentatively reserved for Bit-Manipulation extension". But in latest spec, bit[1] is a reserved bit. Because our CPU followed the old 20211203 spec, bit[1] has the value of 1 which indicates B extension is implemented inside. But now this is regarded as a test failure. It is unclear which privileged spec the RVI20 test supports.
Will RVI20 test support CPUs which followed old spec?

[allenjbaum]

We follow ratified specs only; the proper way to deal with “tentatively reserved” is to hedge your bet and either make the bit RW( and gave your boot code clear it) or have a hidden fuse or control but that sets it. In this particular case, follow the errata waiver procedure, and I suspect ( but don’t guarantee) the waiver will be approved for this.

…

On Wednesday, May 31, 2023, max-starfive ***@***.***> wrote: In privileged spec 20211203, the bit[1] of misa CSR is described as "Tentatively reserved for Bit-Manipulation extension". But in latest spec, bit[1] is a reserved bit. Because our CPU followed the old 20211203 spec, bit[1] has the value of 1 which indicates B extension is implemented inside. But now this is regarded as a test failure. It is unclear which privileged spec the RVI20 test supports. Will RVI20 test support CPUs which followed old spec? — Reply to this email directly, view it on GitHub <#356>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AHPXVJTGNEQQORNYYDMWHLDXI3YFJANCNFSM6AAAAAAYVBFWVA> . You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>

[max-starfive]

Thanks for your reply.
Since users maybe affected by the changes of spec or test suite (the code under riscv-test-suite directory), could there be a version control to let users to choose the spec version and the correspondent test suite to certificate for?

[seechinliang]

Alternatively, would be considering RVI20 following RVA22 and such?
That mean RVI20 for supporting priviledge spec 1.11 while RVI22 supporting 1.12?

[allenjbaum]

The problem you have is that is there is no ratified spec version that defines MISA[1] to be set, so even if we did that it wouldn't help you or anyone else. In general, there are *no* spec versions at all; there are only ratified specs, and once ratified they are not changed. There may be additional extensions, but those extensions have their own names.

…

On Thu, Jun 1, 2023 at 6:57 AM max-starfive ***@***.***> wrote: Thanks for your reply. Since users maybe affected by the changes of spec or test suite (the code under riscv-test-suite directory), could there be a version control to let users to choose the spec version and the correspondent test suite to certificate for? — Reply to this email directly, view it on GitHub <#356 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AHPXVJRVNJS62ZSUCTXHQPTXJCNTJANCNFSM6AAAAAAYVBFWVA> . You are receiving this because you commented.Message ID: ***@***.***>

[max-starfive]

Understood. Thanks a lot.