Re: [PATCH -fixes v2 2/4] dt-bindings: riscv: Add ratified privileged ISA versions

[Samuel Holland]

On 2024-02-13 11:42 AM, Stefan O'Rear wrote:
> On Tue, Feb 13, 2024, at 12:03 PM, Conor Dooley wrote:
>> On Mon, Feb 12, 2024 at 07:37:33PM -0800, Samuel Holland wrote:
>>> The baseline for the RISC-V privileged ISA is version 1.10. Using
>>> features from newer versions of the privileged ISA requires the
>>> supported version to be reported by platform firmware, either in the ISA
>>> string (where the binding already accepts version numbers) or in the
>>> riscv,isa-extensions property. So far two newer versions are ratified.
>>>
>>> Signed-off-by: Samuel Holland <samuel.holland@xxxxxxxxxx>
>>> ---
>>>
>>> Changes in v2:
>>>  - New patch for v2
>>>
>>>  .../devicetree/bindings/riscv/extensions.yaml | 20 +++++++++++++++++++
>>>  1 file changed, 20 insertions(+)
>>>
>>> diff --git a/Documentation/devicetree/bindings/riscv/extensions.yaml b/Documentation/devicetree/bindings/riscv/extensions.yaml
>>> index 63d81dc895e5..7faf22df01af 100644
>>> --- a/Documentation/devicetree/bindings/riscv/extensions.yaml
>>> +++ b/Documentation/devicetree/bindings/riscv/extensions.yaml
>>> @@ -121,6 +121,16 @@ properties:
>>>              version of the privileged ISA specification.
>>>  
>>>          # multi-letter extensions, sorted alphanumerically
>>
>>> +        - const: sm1p11
>>
>> Why are we beholden to this "1p11" format of RVI's? We have free choice
>> of characters here, what's stopping us using "machine-v1.11", for
>> example?
> 
> I'd prefer to use names that are at least somewhat recognizable, e.g. in
> the profiles spec, rather than making up something from whole cloth.

Right, exactly. My two immediate reasons for choosing this are:
1) this is the exact name used in the profiles[1][2], and
2) the same list of extensions is used for the riscv,isa-extensions property and
the ISA string, and this is the expected format for the ISA string.

If we want invent something brand new for the DT binding (which I'm not sure we
do), then I would recommend adding a property like riscv,privileged-isa-version,
because that removes the duplication between the Sm and Ss extensions (since
almost all implementations would have both).

On the other hand, there will likely be other extensions in the future that need
version numbers in riscv,isa-extensions. Adding a special case for the
privileged ISA doesn't help with this, whereas deciding on a syntax for version
numbers in the extension name does.

Regards,
Samuel

[1]:
https://github.com/riscv/riscv-profiles/blob/main/profiles.adoc#rva20s64-mandatory-extensions
[2]:
https://github.com/riscv/riscv-profiles/blob/main/profiles.adoc#rva22s64-mandatory-extensions