[libre-riscv-dev] CSR control of architectural hints

[Jacob Lifshay]

Specifically, there would be a csr that is matched against every
instruction, when it matches, a speculation fence is inserted. We need to
allow for the case that the selected hint has other functionality, such as
a csr write, and still execute the hint as a normal instruction.

I had also suggested we would want functionality to match any instruction
(or group of instructions) and convert it into a trap as a way to
trap-and-emulate allowing us to fix hw bugs discovered after we build the
final silicon.

Jacob

On Sun, Feb 3, 2019, 05:35 Luke Kenneth Casson Leighton <lkcl at lkcl.net
wrote:

> Off-list discussion, documenting.
>
> Speculation barrier hints are needed, for in-process protection. However
> RISCV Foundation is not responding on need to allocate an official hint.
>
> If hardware goes ahead and selects the wrong hint, it will be money wasted.
>
> So the idea raised by Jacob is to allow hints to be *reallocated* to
> different purposes, through CSRs.
>
> I raised the potential hazard of the hint requiring arguments (from src
> registers).
>
> L.
>
>
> --
> ---
> crowd-funded eco-conscious hardware: https://www.crowdsupply.com/eoma68
> _______________________________________________
> libre-riscv-dev mailing list
> libre-riscv-dev at lists.libre-riscv.org
> http://lists.libre-riscv.org/mailman/listinfo/libre-riscv-dev
>