Re: [PATCH] RISC-V: cpu.c: remove nonexistent ext from cpuinfo

[Palmer Dabbelt]

On Wed, 18 May 2022 01:40:18 PDT (-0700), i@xxxxxxxxxxx wrote:
> On Tue, Apr 26, 2022 at 06:08:01AM +0800, Hongren (Zenithal) Zheng wrote:
> > There are no single-letter B/K/J extentions,
> > as they are never ratified. For P, it is still in
> > progress and not ratified.
> >
> > The ordering constraint of these placerholders is now removed
> > from the spec. By commit ("Delete more nonexistent
> > extensions from the naming constraints") of riscv/riscv-isa-manual
>
> Hi, is there any further update on this patch?

IIRC there was some debate as to whether that change constituted an 
incompatible change to the spec, but I guess it's stuck around for long 
enough that maybe we should count on it being canon now?  I added Kito 
and Nelson, there's a GNU toolchain call tomorrow morning so hopefully 
we can remember to talk about it...

> > Signed-off-by: Hongren (Zenithal) Zheng <i@xxxxxxxxxxx>
> > ---
> >  arch/riscv/kernel/cpu.c | 4 +---
> >  1 file changed, 1 insertion(+), 3 deletions(-)
> >
> > diff --git a/arch/riscv/kernel/cpu.c b/arch/riscv/kernel/cpu.c
> > index ccb617791e56..53a061ab0743 100644
> > --- a/arch/riscv/kernel/cpu.c
> > +++ b/arch/riscv/kernel/cpu.c
> > @@ -113,10 +113,8 @@ static void print_isa_ext(struct seq_file *f)
> >  /*
> >   * These are the only valid base (single letter) ISA extensions as per the spec.
> >   * It also specifies the canonical order in which it appears in the spec.
> > - * Some of the extension may just be a place holder for now (B, K, P, J).
> > - * This should be updated once corresponding extensions are ratified.
> >   */
> > -static const char base_riscv_exts[13] = "imafdqcbkjpvh";
> > +static const char base_riscv_exts[9] = "imafdqcvh";
>
> The base_riscv_exts "imafdqcvh" is exactly the spec now, as
> https://github.com/riscv/riscv-isa-manual/commit/db7a4a0dad0e99d1ec1fc67b582624fc0aeae98e
> (Add single-letter "H" extension to the table)
> has shown

Oddly enough I stumbled upon that one this morning, it's another one of 
these like the HPM stuff: we used to say "there's no letter describing 
the hypervisor behavior, so it's part of the base" (see the commentary 
on the binutils patch), but now that there's a letter I'm assuming we 
should split that out?

Not clear if the RISC-V folks want H to be ignored by software like 
those other recent changes, and if so it's also not clear that's a good 
idea.

Anyway, sorry this is taking a while but I think it's going to be too 
late for this merge window -- kind of silly for such a small patch, but 
it's got interface implications and it's all a bit of a hot topic right 
now.

> >  static void print_isa(struct seq_file *f, const char *isa)
> >  {
> > --
> > 2.35.1