Re: Re: [PATCH 3/3] target/riscv: Enable 'B' extension on max CPU type

[Andrew Jones]

On Wed, Jan 10, 2024 at 03:32:21PM -0300, Daniel Henrique Barboza wrote:
> 
> 
> On 1/9/24 14:07, Rob Bradford wrote:
> > Signed-off-by: Rob Bradford <rbradford@rivosinc.com>
> > ---
> >   target/riscv/tcg/tcg-cpu.c | 3 ++-
> >   1 file changed, 2 insertions(+), 1 deletion(-)
> > 
> > diff --git a/target/riscv/tcg/tcg-cpu.c b/target/riscv/tcg/tcg-cpu.c
> > index f10871d352..9705daec93 100644
> > --- a/target/riscv/tcg/tcg-cpu.c
> > +++ b/target/riscv/tcg/tcg-cpu.c
> > @@ -999,7 +999,8 @@ static void riscv_init_max_cpu_extensions(Object *obj)
> >       const RISCVCPUMultiExtConfig *prop;
> >       /* Enable RVG, RVJ and RVV that are disabled by default */
> > -    riscv_cpu_set_misa(env, env->misa_mxl, env->misa_ext | RVG | RVJ | 
> > RVV);
> > +    riscv_cpu_set_misa(env, env->misa_mxl,
> > +                       env->misa_ext | RVG | RVJ | RVV | RVB);
> 
> I'm aware that we decided a while ago the 'max' CPU could only have 
> non-vendor and
> non-experimental extensions enabled. RVB is experimental, so in theory we 
> shouldn't
> enable it.
> 
> But RVB is an alias for zba, zbb and zbs, extensions that the 'max' CPU is 
> already
> enabling. In this case I think it's sensible to enable RVB here since it 
> would just
> reflect stuff that it's already happening.

It's also setting the B bit in misa, which, until this spec is at least
frozen, is a reserved bit and reserved bits "must return zero when read".

I don't want to stand in the way of progress and it seems 99.9% likely
that the spec will be frozen and ratified, but, if we want to stick to
our policies (which we should document), then even the 'max' cpu type
should require x-b be added to the command line if it wants the B bit
set in misa.

Thanks,
drew