Hi Conor,
On Mon, Jul 01, 2024 at 05:31:01PM +0100, Conor Dooley wrote:
> [EXTERNAL MAIL]
> Date: Mon, 1 Jul 2024 17:31:01 +0100
> From: Conor Dooley <conor@xxxxxxxxxx>
> To: Samuel Holland <samuel.holland@xxxxxxxxxx>
> Cc: Charlie Jenkins <charlie@xxxxxxxxxxxx>,
> linux-riscv@xxxxxxxxxxxxxxxxxxx, devicetree@xxxxxxxxxxxxxxx,
> linux-kernel@xxxxxxxxxxxxxxx, linux-sunxi@xxxxxxxxxxxxxxx,
> linux-doc@xxxxxxxxxxxxxxx, linux-kselftest@xxxxxxxxxxxxxxx, Conor Dooley
> <conor.dooley@xxxxxxxxxxxxx>, Rob Herring <robh@xxxxxxxxxx>, Krzysztof
> Kozlowski <krzk+dt@xxxxxxxxxx>, Paul Walmsley <paul.walmsley@xxxxxxxxxx>,
> Palmer Dabbelt <palmer@xxxxxxxxxxx>, Albert Ou <aou@xxxxxxxxxxxxxxxxx>,
> Jisheng Zhang <jszhang@xxxxxxxxxx>, Chen-Yu Tsai <wens@xxxxxxxx>, Jernej
> Skrabec <jernej.skrabec@xxxxxxxxx>, Samuel Holland <samuel@xxxxxxxxxxxx>,
> Jonathan Corbet <corbet@xxxxxxx>, Shuah Khan <shuah@xxxxxxxxxx>, Guo Ren
> <guoren@xxxxxxxxxx>, Evan Green <evan@xxxxxxxxxxxx>, Andy Chiu
> <andy.chiu@xxxxxxxxxx>, Jessica Clarke <jrtc27@xxxxxxxxxx>,
> peterlin@xxxxxxxxxxxxx
> Subject: Re: [PATCH v3 03/13] riscv: dts: allwinner: Add xtheadvector to
> the D1/D1s devicetree
>
> On Mon, Jul 01, 2024 at 11:11:55AM -0500, Samuel Holland wrote:
> > Hi Conor, Charlie,
> >
> > On 2024-07-01 11:07 AM, Conor Dooley wrote:
> > > On Mon, Jul 01, 2024 at 10:27:01AM -0500, Samuel Holland wrote:
> > >> On 2024-06-19 6:57 PM, Charlie Jenkins wrote:
> > >>> The D1/D1s SoCs support xtheadvector so it can be included in the
> > >>> devicetree. Also include vlenb for the cpu.
> > >>>
> > >>> Signed-off-by: Charlie Jenkins <charlie@xxxxxxxxxxxx>
> > >>> Reviewed-by: Conor Dooley <conor.dooley@xxxxxxxxxxxxx>
> > >>> ---
> > >>> arch/riscv/boot/dts/allwinner/sun20i-d1s.dtsi | 3 ++-
> > >>
> > >> The other C906/C910/C920-based SoCs need devicetree updates as well, although
> > >> they don't necessarily need to be part of this series:
> > >>
> > >> - sophgo/cv18xx.dtsi
> > >> - sophgo/sg2042-cpus.dtsi
> > >> - thead/th1520.dtsi
> > >
> > > Yeah, I think I pointed that out before with the same "escape hatch" of
> > > it not needing to be in the same series.
> > >
> > >>
> > >>> 1 file changed, 2 insertions(+), 1 deletion(-)
> > >>>
> > >>> diff --git a/arch/riscv/boot/dts/allwinner/sun20i-d1s.dtsi b/arch/riscv/boot/dts/allwinner/sun20i-d1s.dtsi
> > >>> index 64c3c2e6cbe0..6367112e614a 100644
> > >>> --- a/arch/riscv/boot/dts/allwinner/sun20i-d1s.dtsi
> > >>> +++ b/arch/riscv/boot/dts/allwinner/sun20i-d1s.dtsi
> > >>> @@ -27,7 +27,8 @@ cpu0: cpu@0 {
> > >>> riscv,isa = "rv64imafdc";
> > >>
> > >> The ISA string should be updated to keep it in sync with riscv,isa-extensions.
> > >
> > > This probably looks like this cos I said that the kernel shouldn't parse
> > > vendor extensions from "riscv,isa". My rationale was that we have
> > > basically no control of what a vendor extension means in riscv,isa so
> > > we shouldn't parse them from it (so marginally worse than standard
> > > extensions, where it means what the spec says except when it doesn't).
> > >
> > > Given how we implement the parsing, it also meant we weren't implying
> > > meanings for vendor extensions ACPI-land, where we also can't ensure the
> > > meanings or that they remain stable. That change is in a different
> > > series:
> > > https://patchwork.kernel.org/project/linux-riscv/patch/20240609-support_vendor_extensions-v2-1-9a43f1fdcbb9@xxxxxxxxxxxx/
> > >
> > > Although now that I think about it, this might break xandespmu... I
> > > dunno if the Andes guys switched over to using the new property outside
> > > of the single dts in the kernel tree using their SoC. We could
> > > potentially special-case that extension if they haven't - but my
> > > position on this mostly is that if you want to use vendor extensions you
> > > should not be using riscv,isa (even if the regex doesn't complain if you
> > > add them). I'd like to leave the code in the other patch as-is if we can
> > > help it.
> > >
> > > I added Yu Chien Peter Lin here, maybe they can let us know what they're
> > > doing.
> >
> > OK, that makes sense to me. Then please ignore my original comment.
>
> Should the xandespmu thing be an issue, I'd suggest we just do something
> like the following, in place of the new switch arm added by Charlie:
>
> diff --git a/arch/riscv/kernel/cpufeature.c b/arch/riscv/kernel/cpufeature.c
> index ec4bff7a827c..bb99b4055ec2 100644
> --- a/arch/riscv/kernel/cpufeature.c
> +++ b/arch/riscv/kernel/cpufeature.c
> @@ -628,6 +628,17 @@ static void __init riscv_parse_isa_string(const char *isa, unsigned long *bitmap
> if (unlikely(ext_err))
> continue;
>
> + if (*ext == 'x' && acpi_disabled) {
> + /*
> + * xandespmu predates this "rule", so special case it for
> + * hysterical raisins
> + */
> + if (strncasecmp(ext, "xandespmu", ext_end - ext)) {
> + pr_warn_once("Vendor extensions are ignored in riscv,isa. Use riscv,isa-extensions instead.");
> + break;
> + }
> + }
> +
> match_isa_ext(ext, ext_end, bitmap);
> }
> }
>
Thanks for the hands-up!
We don't use the deprecated riscv,isa to specify xandespmu, so no
need to address this special case.
Regards,
Peter Lin
