Re: [PATCH 5/5] dt-bindings: timer: Add CLINT bindings

[Anup Patel]

On Fri, May 29, 2020 at 4:48 AM Rob Herring <robh@xxxxxxxxxx> wrote:
>
> On Tue, May 26, 2020 at 05:32:30PM -0700, Palmer Dabbelt wrote:
> > On Thu, 21 May 2020 23:29:36 PDT (-0700), seanga2@xxxxxxxxx wrote:
> > > On 5/22/20 1:54 AM, Anup Patel wrote:
> > > > On Fri, May 22, 2020 at 1:35 AM Sean Anderson <seanga2@xxxxxxxxx> wrote:
> > > > >
> > > > > On 5/21/20 9:45 AM, Anup Patel wrote:
> > > > > > +Required properties:
> > > > > > +- compatible : "sifive,clint-1.0.0" and a string identifying the actual
> > > > > > +  detailed implementation in case that specific bugs need to be worked around.
> > > > >
> > > > > Should the "riscv,clint0" compatible string be documented here? This
> > > >
> > > > Yes, I forgot to add this compatible string. I will add in v2.
> > > >
> > > > > peripheral is not really specific to sifive, as it is present in most
> > > > > rocket-chip cores.
> > > >
> > > > I agree that CLINT is present in a lot of non-SiFive RISC-V SOCs and
> > > > FPGAs but this IP is only documented as part of SiFive FU540 SOC.
> > > > (Refer, https://static.dev.sifive.com/FU540-C000-v1.0.pdf)
> > > >
> > > > The RISC-V foundation should host the CLINT spec independently
> > > > under https://github.com/riscv and make CLINT spec totally open.
> > > >
> > > > For now, I have documented it just like PLIC DT bindings found at:
> > > > Documentation/devicetree/bindings/interrupt-controller/sifive,plic-1.0.0.txt
> > >
> > > The PLIC seems to have its own RISC-V-sponsored documentation [1] which
> > > was split off from the older privileged specs. By your logic above,
> > > should it be renamed to riscv,plic0.txt (with a corresponding change in
> > > the documented compatible strings)?
> > >
> > > [1] https://github.com/riscv/riscv-plic-spec
> >
> > Let's propose tagging that PLIC spec as v1.0.0 in the platform spec group, but
> > I don't see a reason why that wouldn't be viable.  Assuming that's all OK, we
> > can start calling this a RISC-V PLIC (in addition to a SiFive PLIC, as they'll
> > be compatible).
> >
> > > >
> > > > If RISC-V maintainers agree then I will document it as "RISC-V CLINT".
> > > >
> > > > @Palmer ?? @Paul ??
> >
> > The CLINT is a SiFive spec.  It has open source RTL so it's been implemented in
> > other designs, but it's not a RISC-V spec.  The CLIC, which is a superset of
> > the CLINT, is a RISC-V spec.  IIRC it's not finished yet (it's the fast
> > interrupts task group), but presumably we should have a "riscv,clic-2.0.0" (or
> > whatever it ends up being called) compat string to go along with the
> > specification.
>
> Whatever you all decide on, note that "sifive,<block><num>" is a SiFive
> thing (as it is documented) and <num> corresponds to tag of the IP
> implmentation (at least it is supposed to). So you can't just copy that
> with 'riscv,<block><num>' unless you have the same IP versioning
> and update the documentation.

I agree that "sifive,<block><num>" is a SiFive thingy. Unfortunately,
lot of RISC-V implementations (SiFive and non-SiFive) have DTS
generated from RTL (not part of Linux sources) where most of them
use "riscv,clint0" as compatible string for CLINT.

>
> Using a spec version is fine, but not standalone. You need
> implementation specific compatible too because no one perfectly
> implements any spec and/or there details a spec may not cover.

Sure, a generic compatible string "riscv,clint0" OR "sifive,clint-1.0.0"
along with an implementation specific compatible string sounds
good to me.

Regards,
Anup