[PATCH v2 2/6] dt-bindings: riscv: Document cboz-block-size

Andrew Jones ajones at ventanamicro.com
Mon Jan 23 07:54:04 PST 2023


On Mon, Jan 23, 2023 at 08:33:56AM -0600, Rob Herring wrote:
> On Sun, Jan 22, 2023 at 1:13 PM Andrew Jones <ajones at ventanamicro.com> wrote:
> >
> > The Zicboz operates on a block-size defined for the cpu-core,
> > which does not necessarily match other cache-sizes used.
> 
> Please use get_maintainers.pl and send patches to the correct lists.

Yup, Conor also pointed out that I forgot to update the CC list when
adding this patch to the series.

> 
> I have no idea what Zicboz is. How does it relate to Zicbom for which
> we already have a block size property? I really hate one by one
> property additions because they lead to poorly designed bindings. So
> what's next? What other information might be needed?

Zicbom and Zicboz are both RISC-V ISA extensions for cache-block operation
(CBO) instructions. Zicbom defines the instructions cbo.inval, cbo.clean,
and cbo.flush while Zicboz only defines cbo.zero. While it's probably
likely that the cache block sizes which Zicbom and Zicboz use will be
the same when both are implemented, the specification does not require it.
With that in mind, it makes sense to me that Zicbom has its own property
and that Zicboz could follow Zicbom's pattern with its own property as
well.

That said, having a generic block size property which is used in the
absence of the per-extension block size properties would allow DTs to only
specify the size once when they're the same. In my reply to Conor, I
suggested introducing a cbo-block-size property for this purpose, but Anup
suggests we just expand the purpose of cbom-block-size. Expanding cbom-
block-size's purpose would allow its size to be used with cbo.zero in the
absence of a cboz-block-size property. Additionally, we could defer the
introduction of the cboz-block-size property until some system needs it,
which may be never.

As far as to what's coming next, I'm not aware of a plan for more of these
types of properties at this time, but the CMO spec also describes prefetch
instructions, which are defined under the Zicbop extension. If Zicbop
support is added, then it should follow the same pattern as we agree for
Zicboz, which is either

 a. Add cboz-block-size and require it (as this series currently does)
 b. Add cboz-block-size, expand the function of cbom-block-size to be
    a fallback, and fallback to cbom-block-size when cboz-block-size is
    absent
 c. Don't add cboz-block-size, only expand the function of cbom-block-size
    and use it. If a need arises for cboz-block-size some day, then it
    can be added at that time.
 d. ??
  
I'm not aware of any additional information needed for these extensions
beyond the block sizes.

Thanks,
drew



More information about the linux-riscv mailing list