Message-ID: <CA+ZOyah1Dwf1L9Mn50+v5tLToyktmmASeA5CutW5uwNybDG_Ow@mail.gmail.com>
Date:   Wed, 31 Aug 2022 13:17:48 +0800
From:   Zong Li <zongbox@...il.com>
To:     Conor.Dooley@...rochip.com
Cc:     ben.dooks@...ethink.co.uk, ben.dooks@...ive.com,
        palmer@...belt.com, paul.walmsley@...ive.com,
        aou@...s.berkeley.edu, greentime.hu@...ive.com,
        linux-kernel@...r.kernel.org, linux-riscv@...ts.infradead.org,
        robh+dt@...nel.org, krzysztof.kozlowski+dt@...aro.org,
        devicetree@...r.kernel.org
Subject: Re: [RESEND PATCH] dt-bindings: sifive-ccache: fix cache level for l3 cache

<Conor.Dooley@...rochip.com> 於 2022年8月31日 週三 凌晨1:09寫道：
>
> On 30/08/2022 17:49, Ben Dooks wrote:
> > On 30/08/2022 14:49, Conor.Dooley@...rochip.com wrote:
> >> On 30/08/2022 13:58, Ben Dooks wrote:
> >>> EXTERNAL EMAIL: Do not click links or open attachments unless you know the content is safe
> >>>
> >>> On 30/08/2022 13:56, Conor.Dooley@...rochip.com wrote:
> >>>> On 30/08/2022 13:51, Ben Dooks wrote:
> >>>>> EXTERNAL EMAIL: Do not click links or open attachments unless you know the content is safe
> >>>>>
> >>>>> With newer cores such as the p550, the SiFive composable cache can be
> >>>>> a level 3 cache. Update the cache level to be one of 2 or 3.
> >>>>>
> >>>>> Signed-off-by: Ben Dooks <ben.dooks@...ive.com>
> >>>>> ---
> >>>>>     Documentation/devicetree/bindings/riscv/sifive-ccache.yaml | 2 +-
> >>>>>     1 file changed, 1 insertion(+), 1 deletion(-)
> >>>>>
> >>>>> diff --git a/Documentation/devicetree/bindings/riscv/sifive-ccache.yaml b/Documentation/devicetree/bindings/riscv/sifive-ccache.yaml
> >>>>> index 1a64a5384e36..6190deb65455 100644
> >>>>> --- a/Documentation/devicetree/bindings/riscv/sifive-ccache.yaml
> >>>>> +++ b/Documentation/devicetree/bindings/riscv/sifive-ccache.yaml
> >>>>> @@ -45,7 +45,7 @@ properties:
> >>>>>         const: 64
> >>>>>
> >>>>>       cache-level:
> >>>>> -    const: 2
> >>>>> +    enum: [2, 3]
> >>>>
> >>>> Do we want to enforce the cache level like we currently do for
> >>>> interrupts and cache-sets?
> >>>
> >>> Not sure on that, for the P550 cores the ccache is going to be level3
> >>> and my colleague has said it does confuse some tooling if the level is
> >>> not set correctly.
> >>
> >> What I meant was:
> >> Currently we enforce the correct cache-sets & interrupts based on the
> >> compatible string. Adding enum: [2, 3] relaxes the enforcement of the
> >> cache-level for existing compatibles and does not prevent someone from
> >> setting an incorrect cache level for p550 cores.
> >>
> >> I think that on top of adding the enum, we should add some enforcement
> >> so that the cache is not incorrectly configured for both existing l2
> >> caches and for the new l3 versions.
> >
> > Ok, we can add some enforcement once we add the new bindings, but I'm
> > not ready for that today and I'd rather get the current queue sorted
> > out first before we come in with newer silicon which hasn't hit the
> > market uet.
>
> SGTM. Maybe this particular patch could come at the start of the
> series. And then we could add something so that the end result looks
> like the following (white space damaged) diff:
>
> diff --git a/Documentation/devicetree/bindings/riscv/sifive-l2-cache.yaml b/Documentation/devicetree/bindings/riscv/sifive-l2-cache.yaml
> index ca3b9be58058..994e4b143485 100644
> --- a/Documentation/devicetree/bindings/riscv/sifive-l2-cache.yaml
> +++ b/Documentation/devicetree/bindings/riscv/sifive-l2-cache.yaml
> @@ -33,6 +33,7 @@ properties:
>      oneOf:
>        - items:
>            - enum:
> +              - sifive,ccache0
>                - sifive,fu540-c000-ccache
>                - sifive,fu740-c000-ccache
>            - const: cache
> @@ -45,7 +46,7 @@ properties:
>      const: 64
>
>    cache-level:
> -    const: 2
> +    enum: [2, 3]
>
>    cache-sets:
>      enum: [1024, 2048]
> @@ -115,6 +116,23 @@ allOf:
>          cache-sets:
>            const: 1024
>
> +  - if:
> +      properties:
> +        compatible:
> +          contains:
> +            const:
> +              sifive,ccache0
> +
> +    then:
> +      properties:
> +        cache-level:
> +         enum: [2, 3]
> +
> +    else:
> +      properties:
> +        cache-level:
> +          const: 2
> +
>  additionalProperties: false
>
>  required:
>
>
> That would keep the enforcement for existing caches and allow you
> the freedome to do w/e you want for the ccache0 compatible.

Thanks you all for bring me here,  we actually have some core series
with 4096 cache set in ccache, should we need to extend the cache set
as follow? or we only need to focus on the DTS which is already in
mainline.

cache-sets:
-    enum: [1024, 2048]
+   enum: [1024, 2048, 4096]

>
> >
> >> @Zong, could you please incorporate Ben's patches into your V2? it
> >> would make it a lot easier to review what's going on here. It may
> >> also make sense to add the compatible for the p550 cache while we are
> >> at it...
> >>
> >> FYI, there is also this patch here outstanding against the l2:
> >> https://lore.kernel.org/linux-riscv/20220825180417.1259360-2-mail@conchuod.ie/
> >>
> >> I intend taking this into 6.0-rc5 or so as a fix, so if you could
> >> rebase the series on that so it is not lost in the dt-binding rename
> >> that would be great.
> >
> > Do we need someone to take charge of this series?
> >
>
> Can I volunteer Zong? (since all but two of the patches are theirs)
>

It is ok to me, but I'm still refining the patchset for V2, and I'm
not sure if we will need the V3. Do you prefer to take V2 patch set
first and replace the dt-binding patch?

> Thanks,
> Conor.
> _______________________________________________
> linux-riscv mailing list
> linux-riscv@...ts.infradead.org
> http://lists.infradead.org/mailman/listinfo/linux-riscv

Powered by blists - more mailing lists