Hi Rob,
On Fri, Jun 8, 2018 at 10:41 PM Rob Herring <robh+dt@xxxxxxxxxx> wrote:
> On Fri, Jun 8, 2018 at 2:47 AM, Geert Uytterhoeven <geert@xxxxxxxxxxxxxx> wrote:
> > On Fri, Jun 8, 2018 at 8:50 AM, Michel Pollet
> > <michel.pollet@xxxxxxxxxxxxxx> wrote:
> >>> > + I add this to the cpu.txt, as a separate patch:
> >>> > # On other systems, the property can be either
> >>> > 32 bits or 64 bits, it is the driver's responsibility
> >>> > to deal with either sizes.
> >>>
> >>> That is definitely not what we want to say. Use of 32-bit should be
> >>> considered out of spec. Yes, we have a few platforms in that category, but
> >>> they already handle that themselves. Would be nice to fix them, but at least
> >>> the STi platforms don't seem too active.
> >>>
> >>> IMO, we should delete whatever text we can here and at most just refer to
> >>> the spec.
> >>
> >> So actually I didn't use 32 bits by plain chance, I read the cpu.txt file which says
> >> that 64 bits systems use 64 bits property, concluded that in my case I ought to
> >> use 32 bits, then grepped around and found other systems using 32 bits, therefore
> >> I went forward and used it..
> >>
> >> Nothing said here that it should be 64 bits everywhere -- So the documentation
> >> needs fixing somehow. Right now it certainly led me wrong.
> >
> > Perhaps we should add to Documentation/devicetree/bindings/ the standard
> > bindings from ePAPR and successors, too?
>
> I hope you mean *reference* here, not duplicate the bindings here. We
> want to move in the other direction and move the common bindings out
> of the kernel and into the spec.
I did mean copy... I usually grep in Documentation/devicetree/bindings/,
and fall back to the spec only rarely, mostly because the spec usually
doesn't cover what I need.
I am aware of the ongoing work on updating the spec. I guess it's a
chicken-and-egg problem...
A list of standardized properties under Documentation/devicetree/bindings/,
referring to the spec may be a good interim solution. So at least it would show
it with git grep.
> The real solution here is validation which I'm working on. I had
> already converted cpus.txt. Here's an example of the results of the
> validation:
>
> arch/arm/boot/dts/stih410-b2120.dt.yaml:1962:7: cpu@0: 'enable-method'
> is a dependency of 'cpu-release-addr'
> arch/arm/boot/dts/stih410-b2120.dt.yaml:1965:26:
> cpu@0:cpu-release-addr: [155254948] is too short
Thanks, nice!
Gr{oetje,eeting}s,
Geert
--
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert@xxxxxxxxxxxxxx
In personal conversations with technical people, I call myself a hacker. But
when I'm talking to journalists I just say "programmer" or something like that.
-- Linus Torvalds
