On Thu, Sep 10, 2020 at 1:09 PM Geert Uytterhoeven <geert@xxxxxxxxxxxxxx> wrote:
> On Thu, Sep 10, 2020 at 12:34 PM Kieran Bingham
> <kieran.bingham+renesas@xxxxxxxxxxxxxxxx> wrote:
> > On 10/09/2020 10:44, Geert Uytterhoeven wrote:
> > > On Mon, Sep 7, 2020 at 5:55 PM Kieran Bingham
> > > <kieran.bingham+renesas@xxxxxxxxxxxxxxxx> wrote:
> > >> On 07/09/2020 03:59, Kuninori Morimoto wrote:
> > >>> From: Kuninori Morimoto <kuninori.morimoto.gx@xxxxxxxxxxx>
> > >>>
> > >>> This patch adds VSP device nodes for R-Car M3-W+ (r8a77961) SoC.
> > >>> This patch is test on R-Car M3-W+ Salvator-XS board.
> > >>> --- a/arch/arm64/boot/dts/renesas/r8a77961.dtsi
> > >>> +++ b/arch/arm64/boot/dts/renesas/r8a77961.dtsi
> > >>> @@ -2056,6 +2056,61 @@ fcpvd2: fcp@fea37000 {
> > >>> iommus = <&ipmmu_vi0 10>;
> > >>> };
> > >>
> > >> The FCP's added are:
> > >>
> > >> fcpf0: fcp@fe950000 {
> > >> fcpf1: fcp@fe951000 {
> > >> fcpvb0: fcp@fe96f000 {
> > >> fcpvb1: fcp@fe92f000 {
> > >> fcpvi0: fcp@fe9af000 {
> > >> fcpvi1: fcp@fe9bf000 {
> > >> fcpvd0: fcp@fea27000 {
> > >> fcpvd1: fcp@fea2f000 {
> > >> fcpvd2: fcp@fea37000 {
> > >>
> > >> So indeed, the first fcpf0 comes before fe960000.
> > >>
> > >> Do we keep the items grouped by the first occurrence? or sort the nodes
> > >> based on address?
> > >>
> > >> for some reason I thought we were ordering based on address, but I see
> > >> other situations where we group too - so I'm confused (and wishing there
> > >> was an automatic tool to get the sorting correct without fuss).
> > >>
> > >> Is there a set policy?
> > >
> > > For nodes with a unit-address, we usually[*] sort by unit-address, but we keep
> > > similar nodes grouped. Hence I prefer this v1 over v2.
> >
> > I assume then the groups are sorted by the first entry,
> >
> > I.e. hypothetically:
> >
> > fdp@0
> > fcp@1
> > vsp@2
> > fdp@3
> > fcp@4
> > vsp@5
> > cmm@6
> > cmm@7
> >
> > would become
> >
> > fdp@0
> > fdp@3
> > fcp@1
> > fcp@4
> > vsp@2
> > vsp@5
> > cmm@6
> > cmm@7
>
> Exactly. That's how we (Reneas SoCs) have been (trying to) doing it.
> I am aware there are some deviations (e.g. do you keep all 4 possible
> SCIF types together (they're all serial@), or do you group them per
> type? And some nodes (ipmmu) may be sorted alphabetically by label).
>
> > Has anyone already created any scripting/validation to automate the
> > sorting requirements?
>
> Not yet, AFAIK. I've been thinking about doing that several times,
> though ;-)
>
> > > [*] Seems like FCP/VSP are interleaved in r8a77990.dsi, doh.
> > >
> >
> > Personally I prefer that - but my opinion doesn't matter here - so as
> > long as the rules are defined (or even better, automatically
> > enforceable) that's fine.
>
> Indeed.
>
> Perhaps creating rules is something to be handled at a higher level
> (i.e. common for all DTS files)?
>
> Summarizing "our" rules:
> 1. Nodes without unit-address are sorted alphabetically, by node name,
> 2. Nodes with unit-address are sorted numerically, by unit-address,
> a. Nodes of the same type are grouped together, i.e. the whole
> group is sorted w.r.t. to other nodes/groups based on the
> unit-address of the first member of the group.
> 3. Anchors are sorted alphabetically, but anchor name.
>
> Do they make sense?
For comparison, scripts/dtc/dtx_diff uses "dtc -s" for sorting, which
sorts everything (nodes and properties) alphabetically.
While I can agree on sorting all nodes alphabetically, sorting
properties alphabetically goes IMHO a bit too far. E.g. it's established
practice to put "compatible" and "reg" first.
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
