On Thu 27 Jan 15:16 CST 2022, Stephen Boyd wrote:
> Quoting Bjorn Andersson (2022-01-25 19:01:31)
> > On Tue 25 Jan 15:46 PST 2022, Doug Anderson wrote:
> >
> > > Hi,
> > >
> > > On Tue, Jan 25, 2022 at 2:58 PM Stephen Boyd <swboyd@xxxxxxxxxxxx> wrote:
> > > >
> > > > Quoting Douglas Anderson (2022-01-25 14:44:22)
> > > > > diff --git a/arch/arm64/boot/dts/qcom/sc7280-herobrine-herobrine-r1.dts b/arch/arm64/boot/dts/qcom/sc7280-herobrine-herobrine-r1.dts
> > > > > new file mode 100644
> > > > > index 000000000000..f95273052da0
> > > > > --- /dev/null
> > > > > +++ b/arch/arm64/boot/dts/qcom/sc7280-herobrine-herobrine-r1.dts
> > > > > @@ -0,0 +1,313 @@
> > > > > +// SPDX-License-Identifier: (GPL-2.0+ OR MIT)
> > > > > +/*
> > > > > + * Google Herobrine board device tree source
> > > > > + *
> > > > > + * Copyright 2022 Google LLC.
> > > > > + */
> > > > > +
> > > > > +/dts-v1/;
> > > > > +
> > > > > +#include "sc7280-herobrine.dtsi"
> > > > > +
> > > > > +/ {
> > > > > + model = "Google Herobrine (rev1+)";
> > > > > + compatible = "google,herobrine", "qcom,sc7280";
> > > >
> > > > Can we stop adding "qcom,sc7280" to the board compatible string? It
> > > > looks out of place. It's the compatible for the SoC and should really be
> > > > the compatible for the /soc node.
> > >
> > > I don't have any objections, but I feel like this is the type of thing
> > > I'd like Bjorn to have the final say on. What say you, Bjorn?
> > >
> >
> > One practical case I can think of right away, where this matters is in
> > cpufreq-dt-plat.c where we blocklist qcom,sc7280.
> >
> > I don't know if we rely on this in any other places, but I'm not keen on
> > seeing a bunch of board-specific compatibles sprinkled throughout the
> > implementation - it's annoying enough having to add each platform to
> > these drivers.
>
> Looking at sc7180, grep only shows cpufreq-dt-plat.c
>
Good, then we handle all other platform specifics in drivers using
platform-specific compatibles.
> $ git grep qcom,sc7180\" -- drivers
> drivers/cpufreq/cpufreq-dt-platdev.c: { .compatible = "qcom,sc7180", },
>
> Simplest solution would be to look at / and /soc for a compatible
> string.
>
You mean that / would contain the device's compatible and /soc the soc's
compatible? I'm afraid I don't see how this would help you - you still
need the compatible in the dts, just now in two places.
Either we leave it as is - which follows my interpretation of what the DT
spec says - or we (and the DT maitainers) agree that it shouldn't be
there (because this dtb won't run on any random qcom,sc7180 anyways) at
all.
Regards,
Bjorn
> $ git grep -W 'soc[^:]*{' -- arch/arm*/boot/dts/ | grep compatible |
> grep -v "simple-bus"
>
> doesn't show many hits. The first hit is "ti,omap-infra" which is
> actually inside an soc node, but even then I don't see anything that
> matches the cpufreq-dt-plat.c lists.
>
> ----8<-----
> diff --git a/drivers/cpufreq/cpufreq-dt-platdev.c
> b/drivers/cpufreq/cpufreq-dt-platdev.c
> index ca1d103ec449..32bfe453f8b4 100644
> --- a/drivers/cpufreq/cpufreq-dt-platdev.c
> +++ b/drivers/cpufreq/cpufreq-dt-platdev.c
> @@ -179,25 +179,29 @@ static bool __init cpu0_node_has_opp_v2_prop(void)
> static int __init cpufreq_dt_platdev_init(void)
> {
> struct device_node *np = of_find_node_by_path("/");
> + struct device_node *soc_np = of_find_node_by_path("/soc");
> const struct of_device_id *match;
> const void *data = NULL;
>
> - if (!np)
> + if (!np && !soc_np)
> return -ENODEV;
>
> match = of_match_node(allowlist, np);
> - if (match) {
> + if (match || (match = of_match_node(allowlist, soc_np))) {
> data = match->data;
> goto create_pdev;
> }
>
> - if (cpu0_node_has_opp_v2_prop() && !of_match_node(blocklist, np))
> + if (cpu0_node_has_opp_v2_prop() && !of_match_node(blocklist, np) &&
> + !of_match_node(blocklist, soc_np))
> goto create_pdev;
>
> + of_node_put(soc_np);
> of_node_put(np);
> return -ENODEV;
>
> create_pdev:
> + of_node_put(soc_np);
> of_node_put(np);
> return PTR_ERR_OR_ZERO(platform_device_register_data(NULL, "cpufreq-dt",
> -1, data,
