Re: [PATCH] pmdomain: core: switch to dynamic root device

[Geert Uytterhoeven]

Hi Johan,

On Fri, 24 Apr 2026 at 12:41, Johan Hovold <johan@xxxxxxxxxx> wrote:
> Driver core expects devices to be dynamically allocated and will, for
> example, complain loudly if a device that lacks a release function is
> ever freed.
>
> Use root_device_register() to allocate and register the root device
> instead of open coding using a static device.
>
> Signed-off-by: Johan Hovold <johan@xxxxxxxxxx>

Thanks for your patch, which is now commit a96e40f4afdcb52a
("pmdomain: core: switch to dynamic root device") in pmdomain/next.
On e.g. R-Car H1, R-Car M2-W, and R-Car H3, this causes scary messages
when systemd-journald.service is started:

    synth uevent: /always-on: failed to send uevent
    genpd_provider always-on: uevent: failed to send synthetic uevent: -22
    synth uevent: /ca15-cpu0: failed to send uevent
    genpd_provider ca15-cpu0: uevent: failed to send synthetic uevent: -22
    [...]

Reverting the commit fixes the issue.

> --- a/drivers/pmdomain/core.c
> +++ b/drivers/pmdomain/core.c
> @@ -33,9 +33,7 @@ static const struct bus_type genpd_provider_bus_type = {
>  };
>
>  /* The parent for genpd_provider devices. */
> -static struct device genpd_provider_bus = {
> -       .init_name = "genpd_provider",
> -};
> +static struct device *genpd_provider_bus;
>
>  #define GENPD_RETRY_MAX_MS     250             /* Approximate */
>
> @@ -2325,7 +2323,7 @@ static int genpd_alloc_data(struct generic_pm_domain *genpd)
>         device_initialize(&genpd->dev);
>         genpd->dev.release = genpd_provider_release;
>         genpd->dev.bus = &genpd_provider_bus_type;
> -       genpd->dev.parent = &genpd_provider_bus;
> +       genpd->dev.parent = genpd_provider_bus;
>
>         if (!genpd_is_dev_name_fw(genpd)) {
>                 dev_set_name(&genpd->dev, "%s", genpd->name);
> @@ -3567,11 +3565,9 @@ static int __init genpd_bus_init(void)
>  {
>         int ret;
>
> -       ret = device_register(&genpd_provider_bus);
> -       if (ret) {
> -               put_device(&genpd_provider_bus);
> -               return ret;
> -       }
> +       genpd_provider_bus = root_device_register("genpd_provider");
> +       if (IS_ERR(genpd_provider_bus))
> +               return PTR_ERR(genpd_provider_bus);
>
>         ret = bus_register(&genpd_provider_bus_type);
>         if (ret)
> @@ -3593,7 +3589,7 @@ static int __init genpd_bus_init(void)
>  err_prov_bus:
>         bus_unregister(&genpd_provider_bus_type);
>  err_dev:
> -       device_unregister(&genpd_provider_bus);
> +       root_device_unregister(genpd_provider_bus);
>         return ret;
>  }
>  core_initcall(genpd_bus_init);

Apparently this is done too late for drivers/pmdomain/renesas/rcar-sysc.c,
which registers PM domains from an early_initcall().  Doing all work in
rcar-sysc.c from the postcore_initcall() is not an option, as the
CPU power domains are needed for secondary CPU startup on R-Car H1.

Note that R-Car Gen4 is not affected, as it uses rcar-gen4-sysc.c
instead.

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