Re: [PATCH] clk: renesas: r9a06g032: Avoid needless probe deferring

[Geert Uytterhoeven]

Hi Phil,

On Fri, Jul 20, 2018 at 2:06 PM Phil Edworthy <phil.edworthy@xxxxxxxxxxx> wrote:
> On 20 July 2018 12:21, Geert Uytterhoeven wrote:
> > On Wed, Jul 18, 2018 at 4:34 PM Phil Edworthy wrote:
> > > To avoid all SoC peripheral drivers deferring their probes, both clock
> > > and pinctrl drivers should already be probed. Since the pinctrl driver
> > > requires a clock to access the registers, the clock driver should be
> > > probed before the pinctrl driver.
> > >
> > > Therefore, move the clock driver from subsys_initcall to core_initcall.
> > >
> > > Signed-off-by: Phil Edworthy <phil.edworthy@xxxxxxxxxxx>
> >
> > Thanks for your patch!
> Thanks for your review!
>
> > The (not yet upstreamed) pinctrl driver uses postcore_initcall(), right?
> No, the pinctrl driver uses subsys_initcall, but postcore_initcall or
> arch_initcall may be better to make it clear about the dependencies.

if the pinctrl driver uses subsys_initcall(), ...

> > > --- a/drivers/clk/renesas/r9a06g032-clocks.c
> > > +++ b/drivers/clk/renesas/r9a06g032-clocks.c
> > > @@ -877,17 +877,18 @@ static const struct of_device_id
> > r9a06g032_match[] = {
> > >         { }
> > >  };
> > >
> > > -static struct platform_driver r9a06g032_clock_driver = {
> > > +static struct platform_driver r9a06g032_clock_driver __refdata = {
> > >         .driver         = {
> > >                 .name   = "renesas,r9a06g032-sysctrl",
> > >                 .of_match_table = r9a06g032_match,
> > >         },
> > > +       .probe = r9a06g032_clocks_probe,
> > >  };
> > >
> > >  static int __init r9a06g032_clocks_init(void)  {
> > > -       return platform_driver_probe(&r9a06g032_clock_driver,
> > > -                       r9a06g032_clocks_probe);
> > > +       platform_driver_register(&r9a06g032_clock_driver);
> > > +       return 0;
> That should be:
> +       return platform_driver_register(&r9a06g032_clock_driver);
>
> > >  }
> >
> > Why are all of the above changes needed?
> > Shouldn't the platform_driver_probe() keep on working?
> > If it does not, it means the clock driver has some other dependency, and
> > cannot be bound immediately.  This is potentially a dangerous situation, as
> > r9a06g032_clocks_probe() is __init, but can still be called at any time later.
> > Hence using platform_driver_probe() is the safe thing to do, possibly with a
> > different reshuffling of the clock and pinctrl initcall priorities.
> No, you cannot call platform_driver_probe() from core_initcall.
> All drivers that are in core_initcall call platform_driver_register().

Hence they cannot have their probe function __init.

>
> Thanks
> Phil
>
> > > -subsys_initcall(r9a06g032_clocks_init);
> > > +core_initcall(r9a06g032_clocks_init);

... using postcore_initcall() or arch_initcall() here, should work with
platform_driver_probe()?

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