Re: [PATCH v2 5/7] staging: board: Add support for devices with complex dependencies

[Geert Uytterhoeven]

	Hi Ulf,

On Fri, 4 Sep 2015, Ulf Hansson wrote:
> On 3 September 2015 at 15:35, Geert Uytterhoeven <geert@xxxxxxxxxxxxxx> wrote:
> > On Thu, Sep 3, 2015 at 2:53 PM, Ulf Hansson <ulf.hansson@xxxxxxxxxx> wrote:
> >> On 17 June 2015 at 10:38, Geert Uytterhoeven <geert+renesas@xxxxxxxxx> wrote:
> >>> Add support for easy registering of one ore more platform devices that
> >>> may:
> >>>   - need clocks that are described in DT,
> >>>   - be part of a PM Domain.
> >
> >>> diff --git a/drivers/staging/board/board.c b/drivers/staging/board/board.c
> >>> index 8712f566b31196e0..29d456e29f38feac 100644
> >>> --- a/drivers/staging/board/board.c
> >>> +++ b/drivers/staging/board/board.c
> >
> >>> +int __init board_staging_register_device(const struct board_staging_dev *dev)
> >>> +{
> >>> +       struct platform_device *pdev = dev->pdev;
> >>> +       unsigned int i;
> >>> +       int error;
> >>> +
> >>> +       pr_debug("Trying to register device %s\n", pdev->name);
> >>> +       if (board_staging_dt_node_available(pdev->resource,
> >>> +                                           pdev->num_resources)) {
> >>> +               pr_warn("Skipping %s, already in DT\n", pdev->name);
> >>> +               return -EEXIST;
> >>> +       }
> >>> +
> >>> +       board_staging_gic_fixup_resources(pdev->resource, pdev->num_resources);
> >>> +
> >>> +       for (i = 0; i < dev->nclocks; i++)
> >>> +               board_staging_register_clock(&dev->clocks[i]);
> >>> +
> >>> +       error = platform_device_register(pdev);
> >>> +       if (error) {
> >>> +               pr_err("Failed to register device %s (%d)\n", pdev->name,
> >>> +                      error);
> >>> +               return error;
> >>> +       }
> >>> +
> >>> +       if (dev->domain)
> >>> +               __pm_genpd_name_add_device(dev->domain, &pdev->dev, NULL);
> >>
> >> Urgh, this managed to slip through my filters.
> >>
> >> It seems like we almost managed to remove all users of the
> >> "..._name_add..." APIs for genpd. If hasn't been for $subject patch.
> >> :-)
> >>
> >> Now, I realize this is already too late here, but let's try to fix
> >> this before it turns into a bigger issue.
> >>
> >> Geert, do you think it's possible to convert into using the non-named
> >> bases APIs?
> >
> > That will be difficult. This code is meant to use drivers that are not yet
> > DT-aware on DT-based systems. Hence it uses platform devices with named PM
> > domains, while the PM domains are described in DT.
> > I don't think there's another way to look up a PM domain by name, is there?
> 
> As a matter of fact there are, especially for those genpds that has
> been created through DT as in this case. The API to use is
> of_genpd_get_from_provider() to find the struct generic_pm_domain.

Thanks!

> Yes, I do realize that you need to manage the parsing of the domain
> name to make sure it's the one you want, but I would rather keep that
> "hack" in this driver than in the generic API.

OK. It turned out not to be that complex, cfr. the patch below.
For now this supports PM domains with "#power-domain-cells = <0>" only,
but of course it can be extended if the need arises.

I've also tried:

    np = of_find_compatible_node(NULL, NULL, "renesas,sysc-rmobile");
    np = of_find_node_by_name(np, "a4lc");

(the second step is needed because of the domain hierarchy in DT), but that
would require adding the compatible string to struct board_staging_dev,
and doesn't work if you have multiple identical power providers.

I hope you like it?