Quoting Wen He (2019-08-19 19:24:25)
>
> >
> > > > > +
> > > > > +static int plldig_clk_probe(struct platform_device *pdev) {
> > > > > + struct clk_plldig *data;
> > > > > + struct resource *mem;
> > > > > + const char *parent_name;
> > > > > + struct clk_init_data init = {};
> > > > > + struct device *dev = &pdev->dev;
> > > > > + int ret;
> > > > > +
> > > > > + data = devm_kzalloc(dev, sizeof(*data), GFP_KERNEL);
> > > > > + if (!data)
> > > > > + return -ENOMEM;
> > > > > +
> > > > > + mem = platform_get_resource(pdev, IORESOURCE_MEM, 0);
> > > > > + data->regs = devm_ioremap_resource(dev, mem);
> > > > > + if (IS_ERR(data->regs))
> > > > > + return PTR_ERR(data->regs);
> > > > > +
> > > > > + init.name = dev->of_node->name;
> > > > > + init.ops = &plldig_clk_ops;
> > > > > + parent_name = of_clk_get_parent_name(dev->of_node, 0);
> > > > > + init.parent_names = &parent_name;
> > > >
> > > > Can you use the new way of specifying clk parents with the
> > > > parent_data member of clk_init?
> > >
> > > Of course, but I don't understand why need recommend to use this member?
> > > Is that the member parent_names will be discard in future?
> > >
> > > Here are definition of the clk-provider.h
> > > /* Only one of the following three should be assigned */
> > > const char * const *parent_names;
> > > const struct clk_parent_data *parent_data;
> > > const struct clk_hw **parent_hws;
> > >
> > > For PLLDIG, it only has one parent.
> >
> > Yes. Can you use clk_parent_data array and specify a DT index of 0 and some
> > name that would go into "clock-names" in the .fw_name member?
>
> OK, but .fw_name used for to registering clk, current it registered with fixed clk in dts .
> I think should be specify a DT index of 0 and specify the unique name for .name member.
>
> I found have two ways to specify:
> 1. declare clk_parent_data variable parent_data, and initialization with clk_init_data, like this:
>
> parent_data.name = of_clk_get_parent_name(dev->of_node, 0);
This isn't preferred because of_clk_get_parent_name() is DT specific and
relies on the parent clk being registered before calling the function so
that we can figure out the globally unique name.
> parent_data.index = 0;
>
> init.name = dev->of_node->name;
> init.ops = &plldig_clk_ops;
> init.parent_data = &parent_data;
> init.num_parents = 1;
>
> 2. Or use a static const array for here? And put the unique name and index like this.
> static const struct clk_parent_data parent_data[] = {
> {.name = "phy_27m", .index = 0},
> };
>
> After then initialization with macro CLK_HW_INIT_PARENTS_DATA?
Yes use option #2. But, don't even specify a .name if this is new code
because .name is a fallback mechanism that is supposed to be used if
you're migrating code from an older DT to a newer DT. I don't think
that's happening here.
