Re: Support for trees of clocks served by one clock manager

[Jared McNeill <jmcneill%invisible.ca@localhost>]

Can you share your dts? Searching up the tree for clock controllers is not 
specified in the clock-bindings documentation afaict.

As far as decode is concerned, it was a mistake on my part not to include 
phandle as a separate parameter. I would rather fix it that way, along 
with making sure that the various decode/acquire callbacks in fdtbus are 
consistent here.

On Wed, 15 Aug 2018, Aymeric Vincent wrote:

> The following patch allows to use one clock manager for a tree of
> clocks when #clock-cells == 0, by looking for a clock manager up the
> device tree until one is found or the root is met.
>
> The decode function can know that the intended clock's phandle is given
> in *data instead of an index thanks to the len == 0 parameter. It's not
> cute but that's how the device tree is specified.
>
> Could someone please commit it after reviewing it?
>
> Regards,
> Aymeric
>
> diff --git a/sys/dev/fdt/fdt_clock.c b/sys/dev/fdt/fdt_clock.c
> index f82e2c86b63a..519698293aef 100644
> --- a/sys/dev/fdt/fdt_clock.c
> +++ b/sys/dev/fdt/fdt_clock.c
> @@ -95,16 +95,18 @@ fdtbus_clock_get_index_prop(int phandle, u_int index, const char *prop)
> 		return NULL;
>
> 	for (n = 0, resid = len; resid > 0; n++) {
> -		const int cc_phandle =
> -		    fdtbus_get_phandle_from_native(be32toh(p[0]));
> +		int cc_phandle = fdtbus_get_phandle_from_native(be32toh(p[0]));
> 		if (of_getprop_uint32(cc_phandle, "#clock-cells", &clock_cells))
> 			break;
> 		if (n == index) {
> -			cc = fdtbus_get_clock_controller(cc_phandle);
> +			do {
> +				cc = fdtbus_get_clock_controller(cc_phandle);
> +				cc_phandle = OF_parent(cc_phandle);
> +			} while (cc == NULL && cc_phandle != -1);
> 			if (cc == NULL)
> 				break;
> 			clk = cc->cc_funcs->decode(cc->cc_dev,
> -			    clock_cells > 0 ? &p[1] : NULL, clock_cells * 4);
> +			    clock_cells > 0 ? &p[1] : &p[0], clock_cells * 4);
> 			break;
> 		}
> 		resid -= (clock_cells + 1) * 4;