On 30/10/2019 21:59, Benoit Parrot wrote:
Tony Lindgren <tony@xxxxxxxxxxx> wrote on Wed [2019-Oct-23 08:56:57 -0700]:
* Tony Lindgren <tony@xxxxxxxxxxx> [191022 16:56]:
* Tero Kristo <t-kristo@xxxxxx> [191022 16:48]:
On 22/10/2019 19:21, Benoit Parrot wrote:
Tony Lindgren <tony@xxxxxxxxxxx> wrote on Tue [2019-Oct-22 08:48:16 -0700]:
* Benoit Parrot <bparrot@xxxxxx> [191016 18:47]:
--- a/arch/arm/boot/dts/am43xx-clocks.dtsi
+++ b/arch/arm/boot/dts/am43xx-clocks.dtsi
@@ -704,6 +704,60 @@
ti,bit-shift = <8>;
reg = <0x2a48>;
};
+
+ clkout1_osc_div_ck: clkout1_osc_div_ck {
+ #clock-cells = <0>;
+ compatible = "ti,divider-clock";
+ clocks = <&sys_clkin_ck>;
+ ti,bit-shift = <20>;
+ ti,max-div = <4>;
+ reg = <0x4100>;
+ };
Here too please describe why the clock names are not generic.
Tero originally had this patch in the kernel so this is somewhat of a
revert. Since these "clock" were removed. If the name syntax is no longer
valid for some reason, then I will need a little more informations to
proceed.
Tero, can you assist here?
This one is just following the naming convention of the rest of the clocks
atm.
If we need to fix all the underscore name clocks, that requires pretty much
complete revamp of both the dts data + clock data under the clock driver,
and it is not backwards compatible either. How should we tackle that one?
We could maybe add support code in kernel to do s/-/_/g for the "new" clocks
so that their parent-child relationships would be retained, and then convert
the clocks in phases.
Well some of them can be fixed by configuring things based
on the compatible value and then the node name can be just
clock like it should be.
Here too one option would be to add custom compatibles like:
compatible = "ti,clkout1-osc-div", "ti,divider-clock";
And then have match data configure the rest.
The other option would be to have lookup tables in the clock
driver based on the SoC and reg address.
This is a hidden mine though.. We've hit it already several times,
and any dts clean-up effort has a chance of breaking things.
Hmm maybe in this case just doing this is enough:
clkout1_osc_div_ck: clock@4100 {
...
}
But then we would end up with 6 clock node with the same name "clock@4100",
doesn't pose a problem somewhere?
Yeah, clk core would not know which one to use then.
Tero?
Benoit
Or do all the TI clocks we have have a dependency to the
node naming?
This is a feature of clock core. Clock parents need to have distinct
names, otherwise it won't work.
-Tero
--
Texas Instruments Finland Oy, Porkkalankatu 22, 00180 Helsinki. Y-tunnus/Business ID: 0615521-4. Kotipaikka/Domicile: Helsinki
