Quoting Jorge Ramirez (2019-09-10 02:40:34) > On 9/10/19 11:34, Jorge Ramirez wrote: > > On 9/10/19 11:14, Stephen Boyd wrote: > >> > >> This is not a backwards compatible change. > >> > >>>>> --- a/arch/arm64/boot/dts/qcom/msm8916.dtsi > >>>>> +++ b/arch/arm64/boot/dts/qcom/msm8916.dtsi > >>>>> @@ -429,7 +429,8 @@ > >>>>> compatible = "qcom,msm8916-apcs-kpss-global", "syscon"; > >>>>> reg = <0xb011000 0x1000>; > >>>>> #mbox-cells = <1>; > >>>>> - clocks = <&a53pll>; > >>>>> + clocks = <&gcc GPLL0_VOTE>, <&a53pll>; > >>>>> + clock-names = "aux", "pll"; > >>>>> #clock-cells = <0>; > >>>>> }; > >>>>> > >> > >> Because the "clocks" property changed from > >> > >> <&a53pll> > >> > >> to > >> > >> <&gcc GPLL0_VOTE>, <&a53pll> > >> > >> and that moves pll to cell 1 instead of cell 0. > >> > >> > > > > what do you mean by backwards compatible? because this change does not > > break previous clients. > > as per the comments I added to the code (in case this helps framing the > discussion) > > [..] > legacy bindings only defined the pll parent clock (index = 0) with no > name; when both of the parents are specified in the bindings, the > pll is the second one (index = 1). The 'clock-names' property is entirely irrelevant to this discussion. The PLL _must_ be index 0 forever so that the binding is left in a backwards compatible state. Moving the PLL to index 1 and then using clock-names to find it is a backwards incompatible change. The order of clks in the 'clocks' property is an ABI.
