On Mon, Apr 25, 2022 at 09:35:13PM +0200, Marek Vasut wrote: > On 4/25/22 21:11, Rob Herring wrote: > > On Fri, Apr 22, 2022 at 06:31:25PM +0200, Marek Vasut wrote: > > > On 4/22/22 17:09, Alexandre Torgue wrote: > > > > In case of "st,stm32mp1-rcc-secure" (stm32mp1 clock driver with RCC > > > > security support hardened), "clocks" and "clock-names" describe oscillators > > > > and are required. > > > > > > > > Signed-off-by: Alexandre Torgue <alexandre.torgue@xxxxxxxxxxx> > > > > > > > > diff --git a/Documentation/devicetree/bindings/clock/st,stm32mp1-rcc.yaml b/Documentation/devicetree/bindings/clock/st,stm32mp1-rcc.yaml > > > > index 7a251264582d..bb0e0b92e907 100644 > > > > --- a/Documentation/devicetree/bindings/clock/st,stm32mp1-rcc.yaml > > > > +++ b/Documentation/devicetree/bindings/clock/st,stm32mp1-rcc.yaml > > > > @@ -58,14 +58,8 @@ properties: > > > > - st,stm32mp1-rcc-secure > > > > - st,stm32mp1-rcc > > > > - const: syscon > > > > - > > > > - clocks: > > > > - description: > > > > - Specifies the external RX clock for ethernet MAC. > > > > - maxItems: 1 > > > > - > > > > - clock-names: > > > > - const: ETH_RX_CLK/ETH_REF_CLK > > > > + clocks: true > > > > + clock-names: true > > > > > > It looks like this should rather be a property than a compatible string -- > > > the compatible string is used by the OS to determine which hardware is > > > represented by a node, but here it is the same hardware in either case, > > > "st,stm32mp1-rcc" and "st,stm32mp1-rcc-secure", it is still the same > > > STM32MP1 RCC block, just configured differently by some bootloader stage. > > > > > > So why not just add one-liner property of the RCC block like ? > > > st,rcc-in-secure-configuration > > > > Because using compatible was already decided. > > I see ... may I ask why compatible is OK in this case even though this is > encoding a policy (secure/non-secure configuration of the same clock IP) > into DT ? I see 'compatible' as an encoding of what is the programming model of the device. Secure vs. non-secure have different models. PCIe hosts vs endpoint mode is a similar case where we mostly have 2 compatibles (but not always). I wouldn't say which way we do things is set in stone, but in this case we already decided something. Rob
