On 27/01/2025 14:56, Dmitry Baryshkov wrote: > On Mon, Jan 27, 2025 at 02:21:04PM +0100, Krzysztof Kozlowski wrote: >> DSI phys, from earliest (28 nm) up to newest (3 nm) generation, provide >> two clocks. The respective clock ID is used by drivers and DTS, so it >> should be documented as explicit ABI. >> >> Signed-off-by: Krzysztof Kozlowski <krzysztof.kozlowski@xxxxxxxxxx> >> >> --- >> >> Patch for Display tree, although with Ack from clock. >> --- >> .../devicetree/bindings/display/msm/dsi-phy-common.yaml | 2 ++ >> MAINTAINERS | 1 + >> include/dt-bindings/clock/qcom,dsi-phy-28nm.h | 9 +++++++++ >> 3 files changed, 12 insertions(+) >> create mode 100644 include/dt-bindings/clock/qcom,dsi-phy-28nm.h >> > >> diff --git a/include/dt-bindings/clock/qcom,dsi-phy-28nm.h b/include/dt-bindings/clock/qcom,dsi-phy-28nm.h >> new file mode 100644 >> index 000000000000..ab94d58377a1 >> --- /dev/null >> +++ b/include/dt-bindings/clock/qcom,dsi-phy-28nm.h > > I think this should be dt-bindings/phy/qcom,foo.h Both options - clock or phy - work for me, although with slight preference of keeping foo constants only in foo (so clock -> clock) because then clock maintainer sees it as well. Also because I would judge by type of constants (so again clock constants -> clock directory), not type of device. We have several MFD devices, like PMICs, which have a clock, so bindings should go to mfd? But mfd is not a real device, but Linux subsystem. For many other archs, e.g Mediatek, pretty often even for the same device, the binding headers are split between clock and reset. I know that Qualcomm GCC has it in one file, with exceptions (ipq, qca8k). Also these other archs have bindings file in e.g. soc or arm, but the header in respective subsystem With exceptions of am654 and pisatchio-usb headers, we don't store clock constants in phy. Unless someone insists or there is existing qcom convention, then I rather prefer to keep it in clock. Best regards, Krzysztof
