Hi,
On Tuesday 24 January 2017 07:35 PM, Vivek Gautam wrote:
> On Tue, Jan 24, 2017 at 3:03 PM, Kishon Vijay Abraham I <kishon@xxxxxx> wrote:
>> Hi,
>>
>> On Friday 20 January 2017 03:12 AM, Stephen Boyd wrote:
>>> On 01/19, Vivek Gautam wrote:
>>>>
>>>> On 01/19/2017 06:10 AM, Stephen Boyd wrote:
>>>>>>
>>>>> Didn't we already move away from subnodes for lanes in an earlier
>>>>> revision of these patches? I seem to recall we did that because
>>>>> lanes are not devices and the whole "phy as a bus" concept not
>>>>> making sense.
>>>>
>>>> Yea, we started out without having any sub-nodes and we
>>>> argued that we don't require them since the qmp device is
>>>> represented by the qmp node itself.
>>>> The lanes otoh are representative of gen_phys and related properties.
>>>>
>>>> In the driver -
>>>> "struct qmp_phy " represents the lanes and holds "struct phy",
>>>> "struct qcom_qmp" represents the qmp block as a whole and holds
>>>> "struct device"
>>>> Does this make lanes qualify to be childs of qmp ?
>>>
>>> Hmm... maybe I was recalling the DSI phy binding. I think there
>>> are lanes there too but we decided to just have one node.
>>>
>>>>
>>>> "phy as a bus" (just trying to understand here) -
>>>> let's say a usb phy controller has one HSIC phy port and one USB2 phy port.
>>>> So, should this phy controller be a bus providing two ports (and so
>>>> we will have
>>>> couple of child nodes to the phy controller) ?
>>>>
>>>
>>> Typically in DT a subnode or collection of subnodes means there's
>>> some sort of bus involved. Usually each node corresponds to a
>>> struct device, and the parent node corresponds to the bus or
>>> controller for the logical bus.
>>>
>>> In this case (only PCIe though? not UFS or USB?) it seems like we
>>> have multiple phys that share a common register space, but
>>> otherwise they have their own register space and power
>>> management. Would you have each PCIe controller point to a
>>> different subnode for their associated phy? I'm trying to
>>> understand the benefit of the subnodes if they aren't treated as
>>> struct devices.
>>
>> Yes, instead of having all the controller having a phandle to the same PHY and
>> then using other mechanisms to differentiate between the PHYs, each controller
>> can have a phandle to the exact port that it is connected to.
>>
>> This also gives a better representation of the hardware and can avoid lot of
>> boilerplate code in the driver.
>
> Below is one binding that works for me.
> --------------------
> phy@34000 {
> compatible = "qcom,msm8996-qmp-pcie-phy";
> reg = <0x034000 0x488>;
> #clock-cells = <1>;
> #address-cells = <1>;
> #size-cells = <1>;
> ranges;
>
> clocks = <&gcc GCC_PCIE_PHY_AUX_CLK>,
> <&gcc GCC_PCIE_PHY_CFG_AHB_CLK>,
> <&gcc GCC_PCIE_CLKREF_CLK>;
> clock-names = "aux", "cfg_ahb", "ref";
>
> vdda-phy-supply = <&pm8994_l28>;
> vdda-pll-supply = <&pm8994_l12>;
>
> resets = <&gcc GCC_PCIE_PHY_BCR>,
> <&gcc GCC_PCIE_PHY_COM_BCR>,
> <&gcc GCC_PCIE_PHY_COM_NOCSR_BCR>;
> reset-names = "phy", "common", "cfg";
>
> pciephy_p0: port@0 {
> reg = <0x035000 0x130>,
> <0x035200 0x200>,
> <0x035400 0x1dc>;
> #phy-cells = <0>;
>
> clocks = <&gcc GCC_PCIE_0_PIPE_CLK>;
> clock-names = "pipe0";
> resets = <&gcc GCC_PCIE_0_PHY_BCR>;
> reset-names = "lane0";
> };
>
> pciephy_p1: port@1 {
> reg = <0x036000 0x130>,
> <0x036200 0x200>,
> <0x036400 0x1dc>;
> #phy-cells = <0>;
>
> clocks = <&gcc GCC_PCIE_1_PIPE_CLK>;
> clock-names = "pipe1";
> resets = <&gcc GCC_PCIE_1_PHY_BCR>;
> reset-names = "lane1";
> };
>
> pciephy_p2: port@2 {
> reg = <0x037000 0x130>,
> <0x037200 0x200>,
> <0x037400 0x1dc>;
> #phy-cells = <0>;
>
> clocks = <&gcc GCC_PCIE_2_PIPE_CLK>;
> clock-names = "pipe2";
> resets = <&gcc GCC_PCIE_2_PHY_BCR>;
> reset-names = "lane2";
> };
> };
> --------------------
>
> let me know if this looks okay.
looks good to me.
Thanks
Kishon
--
To unsubscribe from this list: send the line "unsubscribe devicetree" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
