Re: [PATCH] clk: qcom: gcc-ipq5018: fix swapped MDIO clocks and resets

From: Jie Luo

Date: Mon Nov 16 2026 - 22:44:41 EST




On 11/16/2026 7:41 PM, George Moussalem wrote:
> On 11/12/26 19:57, Davíð Steinn Geirsson wrote:
>> The GCC has a block reset (BCR) and an AHB clock branch for each of the
>> two MDIO controllers. The pair at 0x58000/0x58004 belongs to the
>> controller at 0x90000 (mdio1 in ipq5018.dtsi) and the pair at
>> 0x58010/0x58014 belongs to the controller at 0x88000 (mdio0). The driver
>> maps the GCC_MDIO0_* IDs to the first pair and the GCC_MDIO1_* IDs to
>> the second, so each MDIO node gets the clock and reset of the other
>> controller.
>>
>> ipq5018.dtsi enables mdio0 and leaves mdio1 disabled. On a board that
>> does not enable mdio1, clk_disable_unused() gates 0x58014 and the
>> controller at 0x88000 stops responding. Its registers then read back a
>> stale bus value, so MDIO reads return garbage instead of failing.
>>
>> Observed on a Ubiquiti UniFi 6 Pro, which has the internal GE PHY on
>> mdio0 and nothing on mdio1:
>> - Gating 0x58004 leaves the 0x88000 controller working and every
>> 0x90000 register reads 0xffff. Gating 0x58014 does the reverse.
>> - Asserting the BCR at 0x58000 resets the 0x90000 controller and halts
>> the 0x58004 branch. The BCR at 0x58010 resets the 0x88000 controller
>> and halts the 0x58014 branch.
>>
>> Swap the clock and reset offsets so that each ID controls the
>> controller it is named after. No in-tree DT uses the MDIO resets.
>
> confirmed working on Linksys MX2000 and MR5500. Both devices don't have
> the internal GE PHY on mdio0 wired up. With mdio1 enabled and mdio0
> disabled, ethernet comes up.
>
> Adding Luo Jie: I think this patch can be reverted because not all
> boards use the internal GE PHY and its LDO needs not be enabled for
> ethernet functionality on externally connected PHYs or switches
> controlled by mdio1
> -
> https://lore.kernel.org/all/20260902-qcom_ipq5018_mdio_pinctrl-v3-2-629c7bb05c44@xxxxxxxxxxxxxxxx/
>
> Tested-by: George Moussalem <george.moussalem@xxxxxxxxxxx>
>

Thanks for testing on the MX2000 and MR5500.

All of the Qualcomm RDP boards have the internal GE PHY, so it makes
sense to keep it enabled by default. A board that doesn't use the
internal GE PHY can disable it in its own board DTS, as you already
did for the MX2000 and MR5500:

&mdio0 {
status = "disabled";
};

>>
>> Fixes: e3fdbef1bab8 ("clk: qcom: Add Global Clock controller (GCC) driver for IPQ5018")
>> Signed-off-by: Davíð Steinn Geirsson <david@xxxxxx>
>> ---
>>
>> Tested and confirmed working on the Ubiquiti U6 Pro with both
>> OpenWRT's v6.18.60 kernel and `qcom/clk-for-7.5`.
>>
>> A comment in OpenWRT's `ipq5018-archer-ax55-v1.dts` indicates this
>> bug has bitten people before:
>> https://github.com/openwrt/openwrt/blob/main/target/linux/qualcommax/dts/ipq5018-archer-ax55-v1.dts#L273-L279
>> That workaround should become unnecessary with this patch, though I
>> do not have the right hardware to confirm that.
>>
>> diff --git a/drivers/clk/qcom/gcc-ipq5018.c b/drivers/clk/qcom/gcc-ipq5018.c
>> index 37ccdc44a8..bf1a4745d1 100644
>> --- a/drivers/clk/qcom/gcc-ipq5018.c
>> +++ b/drivers/clk/qcom/gcc-ipq5018.c
>> @@ -1969,28 +1969,28 @@
>> };
>>
>> static struct clk_branch gcc_mdio0_ahb_clk = {
>> + .halt_reg = 0x58014,
>> + .clkr = {
>> + .enable_reg = 0x58014,
>> + .enable_mask = BIT(0),
>> + .hw.init = &(struct clk_init_data) {
>> + .name = "gcc_mdioi0_ahb_clk",
>> + .parent_hws = (const struct clk_hw *[]) {
>> + &pcnoc_clk_src.hw,
>> + },
>> + .num_parents = 1,
>> + .flags = CLK_SET_RATE_PARENT,
>> + .ops = &clk_branch2_ops,
>> + },
>> + },
>> +};
>> +
>> +static struct clk_branch gcc_mdio1_ahb_clk = {
>> .halt_reg = 0x58004,
>> .clkr = {
>> .enable_reg = 0x58004,
>> .enable_mask = BIT(0),
>> .hw.init = &(struct clk_init_data) {
>> - .name = "gcc_mdioi0_ahb_clk",
>> - .parent_hws = (const struct clk_hw *[]) {
>> - &pcnoc_clk_src.hw,
>> - },
>> - .num_parents = 1,
>> - .flags = CLK_SET_RATE_PARENT,
>> - .ops = &clk_branch2_ops,
>> - },
>> - },
>> -};
>> -
>> -static struct clk_branch gcc_mdio1_ahb_clk = {
>> - .halt_reg = 0x58014,
>> - .clkr = {
>> - .enable_reg = 0x58014,
>> - .enable_mask = BIT(0),
>> - .hw.init = &(struct clk_init_data) {
>> .name = "gcc_mdio1_ahb_clk",
>> .parent_hws = (const struct clk_hw *[]) {
>> &pcnoc_clk_src.hw,
>> @@ -3585,8 +3585,8 @@
>> [GCC_GMAC1_SYS_ARES] = { 0x68438, 1 },
>> [GCC_IMEM_BCR] = { 0x0e000, 0 },
>> [GCC_LPASS_BCR] = { 0x2e000, 0 },
>> - [GCC_MDIO0_BCR] = { 0x58000, 0 },
>> - [GCC_MDIO1_BCR] = { 0x58010, 0 },
>> + [GCC_MDIO0_BCR] = { 0x58010, 0 },
>> + [GCC_MDIO1_BCR] = { 0x58000, 0 },
>> [GCC_MPM_BCR] = { 0x2c000, 0 },
>> [GCC_PCIE0_BCR] = { 0x75004, 0 },
>> [GCC_PCIE0_LINK_DOWN_BCR] = { 0x750a8, 0 },

The GCC driver is correct here. GCC_MDIO0_AHB_CLK (0x58004) and
GCC_MDIO1_AHB_CLK (0x58014), along with the matching BCRs, follow the
IPQ5018 hardware register definitions. What is wrong is the clock
assignment in ipq5018.dtsi: the MDIO controller at 0x88000 is clocked
by GCC_MDIO1_AHB_CLK, and the controller at 0x90000 is clocked by
GCC_MDIO0_AHB_CLK.


Please drop this change and fix the DTS instead, for example:

--- a/arch/arm64/boot/dts/qcom/ipq5018.dtsi
+++ b/arch/arm64/boot/dts/qcom/ipq5018.dtsi
@@ -220,7 +220,7 @@ mdio0: mdio@88000 {
#address-cells = <1>;
#size-cells = <0>;

- clocks = <&gcc GCC_MDIO0_AHB_CLK>;
+ clocks = <&gcc GCC_MDIO1_AHB_CLK>;
clock-names = "gcc_mdio_ahb_clk";
clock-frequency = <6250000>;

@@ -243,7 +243,7 @@ mdio1: mdio@90000 {
pinctrl-0 = <&mdio_pins>;
pinctrl-names = "default";

- clocks = <&gcc GCC_MDIO1_AHB_CLK>;
+ clocks = <&gcc GCC_MDIO0_AHB_CLK>;
clock-names = "gcc_mdio_ahb_clk";

status = "disabled";