Re: [PATCH 3/9] usb: dwc3: qcom: Add support for selecting UTMI as reference clock

[Krishna Kurapati <krishna.kurapati@oss.qualcomm.com>]

On 8/26/2026 3:07 PM, George Moussalem wrote:
> On 8/26/26 13:18, Konrad Dybcio wrote:
>> On 8/25/26 12:42 PM, George Moussalem via B4 Relay wrote:
>>> From: George Moussalem <george.moussalem@outlook.com>
>>>
>>> On many Qualcomm chipsets, the UTMI clock is used as the reference clock
>>> (ex. IPQ5018, IPQ6018, and IPQ9574 and more). Since moving to the new
>>> flattened snps-dwc3 model, it is no longer possible to pass the
>>> reference clock to the dwc3 core driver.
>>>
>>> The clock rate is used by dwc3 core to derive the reference clock period
>>> and frame length adjusment. This is needed when the reference clock
>>> deviates from the hardware default (typically 19.2MHz). Above mentioned
>>> SoCs use different rates and, as such, the core initializes with the
>>> wrong values which leads to non-functional USB operation.
>>>
>>> Read the "qcom,select-utmi-as-ref-clk" property to determine if the UTMI
>>> clock should be used as the reference clock. If so, set it as the
>>> reference clock in the dwc3_probe_data structure.
>>> The dwc3 core driver will then use this clock to derive the clock rate.
>>>
>>> Signed-off-by: George Moussalem <george.moussalem@outlook.com>
>>> ---
>>
>> There's a shorter patch in flight:
>>
>> https://lore.kernel.org/linux-arm-msm/20260825-ctrl-v1-2-609e2291814d@oss.qualcomm.com/
> 
> I'm aware and have discussed that with Varada. My concerns are that
> directly setting qcom->dwc->ref_clk from the glue layer is a layering
> violation. This patch passes it to the DWC3 core via the probe_data struct.
> 
> In addition, if ignore_clocks_and_resets is set in the glue layer (which
> is the case for qcom's and others), setting the dwc->ref_clk makes it
> become managed by the core which is the opposite of what the
> ignore_clocks_and_resets property is there for.
> 
I believe doing it from here is better. In [1], refclk is being added as 
a new clock, but I checked with Varada offline and he said that he 
wanted to pass mock_utmi as ref_clk. IMO This patch is better as it 
reuses mock_utmi instead of adding another new clock.

[1]: 
https://lore.kernel.org/all/20260825-ctrl-v1-2-609e2291814d@oss.qualcomm.com/

Regards,
Krishna,