Re: [PATCH 2/4] phy: qcom-qusb2: Fix SM6115 init sequence

[Konrad Dybcio <konrad.dybcio@oss.qualcomm.com>]

On 6/17/26 2:48 PM, Iskren Chernev wrote:
> 
> 
> On 6/15/26 1:44 PM, Konrad Dybcio wrote:
>> On 6/14/26 2:29 PM, Iskren Chernev wrote:
>>>
>>>
>>> On 6/10/26 3:04 PM, Konrad Dybcio wrote:
>>>> From: Konrad Dybcio <konrad.dybcio@oss.qualcomm.com>
>>>>
>>>> I don't know where the existing one came from, but it's apparently
>>>> wrong, according to both docs and a downstream DT [1]. Fix it up.
>>>
>>> They came from DTB extracted from a running billie2 (OnePlus Nord N100):
>>> [1] https://mainlining.dev/wp-content/uploads/2021/02/03_dtbdump_Qualcomm_Technologies_Inc._Bengal_SoC.dts
>>>
>>> The phone was bough early after launch, so it could have been wrong/updated later.
>>
>> Good to see you're still around!
>>
>> Looks like vendor tuning. I see that even the initial commit for
>> 6115 had the init sequence I posted. And the OnePlus sources have
>> what seems like a project-specific local copy of the DTSI:
>>
>> https://github.com/OnePlusOSS/android_kernel_oneplus_sm4250/blob/oneplus/SM4250_Q_10.0/arch/arm64/boot/dts/vendor/qcom/bengal-usb.dtsi#L145
>> https://github.com/OnePlusOSS/android_kernel_oneplus_sm4250/blob/oneplus/SM4250_Q_10.0/arch/arm64/boot/dts/vendor/20882/bengal-usb.dtsi#L148
>>
>> To support that, we should add a new property to override the TUNEx
>> registers - like e.g. qcom,hstx-trim-value that's already consumed
> 
> My 2 cents - I never understood why init sequences are taboo in mainline
> and widely used in downstream. I guess if it doesn't change (but across
> what and who decides) it should be in code, but if it's "tuning"
> - whatever that means, possibly depends on other components around, it
> should be "configurable" via DT.

The PHY has some electrical characteristics of its own, and then atop
that are the characteristics of what's on the other end of it. Making
all parameters configurable (i.e. raw init sequence) leads to duplication
and pure blob seqeuences, whereas making everything constant leads to
polluting the driver (if every device-specific seq was to be in C files)

I think the current model of "override as necessary" is OK, especially
since we can use the upstream leverage to require describing what the
altered parameters actually change

>> Would you like to look into that, or should I take this?
> 
> You can take it, the other option is to mark a TODO, and if somebody
> feels strongly about the binary value in a usb tune register s/he can
> take up the task.

Seems like OnePlus does.. actually, a number of vendors do. Sony
does/used to do some tuning there too.

> I just wanted to point out that the number didn't come from a random
> number generator (or AI).

I'm sorry if my language was too harsh. You used the best sources
you had and had no reason to believe they were not the expected values.

Konrad