Re: [PATCH 4/5] i2c: qcom-cci: Add OPP table support and enforce FAST_PLUS requirements

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

On 9/8/25 12:09 PM, Stephan Gerhold wrote:
> On Mon, Sep 08, 2025 at 12:00:13PM +0200, Konrad Dybcio wrote:
>> On 9/8/25 11:57 AM, Stephan Gerhold wrote:
>>> On Mon, Sep 08, 2025 at 11:49:52AM +0200, Konrad Dybcio wrote:
>>>> On 9/8/25 10:46 AM, Stephan Gerhold wrote:
>>>>> On Mon, Sep 08, 2025 at 10:43:50AM +0200, Konrad Dybcio wrote:
>>>>>> On 9/8/25 10:36 AM, Stephan Gerhold wrote:
>>>>>>> On Thu, Sep 04, 2025 at 04:31:23PM +0200, Konrad Dybcio wrote:
>>>>>>>> From: Konrad Dybcio <konrad.dybcio@oss.qualcomm.com>
>>>>>>>>
>>>>>>>> The CCI clock has voltage requirements, which need to be described
>>>>>>>> through an OPP table.
>>>>>>>>
>>>>>>>> The 1 MHz FAST_PLUS mode requires the CCI core clock runs at 37,5 MHz
>>>>>>>> (which is a value common across all SoCs), since it's not possible to
>>>>>>>> reach the required timings with the default 19.2 MHz rate.
>>>>>>>>
>>>>>>>> Address both issues by introducing an OPP table and using it to vote
>>>>>>>> for the faster rate.
>>>>>>>>
>>>>>>>> Signed-off-by: Konrad Dybcio <konrad.dybcio@oss.qualcomm.com>
>>>>>>>
>>>>>>> Using an OPP table for a single static rate that remains the same over
>>>>>>> the whole lifetime of the driver feels like overkill to me. Couldn't you
>>>>>>> just put the "required-opps" directly into the device node so that it is
>>>>>>> automatically applied when the device goes in/out of runtime suspend?
>>>>>>>
>>>>>>> And since you need to make DT additions anyway, couldn't you just use
>>>>>>> "assigned-clock-rates" to avoid the need for a driver patch entirely? We
>>>>>>> use that for e.g. USB clocks as well.
>>>>>>
>>>>>> This is futureproofing, in case someone invents FastMode++ with a higher
>>>>>> dvfs requirement or for when the driver adds presets for a 19.2 MHz CCI
>>>>>> clock which would (marginally) decrease power consumption
>>>>>>
>>>>>
>>>>> If 19.2 MHz CCI clock is feasible and has lower voltage requirements,
>>>>> then I would expect a separate entry for 19.2 MHz in the OPP table of
>>>>> PATCH 5/5? The DT is unrelated to what functionality you implement in
>>>>> the driver, and that would make the OPP table look less useless. :-)
>>>>
>>>> The frequency plan for 8280 does not recommend any rate != 37.5 MHz
>>>>
>>>> For x1e80100 however, the lovsvs_d1 corner is recommended to be 30
>>>> (yes, thirty) MHz, sourced from CAM_PLL8 for $reasons
>>>>
>>>
>>> The 37.5 MHz rate still exists on X1E I presume, or are you saying we
>>> need more changes to support those odd 30 MHz?
>>
>> Yes, any corner over lowsvs_d1 is 37.5, sourced from cam_pll0
>>
>>> Personally, I'm not fully convinced there is ever going to be a use case
>>> of someone using a "non-standard" frequency. Even if "FastMode++" is
>>> invented most devices will probably want to use it.
>>
>> Not really, there's no reason to make your i2c bus go fastfastfast if
>> the devices on the other end can't cope with it
>>
>>> And the voltage
>>> requirements we're currently talking about here like "low svs" during
>>> camera use cases are kind of negligible compared to others too.
>>
>> Again, this is an I2C controller that seems to be associated with
>> cameras.. No image data has to actually be processed for the
>> communications to take place and you can attach any odd device
>>
> 
> My point is: In the unlikely case that support for faster I2C speeds is
> added in newer SoCs, I think you'd just get a new "standard" base clock
> frequency, add a new cci_data struct with adjusted timings and everyone
> will use that (even for the lower I2C speeds). I doubt anyone will
> bother adjusting and validating this for just one "corner"/voltage level
> less. There are much more effective targets for power optimization than
> the few bytes of I2C communication. :-)

There are OEMs that customize some of the timings (e.g. Sony) and I would
expect SV efforts to at least cover the recommended frequency plan (which
like for x1e sometimes contains >1 frequency)..

That said, I do agree with you it's a rather niche/improbable corner of
the SoC to worry about.. But using required-opps in dt-bindings for
non-trivial hardware (CCI is a little bit more "fun" than the today's
state of the upstream driver makes it seem) simply feels like asking for
trouble (i.e. a ""real need"" for an opp table will probably come around
one day, so I don't think the additional 10 lines of code to support it
are that much of an issue either).

Konrad