Re: [PATCH] clk: qcom: gcc: Update the SDCC clock to use shared_floor_ops

[Taniya Das]

On 8/7/2025 10:32 PM, Konrad Dybcio wrote:
> On 8/6/25 11:39 AM, Taniya Das wrote:
>>
>>
>> On 8/6/2025 3:00 PM, Konrad Dybcio wrote:
>>> On 8/6/25 11:27 AM, Taniya Das wrote:
>>>>
>>>>
>>>> On 8/5/2025 10:52 AM, Dmitry Baryshkov wrote:
>>>>> On Mon, Aug 04, 2025 at 11:59:21PM +0530, Taniya Das wrote:
>>>>>> gcc_sdcc2_apps_clk_src: rcg didn't update its configuration" during
>>>>>> boot. This happens due to the floor_ops tries to update the rcg
>>>>>> configuration even if the clock is not enabled.
>>>>>
>>>>> This has been working for other platforms (I see Milos, SAR2130P,
>>>>> SM6375, SC8280XP, SM8550, SM8650 using shared ops, all other platforms
>>>>> seem to use non-shared ops). What's the difference? Should we switch all
>>>>> platforms? Is it related to the hypervisor?
>>>>>
>>>>
>>>> If a set rate is called on a clock before clock enable, the
>>>
>>> Is this something we should just fix up the drivers not to do?
>>>
>>
>> I do not think CCF has any such limitation where the clock should be
>> enabled and then a clock rate should be invoked. We should handle it
>> gracefully and that is what we have now when the caching capabilities
>> were added in the code. This has been already in our downstream drivers.
> 
> Should we do CFG caching on *all* RCGs to avoid having to scratch our
> heads over which ops to use with each clock individually?
> 

Yes, Konrad, that’s definitely the cleanest approach. If you're okay
with it, we can proceed with the current change first and then follow up
with a broader cleanup of the rcg2 ops. As part of that, we can also
transition the relevant SDCC clock targets to use floor_ops. This way,
we can avoid the rcg configuration failure logs in the boot sequence on
QCS615.

>>
>> We can add the fix to do a check 'clk_hw_is_enabled(hw)' in the normal
>> rcg2_ops/rcg2_floor/ceil_ops as well, then we can use them.
> 
> FWIW this is not the first time this issue has popped up..
> 
> I don't remember the details other than what I sent in the thread
> 
> https://lore.kernel.org/linux-arm-msm/20240427-topic-8450sdc2-v1-1-631cbb59e0e5@xxxxxxxxxx/
> 

Yes, but as I mentioned the new ops looks much cleaner, so wanted to
take this approach.

> Konrad
>>
>> AFAIK the eMMC framework has this code and this is not limited to drivers.
>>
> 

-- 
Thanks,
Taniya Das