[PATCH v3 6/9] coresight: Avoid enable programming clock duplicately

[Anshuman Khandual]

On 10/06/25 5:33 PM, Leo Yan wrote:
> On Tue, Jun 10, 2025 at 04:09:32PM +0530, Anshuman Khandual wrote:
>>
>>
>> On 09/06/25 10:44 PM, Leo Yan wrote:
>>> On Mon, Jun 09, 2025 at 05:58:34PM +0100, Suzuki Kuruppassery Poulose wrote:
>>>
>>> [...]
>>>
>>>>>   static inline struct clk *coresight_get_enable_apb_pclk(struct device *dev)
>>>>>   {
>>>>> -	struct clk *pclk;
>>>>> +	struct clk *pclk = NULL;
>>>>> -	pclk = devm_clk_get_enabled(dev, "apb_pclk");
>>>>> -	if (IS_ERR(pclk))
>>>>> -		pclk = devm_clk_get_enabled(dev, "apb");
>>>>> +	if (!dev_is_amba(dev)) {
>>>>> +		pclk = devm_clk_get_enabled(dev, "apb_pclk");
>>>>> +		if (IS_ERR(pclk))
>>>>> +			pclk = devm_clk_get_enabled(dev, "apb");
>>>>
>>>> AMBA driver doesn't handle "apb" clock ? So we may need to retain that here
>>>> ?
>>>
>>> Here checks the condition "if (!dev_is_amba(dev))", it means the device
>>> is not an AMBA device (e.g., a platform device), the APB clock is
>>> enabled at here.
>>
>> Just exit early for AMBA devices when 'pclk' clock is still NULL ?
>>
>> 	if (dev_is_amba(dev))
>> 		return pclk;
> 
> If it is an AMBA device, we should return a NULL pointer, as this
> indicates that the APB clock is not managed by the CoreSight driver.
> 
> In this patch, I did not perform any refactoring and simply made a
> straightforward changed.  The refactoring is done in the patch 07, as
> you suggested, where the function is refined as:
> 
>     if (dev_is_amba(dev)) {
>         return NULL;
>     } else {
>         pclk = devm_clk_get_enabled(dev, "apb_pclk");
>         ...
>     }
> 
> Would it be acceptable to keep this patch as it is?

Agreed. Let's avoid the churn here as the refactoring happens
later anyway.