Re: [PATCH 1/2] clk: imx93: Drop macro IMX93_CLK_END

[Krzysztof Kozlowski <krzk@kernel.org>]

On 28/06/2024 03:17, Peng Fan wrote:
>> Subject: Re: [PATCH 1/2] clk: imx93: Drop macro IMX93_CLK_END
>>
>> On 25/06/2024 12:43, Pengfei Li wrote:
>>> On Tue, Jun 25, 2024 at 09:44:42AM +0200, Krzysztof Kozlowski
>> wrote:
>>>> On 25/06/2024 19:51, Pengfei Li wrote:
>>>>> IMX93_CLK_END was previously defined in imx93-clock.h to
>> indicate
>>>>> the number of clocks, but it is not part of the ABI, so it should be
>>>>> dropped.
>>>>>
>>>>> Now, the driver gets the number of clks by querying the maximum
>>>>> index in the clk array. Due to the discontinuity in the definition
>>>>> of clk index, with some gaps present, the total count cannot be
>>>>> obtained by summing the array size.
>>>>>
>>>>> Signed-off-by: Pengfei Li <pengfei.li_1@nxp.com>
>>>>> ---
>>>>>  drivers/clk/imx/clk-imx93.c | 25 +++++++++++++++++++++----
>>>>>  1 file changed, 21 insertions(+), 4 deletions(-)
>>>>>
>>>>> diff --git a/drivers/clk/imx/clk-imx93.c
>>>>> b/drivers/clk/imx/clk-imx93.c index c6a9bc8ecc1f..68c929512e16
>>>>> 100644
>>>>> --- a/drivers/clk/imx/clk-imx93.c
>>>>> +++ b/drivers/clk/imx/clk-imx93.c
>>>>> @@ -257,6 +257,20 @@ static const struct imx93_clk_ccgr
>> {  static
>>>>> struct clk_hw_onecell_data *clk_hw_data;  static struct clk_hw
>>>>> **clks;
>>>>>
>>>>> +static int imx_clks_get_num(void)
>>>>> +{
>>>>> +	u32 val = 0;
>>>>> +	int i;
>>>>> +
>>>>> +	for (i = 0; i < ARRAY_SIZE(root_array); i++)
>>>>> +		val = max_t(u32, val, root_array[i].clk);
>>>>> +
>>>>> +	for (i = 0; i < ARRAY_SIZE(ccgr_array); i++)
>>>>> +		val = max_t(u32, val, ccgr_array[i].clk);
>>>>> +
>>>>> +	return val + 1;
>>>>> +}
>>>>> +
>>>>>  static int imx93_clocks_probe(struct platform_device *pdev)  {
>>>>>  	struct device *dev = &pdev->dev;
>>>>> @@ -264,14 +278,17 @@ static int imx93_clocks_probe(struct
>> platform_device *pdev)
>>>>>  	const struct imx93_clk_root *root;
>>>>>  	const struct imx93_clk_ccgr *ccgr;
>>>>>  	void __iomem *base, *anatop_base;
>>>>> +	int clks_num;
>>>>>  	int i, ret;
>>>>>
>>>>> +	clks_num = imx_clks_get_num();
>>>>> +
>>>>>  	clk_hw_data = devm_kzalloc(dev, struct_size(clk_hw_data,
>> hws,
>>>>> -					  IMX93_CLK_END),
>> GFP_KERNEL);
>>>>> +					  clks_num), GFP_KERNEL);
>>>>>  	if (!clk_hw_data)
>>>>>  		return -ENOMEM;
>>>>>
>>>>> -	clk_hw_data->num = IMX93_CLK_END;
>>>>> +	clk_hw_data->num = clks_num;
>>>>
>>>> Why so complicated code instead of pre-processor define or array
>> size?
>>>>
>>>> Best regards,
>>>> Krzysztof
>>>>
>>>>
>>>
>>> Hi Krzysztof,
>>>
>>> Thanks for the comment, here are some of our thoughts.
>>>
>>> Regarding the predefined method, it's easy to forget to update the
>>> macro definition when adding some new clocks to imx93-clock.h in
>> the future.
>>
>> Somehow most developers in most platforms can do it... Anyway, that
>> would be build time detectable so no problem at all.
>>
>>>
>>> Also, we cannot use the array size method in this scenario, as some
>>> unnecessary clocks have been removed in the past, resulting in
>>> discontinuous definitions of clock indexes. This means that the
>>> maximum clock index can be larger than the allocated clk_hw array
>> size. At this point, using the maximum index to access the clk_hw array
>> will result in an out of bounds error.
>>
>> You mix bindings with array entries. That's independent or just clock
>> drivers are broken.
> 
> But there is issue that binding update and clock driver are normally in
> two patches.  So if just use the IMX93_CLK_END in this file,
> it will be easy to break `git bisect`.

There is no issue. Srsly, this would be the only, only one driver having
that issue.

How is this even possible? How adding one new define for pre-processor
would cause driver issues or some sort of bisectability problems?

These are basics of C we talk about now...

Best regards,
Krzysztof