Re: [PATCH v2 2/4] hwrng: stm32 - implement support for STM32MP25x platforms

[Gatien CHEVALLIER <gatien.chevallier@foss.st.com>]

On 10/14/24 20:55, Marek Vasut wrote:
> On 10/14/24 2:36 PM, Gatien CHEVALLIER wrote:
>>
>>
>> On 10/14/24 10:52, Marek Vasut wrote:
>>> On 10/14/24 10:38 AM, Gatien CHEVALLIER wrote:
>>>>
>>>>
>>>> On 10/11/24 18:17, Marek Vasut wrote:
>>>>> On 10/11/24 5:41 PM, Gatien Chevallier wrote:
>>>>>
>>>>> [...]
>>>>>
>>>>>> @@ -551,6 +565,41 @@ static int stm32_rng_probe(struct 
>>>>>> platform_device *ofdev)
>>>>>>       priv->rng.read = stm32_rng_read;
>>>>>>       priv->rng.quality = 900;
>>>>>> +    if (!priv->data->nb_clock || priv->data->nb_clock > 2)
>>>>>> +        return -EINVAL;
>>>>>> +
>>>>>> +    priv->clk_bulk = devm_kzalloc(dev, priv->data->nb_clock * 
>>>>>> sizeof(*priv->clk_bulk),
>>>>>> +                      GFP_KERNEL);
>>>>>> +    if (!priv->clk_bulk)
>>>>>> +        return -ENOMEM;
>>>>>
>>>>> Try this:
>>>>>
>>>>> ret = devm_clk_bulk_get(dev, priv->data->nb_clock, priv->clk_bulk);
>>>>> ...
>>>>> // Swap the clock if they are not in the right order:
>>>>> if (priv->data->nb_clock == 2 &&
>>>>>      strcmp(__clk_get_name(priv->clk_bulk[0].clk), "core"))
>>>>> {
>>>>>   const char *id = priv->clk_bulk[1].id;
>>>>>   struct clk *clk = priv->clk_bulk[1].clk;
>>>>>   priv->clk_bulk[1].id = priv->clk_bulk[0].id;
>>>>>   priv->clk_bulk[1].clk = priv->clk_bulk[0].clk;
>>>>>   priv->clk_bulk[0].id = id;
>>>>>   priv->clk_bulk[0].clk = clk;
>>>>> }
>>>>>
>>>>
>>>> Hi Marek,
>>>>
>>>> This won't work as the name returned by this API is clk->core->name.
>>>> AFAICT, it doesn't correspond to the names present in the device tree
>>>> under the "clock-names" property.
>>>> Any other idea or are you fine with what's below?
>>> Hmmm, it is not great, but at least it reduces the changes throughout 
>>> the driver, so that is an improvement.
>>>
>>> I guess one could do some of_clk_get() and clk_is_match() in probe to 
>>> look up the clock in OF by name and then compare which clock is which 
>>> before swapping them in clk_bulk[] array, but that might be too 
>>> convoluted?
>>
>> Yes, probably too much. What's present in the patch is not close to
>> perfection but has the advantage of being straightforward. If we agree
>> on that, I'll send a V3 containing the modifications in the bindings
>> file.
> Errr, I'm sorry, maybe there is a way to do this better. Look at 
> drivers/clk/clk-bulk.c :
> 
>   15 static int __must_check of_clk_bulk_get(struct device_node *np, int 
> num_clks,
>   16                                         struct clk_bulk_data *clks)
>   17 {
>   18         int ret;
>   19         int i;
>   20
>   21         for (i = 0; i < num_clks; i++) {
>   22                 clks[i].id = NULL;
>   23                 clks[i].clk = NULL;
>   24         }
>   25
>   26         for (i = 0; i < num_clks; i++) {
>   27                 of_property_read_string_index(np, "clock-names", i, 
> &clks[i].id);
>   28                 clks[i].clk = of_clk_get(np, i);
> 
> If I read this right, then clks[i].id should be the DT clock name. So 
> the swap conditional above could use .id to identify whether the first 
> position is core clock or not, like this:
> 
> if (priv->data->nb_clock == 2 &&
>      strcmp(__clk_get_name(priv->clk_bulk[0].id), "core"))
>                                              ^^
> 
> You might need to use devm_clk_bulk_get_all() to access the 
> of_clk_bulk_get() .
> 
> Or am I missing something still ?

Oooooh I see, devm_clk_bulk_get() and devm_clk_bulk_get_all() use
a different path. I don't understand why, to be honest... The doc
doesn't state this difference either.

I'll give this a try while also correcting the issue that the robot
highlighted.

Best regards,
Gatien