[PATCH v2 5/8] phy: phy-imx8m-pcie: Add support for i.MX8M{M/P} PCIe PHY

[Marek Vasut]

On 2/26/24 2:49 PM, Sumit Garg wrote:
> On Mon, 26 Feb 2024 at 19:06, Marek Vasut <marex at denx.de> wrote:
>>
>> On 2/26/24 2:30 PM, Sumit Garg wrote:
>>> On Mon, 26 Feb 2024 at 18:34, Marek Vasut <marex at denx.de> wrote:
>>>>
>>>> On 2/26/24 1:50 PM, Sumit Garg wrote:
>>>>
>>>> [...]
>>>>
>>>>>>> +     ret = reset_get_by_name(dev, "pciephy", &imx8_phy->reset);
>>>>>>> +     if (ret) {
>>>>>>> +             dev_err(dev, "Failed to get PCIEPHY reset control\n");
>>>>>>> +             return ret;
>>>>>>> +     }
>>>>>>
>>>>>> Some clk_put() and co. fail path is missing here.
>>>>>
>>>>> Ditto for clk_put() not being available. However, I can add the fail
>>>>> path for the other reset.
>>>>
>>>> Oh, maybe reset_free() and clk_release*() then ?
>>>
>>> AFAICS, clk_release*() APIs are internally only a wrapper around
>>> clk_disable(). So actually those APIs don't seem to do anything useful
>>> in the fail path here.
>>>
>>> However, I will let clock tree maintainers chime in here if they plan
>>> to remove clk_release*() APIs on similar lines as with clk_free().
>>
>> Maybe they should be in place so in case in the future clock should be
>> torn down, it wouldn't be necessary to fix so many drivers ?
> 
> See the following comparison:
> 
> $ git grep -nr clk_get_by_name | wc -l
> 229
> 
> $ git grep -nr clk_release | wc -l
> 59
> 
> There are already many driver instances which don't use clk_release*() APIs.

And that's a good argument to make the situation worse ?