Re: [PATCH next v9 1/4] i2c: hisi: Add initial device tree support

[chenweilong <chenweilong@huawei.com>]

On 2022/10/31 23:42, Andy Shevchenko wrote:
> On Mon, Oct 31, 2022 at 09:57:51AM +0800, chenweilong wrote:
>> On 2022/10/31 6:01, Andy Shevchenko wrote:
>>> On Sat, Oct 29, 2022 at 07:59:36PM +0800, Weilong Chen wrote:
>>>> The HiSilicon I2C controller can be used on embedded platform, which
>>>> boot from devicetree.
>>> ...
>>>
>>>> +#include <linux/acpi.h>
>>>> +#include <linux/of.h>
>>> Why?
>>>
>>> ...
>>>
>>>> +#ifdef CONFIG_ACPI
>>> Why?
>>>
>>> ...
>>>
>>>> +#ifdef CONFIG_OF
>>> Why?
>>>
>>> ...
>>>
>>>> -		.acpi_match_table = hisi_i2c_acpi_ids,
>>>> +		.acpi_match_table = ACPI_PTR(hisi_i2c_acpi_ids),
>>> Why?
>>>
>>> ...
>>>
>>>> +		.of_match_table = of_match_ptr(hisi_i2c_dts_ids),
>>> Why of_match_ptr()?
>> There's a lot of drivers use of_match_ptr/ACPI_PTR to protect the of_device_id and
>> have explicit headers file references to linux/acpi.h or linux/of.h, such as
>> drivers/media/platform/intel/pxa_camera.c,
>> bluetooth/hci_intel.c, 
>> platform/x86/intel/chtwc_int33fe.c,
>> platform/x86/intel/pmc/core.c and so on.
> We have a lot of the legacy or not-up-to-dated to all new kernel APIs code.
> Does it justify not to use the new approach in the new contribution?
>
> ...
>
>> The acpi.h and of.h have a nice function or macro definition if CONFIG_OF/ACPI is not satisfy,
>> for example:
>>
>> #define ACPI_PTR(_ptr)  (_ptr)  vs  #define ACPI_PTR(_ptr)  (NULL)
>>
>> and also a lot of 'static inline' function there.
> And why do you need it?
>
> ...
>
>> Seems a good idea to remove all of them, the codes your noted may look a bit
>> verbose there. But I think it is valuable for a driver and device ,telling
>> users it support acpi boot or is it just embedded.
> So, what do we gain here?
>
> (Fill the "Advantages of your code" section below)
>
> Disadvantages of your code:
> - ugly ifdeffery which we usually do not appreciate
> - in some cases it's good to have OF ID table on ACPI platforms (see what
>   PRP0001 trick is)
> - use old approach for the compiler on how to avoid warnings of the static
>   variables being defined and not used (note, neither ACPI_PTR() nor
>   of_match_ptr() provides a new approach on that, so you have to amend them
>   first)
> - as a side effect additional headers to be included that are used for 1% or
>   less of their capacity and slow down the compilation

Thanks very much for your detailed explanation.

By the way,  is it valuable to make a cleanup for the legacy not-up-to-dated drivers?

There's lots of of_match_ptr or ACPI_PTR...


Best Regards.