Re: [PATCH 01/16] spi: armada-3700: Drop of_match_ptr for ID table

[Krzysztof Kozlowski]

On 13/03/2023 20:22, Mark Brown wrote:
> On Mon, Mar 13, 2023 at 07:39:45PM +0100, Krzysztof Kozlowski wrote:
>> On 13/03/2023 14:55, Mark Brown wrote:
>>> On Fri, Mar 10, 2023 at 11:28:42PM +0100, Krzysztof Kozlowski wrote:
> 
>>>>   drivers/spi/spi-armada-3700.c:807:34: error: ‘a3700_spi_dt_ids’ defined but not used [-Werror=unused-const-variable=]
> 
>>> It would be much better to fix of_match_ptr() and/or the module stuff
>>> that also references the match table here.
> 
>> Why? The recommendation is in general not to use of_match_ptr, because
>> there are little benefits but it disables matching via PRP0001. Jonathan
>> in parallel thread explicitly said of_match_ptr should disappear and he
>> is not accepting any new code with it. And in general he is right.
> 
> If that's the case then why are you adding maybe unused annotations for
> half the drivers rather than removing their of_match_ptr() usages?
> There doesn't seem to be any logic here, it's just randomly making
> changes as far as I can tell.

These are not random but depend on whether OF is the only matching
method or one of few. Although for IIO all of my patches would be
dropping the of_match_ptr... For some cases in other patchsets I added
__maybe_unused also because of_match_node(), when the table is not used
in driver of_match_table.

> 
> The PRP0001 stuff isn't an issue, of_match_ptr() can just be changed to
> do the right thing for CONFIG_ACPI.

That's actually interesting idea, kind of obvious so I wonder why it
wasn't done like this in the first place in 886ca88be6b3 ("ACPI / bus:
Respect PRP0001 when retrieving device match data"). Maybe not to
populate OF device ID tables for the ACPI systems which do not care
about PRP0001?


>  It doesn't buy us huge amounts but
> it also costs us very little and may be useful in future.  When there's
> missing annotations it's not causing issues for practical configurations
> as far as I can tell, and if the macro were updated for CONFIG_ACPI it'd
> be even less of an issue.

Best regards,
Krzysztof