Re: [PATCH 01/12] net: dsa: lantiq_gswip: mark OF related data as maybe unused

[Vladimir Oltean <olteanv@gmail.com>]

On Sun, Mar 12, 2023 at 11:20:38AM +0100, Krzysztof Kozlowski wrote:
> On 11/03/2023 19:14, Vladimir Oltean wrote:
> > On Sat, Mar 11, 2023 at 06:32:52PM +0100, Krzysztof Kozlowski wrote:
> >> The driver can be compile tested with !CONFIG_OF making certain data
> >> unused:
> >>
> >>   drivers/net/dsa/lantiq_gswip.c:1888:34: error: ‘xway_gphy_match’ defined but not used [-Werror=unused-const-variable=]
> >>
> >> Signed-off-by: Krzysztof Kozlowski <krzysztof.kozlowski@linaro.org>
> >> ---
> > 
> > Do you happen to have any context as to why of_match_node() without
> > CONFIG_OF is implemented as:
> > 
> > #define of_match_node(_matches, _node)	NULL
> > 
> > and not as:
> > 
> > static inline const struct of_device_id *
> > of_match_node(const struct of_device_id *matches,
> > 	      const struct device_node *node)
> > {
> > 	return NULL;
> > }
> > 
> > ?
> > 
> > Generally, the static inline shim function model is nicer, because it
> > allows us to not scatter __maybe_unused all around.
> 
> Sorry, I don't follow. I don't touch that wrappers, just fix errors
> related to OF device ID tables, although in few cases it is indeed
> related to of_match_node.

I'm saying this because in lantiq_gswip.c, xway_gphy_match is accessed
through of_match_node(). If the shim definition for of_match_node() was
different, the variable wouldn't have been unused with CONFIG_OF=n.
I guess it's worth considering changing that wrapper instead of adding
the __maybe_unused.