On Sat, 4 Sept 2021 at 04:23, Rob Herring <robh+dt@kernel.org> wrote:
>
> On Thu, Sep 2, 2021 at 4:27 AM Chunyan Zhang <zhang.lyra@gmail.com> wrote:
> >
> > From: Chunyan Zhang <chunyan.zhang@unisoc.com>
> >
> > If CONFIG_OF is selected, but CONFIG_OF_ADDRESS is not, when compiling
> > files (sunch as timer_of.c) in which the function of_iomap() is invoked,
> > compiler would report 'undefined reference to of_iomap', the same case
> > is for of_address_to_resource().
>
> What arch and config are you building?

sparc, and the config is attached.

>
> > This patch also makes sure that of_iomap() and of_address_to_resource()
> > are declared for sparc so that sparc can have its specific
> > implementations in arch/sparc/kernel/of_device_common.c, even if
> > including include/linux/of_address.h in it.
> >
> > Signed-off-by: Chunyan Zhang <chunyan.zhang@unisoc.com>
> > ---
> >  include/linux/of_address.h | 33 +++++++++++++++++----------------
> >  1 file changed, 17 insertions(+), 16 deletions(-)
> >
> > diff --git a/include/linux/of_address.h b/include/linux/of_address.h
> > index 45598dbec269..02a719d58466 100644
> > --- a/include/linux/of_address.h
> > +++ b/include/linux/of_address.h
> > @@ -80,6 +80,23 @@ static inline u64 of_translate_address(struct device_node *np,
> >         return OF_BAD_ADDR;
> >  }
> >
> > +#ifdef CONFIG_SPARC
> > +extern int of_address_to_resource(struct device_node *dev, int index,
> > +                                 struct resource *r);
> > +void __iomem *of_iomap(struct device_node *device, int index);
>
> This is now in the !CONFIG_OF_ADDRESS section. So for
> CONFIG_OF_ADDRESS, we'd never have a function declaration.

Not sure I got your point.
There are these two function declarations both in
arch/sparc/kernel/of_device_common.c (CONFIG_SPARC &&
!CONFIG_OF_ADDRESS) and drivers/of/address.c (!CONFIG_SPARC &&
CONFIG_OF_ADDRESS).

> This change does not look right at all to me.
>
> > +#else
> > +static inline int of_address_to_resource(struct device_node *dev, int index,
> > +                                        struct resource *r)
> > +{
> > +       return -EINVAL;
> > +}
> > +
> > +static inline void __iomem *of_iomap(struct device_node *device, int index)
> > +{
> > +       return NULL;
> > +}
> > +#endif

Now, I'm also not sure if this change is right, I learned in another
patch recently that Linus prefers to avoid the extra code like this.

https://lkml.org/lkml/2021/7/2/511



> > +
> >  static inline const __be32 *__of_get_address(struct device_node *dev, int index, int bar_no,
> >                                              u64 *size, unsigned int *flags)
> >  {
> > @@ -124,22 +141,6 @@ static inline bool of_dma_is_coherent(struct device_node *np)
> >  }
> >  #endif /* CONFIG_OF_ADDRESS */
> >
> > -#ifdef CONFIG_OF
> > -extern int of_address_to_resource(struct device_node *dev, int index,
> > -                                 struct resource *r);
> > -void __iomem *of_iomap(struct device_node *node, int index);
> > -#else
> > -static inline int of_address_to_resource(struct device_node *dev, int index,
> > -                                        struct resource *r)
> > -{
> > -       return -EINVAL;
> > -}
> > -
> > -static inline void __iomem *of_iomap(struct device_node *device, int index)
> > -{
> > -       return NULL;
> > -}
> > -#endif
> >  #define of_range_parser_init of_pci_range_parser_init
> >
> >  static inline const __be32 *of_get_address(struct device_node *dev, int index,
> > --
> > 2.25.1
> >