Re: [PATCH v2] OF: of_address: clean up OF stub & extern functions

[Rob Herring <robh+dt@kernel.org>]

On Mon, May 24, 2021 at 2:09 PM Randy Dunlap <rdunlap@infradead.org> wrote:
>
> Adjust <linux/of_address.h> so that stubs are present when
> CONFIG_OF is not set *or* OF is set but OF_ADDRESS is not set.
>
> This eliminates 2 build errors on arch/s390/ when HAS_IOMEM
> is not set (so OF_ADDRESS is not set).
> I.e., it provides a stub for of_iomap() when one was previously
> not provided as well as removing some duplicate externs.

Personally, I think we should get rid of HAS_IOMEM or at least most of
its usage in kconfig. It has little purpose beyond hiding drivers in
kconfig and mainly for UML though I think UML no longer needs that
IIRC. (I'm not wild about 'depends on OF' either).

> s390-linux-ld: drivers/irqchip/irq-al-fic.o: in function `al_fic_init_dt':
> irq-al-fic.c:(.init.text+0x7a): undefined reference to `of_iomap'
> s390-linux-ld: drivers/clocksource/timer-of.o: in function `timer_of_init':
> timer-of.c:(.init.text+0xa4): undefined reference to `of_iomap'
>
> Tested with many randconfig builds, but there could still be some
> hidden problem here.
>
> Fixes: 4acf4b9cd453 ("of: move of_address_to_resource and of_iomap declarations from sparc")
> Signed-off-by: Randy Dunlap <rdunlap@infradead.org>
> Cc: Rob Herring <robh+dt@kernel.org>
> Cc: Frank Rowand <frowand.list@gmail.com>
> Cc: devicetree@vger.kernel.org
> Cc: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
> Reported-by: kernel test robot <lkp@intel.com>
> ---
> v2: handle SPARC as a special case since it provides its own versions of
>     of_address_to_resource() and of_iomap();
>     fix build errors reported by lkp/ktr and address comments from Laurent;
>     do more randconfig build testing;
>
>  include/linux/of_address.h |   11 +++++++----
>  1 file changed, 7 insertions(+), 4 deletions(-)
>
> --- linux-next-20210524.orig/include/linux/of_address.h
> +++ linux-next-20210524/include/linux/of_address.h
> @@ -106,11 +106,12 @@ static inline bool of_dma_is_coherent(st
>  }
>  #endif /* CONFIG_OF_ADDRESS */
>
> -#ifdef CONFIG_OF
> +#ifdef CONFIG_SPARC /* SPARC has its own versions of these */

The whole point of CONFIG_OF_ADDRESS is really just for SPARC. So I
don't really like the mixture of the ifdefs here and in kconfig. It
looks only more fragile.

Can we drop the HAS_IOMEM dependency from CONFIG_OF_ADDRESS and then
fix the fallout from that? That would also remove all the other build
time dependencies on HAS_IOMEM.

>  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
> +extern void __iomem *of_iomap(struct device_node *device, int index);
> +#else /* !CONFIG_SPARC */
> +#if (defined(CONFIG_OF) && !defined(CONFIG_OF_ADDRESS)) || !defined(CONFIG_OF)
>  static inline int of_address_to_resource(struct device_node *dev, int index,
>                                          struct resource *r)
>  {
> @@ -121,7 +122,9 @@ static inline void __iomem *of_iomap(str
>  {
>         return NULL;
>  }
> -#endif
> +#endif /* (defined(CONFIG_OF) && !defined(CONFIG_OF_ADDRESS)) || !defined(CONFIG_OF) */
> +#endif /* CONFIG_SPARC */
> +
>  #define of_range_parser_init of_pci_range_parser_init
>
>  #if defined(CONFIG_OF_ADDRESS) && defined(CONFIG_PCI)