Re: [PATCH 1/3] avr32: fix build failure

[Sudip Mukherjee <sudipm.mukherjee@gmail.com>]

On Wed, Sep 23, 2015 at 07:15:16PM +0200, Hans-Christian Egtvedt wrote:
> Around Wed 23 Sep 2015 21:26:01 +0530 or thereabout, Sudip Mukherjee wrote:
> > On Mon, Sep 21, 2015 at 01:31:44PM +0530, Sudip Mukherjee wrote:
> >> On Mon, Sep 21, 2015 at 09:33:00AM +0200, Hans-Christian Egtvedt wrote:
> >> > Around Mon 21 Sep 2015 12:09:01 +0530 or thereabout, Sudip Mukherjee wrote:
> >> > > On Mon, Sep 21, 2015 at 08:09:42AM +0200, Hans-Christian Egtvedt wrote:
> >> > >> Around Sat 19 Sep 2015 22:42:57 +0530 or thereabout, Sudip Mukherjee wrote:
> >> > >> > While building avr32 with allmodconfig, the build used to fail with the
> >> > >> > message:
> >> > >> > error: implicit declaration of function 'pci_iomap'
> >> > >> > error: implicit declaration of function 'pci_iounmap'
> >> > >> 
> >> > >> What has changed recently that start pulling in these? AVR32 does not have
> >> > >> PCI at all, and will never have it either. Is this exposing a bug somewhere
> >> > >> else?
> >> > > It looks like pci_iomap and pci_iounmap doesnot depend on CONFIG_PCI.
> >> > > So drivers/net/ethernet/via/via-rhine.c is calling these functions even
> >> > > if PCI is disabled. The build log is at:
> >> > > https://travis-ci.org/sudipm-mukherjee/parport/jobs/81127188
> >> > > 
> >> > > You can find a similar discussion at:
> >> > > http://www.linux-mips.org/archives/linux-mips/2013-06/msg00510.html
> >> > 
> >> > If multiple architectures have similar problems, then a more global
> >> > definition of these unused functions could be added.
> >> > 
> >> > Just move the implementation for MIPS into include/asm-generic/io.h ?
> >> > 
> >> > That way we do not have to implement and identical stub for every device not
> >> > supporting PCI.
> >> 
> >> include/asm-generic/io.h is having:
> >> #ifndef CONFIG_GENERIC_IOMAP
> >> struct pci_dev;
> >> extern void __iomem *pci_iomap(struct pci_dev *dev, int bar, unsigned
> >> 		long max);
> >> 
> >> #ifndef pci_iounmap
> >> #define pci_iounmap pci_iounmap
> >> static inline void pci_iounmap(struct pci_dev *dev, void __iomem *p)
> >> {
> >> }
> >> #endif
> >> #endif /* CONFIG_GENERIC_IOMAP */
> >> 
> >> and CONFIG_GENERIC_IOMAP is not defined when I do allmodconfig with
> >> avr32. I have not seen this earlier, maybe in my patch pci_iounmap() was
> >> not required. But pci_iomap is declared as extern so we need have it
> >> somewhere.
> > Hi Hans,
> > Please suggest.
> 
> I can ack the original change, but it will only lead to a cluttered code
> base. The real problem is somewhere else, where it looks like the
> CONFIG_GENERIC_IOMAP symbol assumes PCI is for everybody. I would like if
> somebody addressed that problem instead.

I was having a look and it looks like generic empty pci_iomap and
pci_iounmap are already defined and is used only if CONFIG_PCI is not
defined and CONFIG_GENERIC_PCI_IOMAP is defined. So instead of adding
empty functions to each architecture where PCI is not supported maybe
the driver can have a dependency on CONFIG_GENERIC_PCI_IOMAP.
I have submitted the patch and you are in CC list. arch/sh had a similar
problem with the same driver.

regards
sudip