Re: [PATCH] bcma: Prevent build of PCI host features in module

[Matt Redfearn]

Hi Kalle,

On 15/01/18 10:07, Kalle Valo wrote:
> Matt Redfearn <matt.redfearn@xxxxxxxx> writes:
>
> > Attempting to build bcma.ko with BCMA_DRIVER_PCI_HOSTMODE=y results in
> > a build error due to use of symbols not exported from vmlinux:
> >
> > ERROR: "pcibios_enable_device" [drivers/bcma/bcma.ko] undefined!
> > ERROR: "register_pci_controller" [drivers/bcma/bcma.ko] undefined!
> > make[1]: *** [scripts/Makefile.modpost:92: __modpost] Error 1
> >
> > To prevent this, don't allow the host mode feature to be built if
> > CONFIG_BCMA=m
> >
> > Signed-off-by: Matt Redfearn <matt.redfearn@xxxxxxxx>
> >
> > ---
> >
> >   drivers/bcma/Kconfig | 2 +-
> >   1 file changed, 1 insertion(+), 1 deletion(-)
> >
> > diff --git a/drivers/bcma/Kconfig b/drivers/bcma/Kconfig
> > index 02d78f6cecbb..4294784b9cf1 100644
> > --- a/drivers/bcma/Kconfig
> > +++ b/drivers/bcma/Kconfig
> > @@ -55,7 +55,7 @@ config BCMA_DRIVER_PCI
> >   
> >   config BCMA_DRIVER_PCI_HOSTMODE
> >   	bool "Driver for PCI core working in hostmode"
> > -	depends on MIPS && BCMA_DRIVER_PCI
> > +	depends on MIPS && BCMA_DRIVER_PCI && BCMA = y
>
> Is this a new regression? Do you know the commit which broke this?

As far as I can see, pcibios_enable_device and register_pci_controller 
have never being exported symbols from vmlinux, and an allmodconfig 
build with CONFIG_MIPS_COBALT=y (MIPS cobalt platform) which attempts to 
put this this functionality into a module would never have linked 
successfully. As such it can be traced back to when this functionality 
was added, 49dc9577155576b10ff79f0c1486c816b01f58bf ("bcma: add PCIe 
host controller").

> Is it somehow related to this:
>
> bcma: Fix 'allmodconfig' and BCMA builds on MIPS targets
> https://patchwork.kernel.org/patch/10162839/
>
> Or are these two separate issues?

Separate issues - that one fixes allmodconfig when CONFIG_MIPS_GENERIC=y 
(MIPS generic platform) which does not define the struct pci_controller 
- that error was really introduced when that struct definition was 
removed for the generic platform by 
c5611df968047fb0b38156497b4242730ef66108 ("MIPS: PCI: Introduce 
CONFIG_PCI_DRIVERS_LEGACY").

I think both patches are valid since they fix errors building 
allmodconfig on 2 separate MIPS platforms.

Thanks,
Matt