PLEA: Please fix mach/gpio.h includes (was: Re: [RFC PATCH 2/2] GPIO: add gpiolib and irqchip for CSR SiRFprimaII GPIO controller)

[Nicolas Pitre]

On Wed, 27 Jul 2011, Barry Song wrote:

> 2011/7/27 Nicolas Pitre <nicolas.pitre at linaro.org>:
> > On Tue, 26 Jul 2011, Russell King - ARM Linux wrote:
> >
> >> I don't think there's any plans to break the:
> >>
> >> linux/gpio.h -> asm/gpio.h -> mach/gpio.h
> >>
> >> include path at the moment, as platforms do define ARCH_NR_GPIO,
> >> which has to be done before asm-generic/gpiolib.h is included.
> >
> > Well, this GPIO business is the biggest hurdle towards a single kernel
> > image that can support multiple SOCs at the moment, and the only one for
> > which I have no solution yet.  Anything that could help removing
> > mach/gpio.h from asm/gpio.h would be welcome.
> 
> After reading all the following patches,
> 
> 1. use CONFIG_PHYS_OFFSET to prepare for removal of a bunch of
> <mach/memory.h> files
> 
> 2. move from ARM_DMA_ZONE_SIZE to mdesc->dma_zone_size
> 
> 3. convert boot_params to atag_offset
> 
> Have i lost anything for the work on 1 and 2? considering some
> platforms need CONSISTENT_DMA_SIZE more than the default 2MB in
> arch/arm/include/asm/memory.h, we can't remove all memory.h only by
> defining CONFIG_PHYS_OFFSET and moving ARM_DMA_ZONE_SIZE to
> mdesc->dma_zone_size. So CONSISTENT_DMA_SIZE should also be a left
> issue except gpio.h.

CONSISTENT_DMA_SIZE can be solved the same way as ARM_DMA_ZONE_SIZE, 
plus a few changes in dma-mapping.c to dynamically allocate the 
consistent_pte array.

But that's far from the end of it.  In a nutshell:

- We have to get rid of mach/vmalloc.h (trivial)

- We have to kill every SOC specific definition for CLOCK_TICK_RATE and 
  remove mach/timex.h.

- All mach/io.h need a similar refactoring.

- asm/irq.h must stop including mach/irqs.h, and everyone should 
  initialize mdesc->nr_irqs

- mach/entry-macro.S should be translated into something that hooks to 
  mdesc->handle_irq.

- mach/system.h content needs to be moved to out-of-line indirect calls.

- A multy-SOC kernel binary might not support CONFIG_DEBUG_LL out of the 
  box.

- And there are a few odd cases we probably might not find sufficient 
  motivation to fix them and simply exclude them from a single image 
  config.

But there is mach/gpio.h which is the real bummer at the moment.


Nicolas