Linus,
	I think I'm getting what you're aiming for here.  See if this patch comes close 
to what you're looking for.

	The basic idea is we get rid of the byte/word/dword functions and just put a 
size arg in the generic calls.  I put back the different struct pci_ops for 
conf1 and conf2, so that drivers and other code could use either config option. 
  I also fixed the macros in access.h so that it correctly calls the new functions.

Please *don't* apply this as it will break a bunch of stuff (just about all PCI 
code ;)

Cheers!

-Matt


Linus Torvalds wrote:
> On 13 Aug 2002, Alan Cox wrote:
> 
>>>OK, that IDE thing smacks of unmitigated evil to me, but if things are relying 
>>>on it, we shouldn't change it.
>>
>>It wants to force its own conf1/conf2 over the BIOS even if BIOS is
>>preferred because some BIOSes dont honour the size requested and the
>>hardware has bugs.
>>
>>That to me says there may well be cleaner approaches.
> 
> 
> The thing I liked about the separate structures for function pointers for 
> conf1/conf2 is that I could at least _see_ that the IDE driver might some 
> day be changed to just do
> 
> 	..
> 	conf2_struct->pci_config_read_byte(..)
> 	..
> 
> even if (judging by past performance) this would never happen ;)
> 
> This is why I'd like to continue with the notion of having a well-defined 
> structure that contains all the pointers (and one default case). Now, 
> shrinking those structures down to 2 entries instead of 6 sounds like a 
> fine idea to me, but short-circuiting them internally sounds bad because 
> it loses the ability to use the pci config space functions independently 
> of each other.
> 
> 		Linus
> 
>